Investigation and Development of Novel Methodologies for Disease Diagnosis and Drug Monitoring Utilizing Mass Spectrometry and Field Asymmetric Ion Mobility Spectrometry

Material Information

Investigation and Development of Novel Methodologies for Disease Diagnosis and Drug Monitoring Utilizing Mass Spectrometry and Field Asymmetric Ion Mobility Spectrometry
Costanzo, Michael T
Place of Publication:
[Gainesville, Fla.]
University of Florida
Publication Date:
Physical Description:
1 online resource (244 p.)

Thesis/Dissertation Information

Doctorate ( Ph.D.)
Degree Grantor:
University of Florida
Degree Disciplines:
Committee Chair:
Committee Co-Chair:
Committee Members:
Graduation Date:


Subjects / Keywords:
Cell culture techniques ( jstor )
Electric fields ( jstor )
Ionization ( jstor )
Ions ( jstor )
Mass spectroscopy ( jstor )
Melanoma ( jstor )
Molecules ( jstor )
Principal components analysis ( jstor )
Solvents ( jstor )
Waveforms ( jstor )
Chemistry -- Dissertations, Academic -- UF
ambient -- analysis -- analytical -- breath -- chemistry -- detection -- disease -- drug -- faims -- ion -- mass -- melanoma -- metabolomics -- mobility -- spectrometry
bibliography ( marcgt )
theses ( marcgt )
government publication (state, provincial, terriorial, dependent) ( marcgt )
born-digital ( sobekcm )
Electronic Thesis or Dissertation
Chemistry thesis, Ph.D.


In a society of ever increasing medical costs and invasive procedures, the development of quick, cheap, and painless diagnostic tests is crucial for not only the doctor, but also the patient. By utilizing ambient analytical techniques, medical professionals could directly analyze samples in a more efficient, cost effective manner, more applicable to the clinical setting; i.e., no need to send off to the lab. By cutting down on analysis time and cost, and increasing accessibility of diagnostic results to nonexperts, visits to physicians would become quicker and more beneficial to patients who would no longer have to submit to a variety of disease panels and wait days or weeks for results. Mass spectrometry (MS) has long been used for the identification of compounds, as both a standalone instrument and coupled to any variety of supplemental techniques. This research employed several different types of mass spectrometers and numerous ionization methods (e.g., APCI, DART, EESI, ESI) to identify compounds of interest for multiple diagnostic applications. MS can prove especially valuable when coupled to ultra high performance liquid chromatography (UHPLC) for global analysis methods. Utilizing chromatographic separation in conjunction with mass spectrometric analysis can produce a list of thousands of compounds; multivariate data analysis can potentially reduce this list to only those that may be of importance to a specific study. Moreover, MS can be combined with high field asymmetric waveform ion mobility spectrometry (FAIMS) to perform targeted analysis on select compounds of known significance. Standalone FAIMS has the added benefit of portability, providing an easy transition to the clinical setting. With this in mind, FAIMS MS would be used to produce fingerprint spectra that correlate to specific analytes of interest, then MS could be decoupled and standalone FAIMS could provide rapid, noninvasive results in and outside of the lab. The primary focus of the work presented will be two related projects, comprehensive metabolomics of malignant melanoma by UHPLC and direct analysis MS, and analysis of human breath by FAIMS and FAIMS MS. In addition, the use of UV spectroscopy to characterize capsaicin solutions will be described. ( en )
General Note:
In the series University of Florida Digital Collections.
General Note:
Includes vita.
Includes bibliographical references.
Source of Description:
Description based on online resource; title from PDF title page.
Source of Description:
This bibliographic record is available under the Creative Commons CC0 public domain dedication. The University of Florida Libraries, as creator of this bibliographic record, has waived all rights to it worldwide under copyright law, including all related and neighboring rights, to the extent allowed by law.
Thesis (Ph.D.)--University of Florida, 2015.
Co-adviser: BASSO,KARI B.
Electronic Access:
Statement of Responsibility:
by Michael T Costanzo.

Record Information

Source Institution:
Rights Management:
Applicable rights reserved.
Embargo Date:
LD1780 2015 ( lcc )


This item has the following downloads:

Full Text




© 2015 Michael Thomas Costanzo


To my family a nd friends


4 ACKNOWLEDGMENTS I would like to extend my gratitude and thanks to all of those who have helped me along the way with their support and encouragement . I must first recognize Dr. Richard Yost, my advisor and mentor of the past five years. Dr. Yost is a man whose commitment and excitement for science is infectious, without whose guidance would have made achieving this degree much more difficult. I would also like to acknowledge the members of my graduate committee for their direction and consid eration: Dr. Kari Basso, Dr. Paul Davenport, Dr. Stephen Miller , and Dr. Benjamin Smith; as well as my previous committee member Dr. David Powell, and Dr. James Horvath whom I taught under for most of my years of graduate schooling. I must also thank my f ellow Yost group members, most notably Chia Wei Tsai and Robert Menger , who showed me the ropes after I joined the Yost team and instilled in me what it takes to be a successful graduate student. For their ideas in the lab and friendship outside of it, I w ould like to thank other members of the Yost group : Emily Yu Hsuan Tsai, Liz Muffly , Candice Ulmer, Rainey Patte rson, Chris Chouinard , and Whitney Stutts ; with a special shout out to my diligent undergrad, Matthew Kazaleh. y appreciation to my fellow chemists Gianna DiFrancesco, Andrew Mowson, Andrew Powers, and Russell Winkel. I also thank those working tirelessly to ensure the continued efforts of the graduate and analytical offices, in particular, Lori Clark and Antoinett e Knight. Furthermore, I would like to give a huge thanks to all of my collaborators throughout my year s as a graduate student at UF. Without their gracious donations of sample s and instrument s, software provisions, research funding, or valuable insight, this dissertation would certainly not be possible. A special thanks to Owlstone and Cannabix


5 Technologies and Dr. Nikolaus Gravenstein for sharing his fantastic ideas on melanoma detection, Dr. Matt hew Booth for getting me started with breath dopants, and Dr. Timothy Garret for the ideas and discussions that greatly improved my understanding of chromatography and mass spectrometry in general. And so, w hile I gaze ahead to the future, I must first loo k back at everyone who l ed me down this path to graduate school. T o Dr. Troy Wood, my undergraduate advisor who gave me my first glimpse into chemistry research, to Anne Ruppert who, before anyone else, recognized my potential in chemistry, and to my frien ds and family who . Gainesville, I have unfortunately left many out. However, the number of people I thank does not detract from the amount of thanks and praise I have toward everyone. Whether their c ontributions were academic or personal last five years have become a huge part of my memory and work at UF, and wi thout all of them, this would not be possible. So again, to you all, I say thank you.


6 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 9 LIST OF FIGURES ................................ ................................ ................................ ........ 10 ABSTRACT ................................ ................................ ................................ ................... 14 CHAPTER 1 A BRIEF INTRODUCTION TO A VARIETY OF BIOMEDICALLY APPLICABLE ANALYTICAL TECHNIQUES AND METHODOLOGIES ................................ ........ 16 Ultraviolet Visible Molecular Absorption Spectrometry ................................ ........... 16 Background ................................ ................................ ................................ ...... 16 Principles of M olecular Absorption Spectrometry ................................ ............. 16 Ambient Ionization Mass Spectrometry ................................ ................................ ... 18 Ionization Techniques ................................ ................................ ...................... 18 Atmospheric pressure chemical ionization ................................ ................. 18 Electrospray ionization ................................ ................................ ............... 20 Mass Analyzers ................................ ................................ ................................ 24 Ion trap ................................ ................................ ................................ ....... 25 Orbitrap ................................ ................................ ................................ ...... 27 Time of flight ................................ ................................ .............................. 29 Compound Identification ................................ ................................ ................... 31 Accurate mass ................................ ................................ ........................... 32 Tandem mass spectrometry ................................ ................................ ....... 32 High Field Asymmetric Waveform Ion Mobility Spectrometry FAIMS .................. 33 Principles of IMS and FAIMS ................................ ................................ ............ 33 Modifications to Conventional FAIMS ................................ ............................... 36 Cell geometry ................................ ................................ ............................. 36 Miniature FAIMS cells and micromachined FAIMS chip ............................ 37 Addition of solvent vapor ................................ ................................ ............ 39 Application of a square waveform ................................ .............................. 40 High Performance Liquid Chromatography HPLC ................................ ............... 41 Normal Phase vs. Reversed Phase Chromatography ................................ ...... 42 High Performance Liquid Chromatography ................................ ...................... 43 Multivariate Data Analysis ................................ ................................ ....................... 45 Principal Component Analysis ................................ ................................ .......... 47 Partial Lea st Squares Discriminant Analysis ................................ .................... 49 Applying Statistical Analysis to Biomarker Detection ................................ ........ 49 Scope of the Dissertation ................................ ................................ ........................ 50


7 2 STANDARDIZED METHOD FOR SOLUBILITY AND STORAGE OF CAPSAICIN BASED SOLUTIONS FOR COUGH INDUCTION .............................. 76 Introduction ................................ ................................ ................................ ............. 76 Experimental ................................ ................................ ................................ ........... 78 Chemicals and Reagents ................................ ................................ ................. 78 Solubility of CAP in Different Solvent Systems ................................ ................. 78 Stability of CAP in 10% EtOH Solutions ................................ ........................... 79 Determination of CAP Concentration ................................ ............................... 80 Results and Discussion ................................ ................................ ........................... 80 Preliminary Analysis of CAP ................................ ................................ ............. 81 Selection of Wavelength for UV Spectroscopic Analysis ................................ .. 82 Solubility of CAP in Different Solvent Systems ................................ ................. 83 Recommendations for Preparation of CAP Solutions in EtOH Solvent System ................................ ................................ ................................ .......... 85 Stability of CAP in 10% EtOH Solutions ................................ ........................... 86 Conclusions ................................ ................................ ................................ ............ 88 3 DEVELOPMENT O F NOVEL METHODOLOGIES FOR ANALYSIS OF HUMAN BREATH BY HIGH FIELD ASYMMETRIC WAVEFORM ION MOBILITY SPECTROMETRY (FAIMS) AND MASS SPECTROMETRY (MS) ...................... 102 Introduction ................................ ................................ ................................ ........... 102 Experimental ................................ ................................ ................................ ......... 109 Chemicals and Reagents ................................ ................................ ............... 109 Instrumentation ................................ ................................ ............................... 109 Flavorant Analysis ................................ ................................ .......................... 111 THC Analysis ................................ ................................ ................................ .. 112 Results and Discussion ................................ ................................ ......................... 112 Development of Breath Sampling Methodologies ................................ ........... 112 On line introduction and collection of breath ................................ ............ 112 Simulation of breath ................................ ................................ ................. 114 Flavorant and THC Analysis ................................ ................................ ........... 115 MS direct infusion of standards ................................ ............................. 115 MS breath sampling ................................ ................................ .............. 118 Standalone FAIMS direct detection of standards ................................ .. 123 Standalone FAIMS bre ath sampling ................................ ..................... 126 FAIMS/MS direct infusion of standards ................................ ................. 127 FAIMS/MS ................................ ................................ 128 Conclusion ................................ ................................ ................................ ............ 129 4 METABOLOMICS OF MELANOMA: IDENTIFYING AND CHARACTERIZING POTENIAL BIOMARKERS OF DISEASE BY MASS SPECTROMETRY ............. 170 Introduction ................................ ................................ ................................ ........... 170 Experimental ................................ ................................ ................................ ......... 174 Chemicals, Reagents, and Biological Samples ................................ .............. 174


8 Solvents and standards ................................ ................................ ........... 174 Mammalian cell cultures and media ................................ ......................... 174 Biopsied tissue ................................ ................................ ......................... 177 Protein quantitation ................................ ................................ .................. 177 Instrumentation ................................ ................................ ............................... 178 Biological Sample Prepar ation ................................ ................................ ....... 179 Mammalian cell metabolite isolation ................................ ........................ 179 Biopsied tissue preparation ................................ ................................ ...... 180 Multivariate Data Analysis ................................ ................................ .............. 180 Chromatographic statistical analysis workflow ................................ ......... 181 Non chromatographic statistical an alysis workflow ................................ .. 182 Results and Discussion ................................ ................................ ......................... 183 High Performance Liquid Chromatography, High Resolution Mass Spectrometry ................................ ................................ ............................... 184 First cell culture iteration ................................ ................................ .......... 18 4 Second cell culture iteration ................................ ................................ ........... 185 Chrom atography column improvements ................................ .................. 188 Atmospheric Pressure Chemical Ionization, Tandem Mass Spectrometry ..... 189 Direct Analysis in Real T ime, High Resolution Mass Spectrometry ............... 191 First cell culture iteration ................................ ................................ .......... 191 Biopsied human tissue ................................ ................................ ............. 192 Compound Identification ................................ ................................ ................. 194 Selection of Putative Biomarkers ................................ ................................ .... 197 Conclusion ................................ ................................ ................................ ............ 199 5 CONCLUSIONS AND FUTURE WORK ................................ ............................... 225 Summary ................................ ................................ ................................ .............. 225 Future Work ................................ ................................ ................................ .......... 227 LIST OF REFERENCES ................................ ................................ ............................. 235 BIOGRAPHICAL SKETCH ................................ ................................ .......................... 244


9 LIST OF TABLES Table page 1 1 Advantages and disadvantages betwe en LTQ, QE, and TOF instruments ........ 59 1 2 Co mparison of FAIMS cell features ................................ ................................ .... 70 2 1 CAP concentration in EtOH vs . Tween based solvent systems ........................ 94 2 2 Recommended preparation of 1L CAP solutions ................................ ................ 98 3 1 Comparison of select analytical met hods for use in breath analysis ................. 134 3 2 MS parameters for LTQ experime ntation on human exhaled breath ................ 136 3 3 MS and FAIMS/MS par ameters for LCQ experimentation ................................ 137 3 4 FAIMS cell dimensions and parameters ................................ ........................... 138 3 5 LOD a nd LOQ for Car, MethAn , and Van by APCI/MS on the LTQ .................. 149 4 1 MS and chromatographic pa rameters for QE experimentation ......................... 207 4 2 MS parameters for LTQ e xperimentation on cell cultures ................................ . 208 4 3 MS par ameters for TOF experimentation ................................ ......................... 209 4 4 Comparison of inst rument al methodologies for metabolomics studies ............. 210 4 5 Potential biomark ers for melanoma ................................ ................................ .. 224


10 LIST OF FIGURES Figure page 1 1 Transmittance interferences caused by an a lyte container ................................ .. 51 1 2 Diagram of the double beam geometry spectrophotometer ............................... 52 1 3 APCI process in positive polarity mode ................................ .............................. 53 1 4 Applicable analyte range of several ionization methods ................................ ..... 54 1 5 APCI process for breath sampling in positive polarity mode ............................... 55 1 6 ESI ionization and desolvation process diagram ................................ ................ 56 1 7 EE SI process diagram ................................ ................................ ........................ 57 1 8 Types of liquid liquid droplet interactions ................................ ............................ 58 1 9 Schematic of Thermo Finnigan LTQ XL ................................ ............................. 60 1 10 Stability diagram for ions in a quadrupole mass filter ................................ ......... 61 1 11 Schematic of Thermo Scientific Q Exactive ................................ ........................ 62 1 12 Schematic of Agilent 6220 TOF ................................ ................................ .......... 63 1 13 Drift tube IMS diagram ................................ ................................ ........................ 64 1 14 FAIMS diagram ................................ ................................ ................................ ... 65 1 15 FAIMS with CV diagram ................................ ................................ ..................... 66 1 16 FAIMS ion classification ................................ ................................ ..................... 67 1 17 3D rep resentations of FAIMS cell geometries ................................ .................... 68 1 18 Relationship of frequency and amplitude of FAIMS waveform ........................... 69 1 19 Photographs of The rmo waveform generator and Lonestar ............................... 71 1 20 Photograph comparing Thermo waveform generator to Lonestar ...................... 72 1 21 Various FAIMS wavef orms ................................ ................................ ................. 73 1 22 Conceptual chromatographic peaks produced by LC ................................ ......... 74 1 23 Hypothetical example of PC A applied to a 2 variable dataset ............................ 75


11 2 1 Molecular structure for CAP. ................................ ................................ ............... 90 2 2 Molecular structures of Polysorbate 80 and 20 ................................ ................. 91 2 3 Absorbance of 95% EtOH, Tween solution, and CAP in Tween solution ........... 92 2 4 Absorbance of CAP (50 500 µM) in 95% EtOH ................................ .................. 93 2 5 CAP (200 µM) produces maxima at 227 and 281 nm ................................ ......... 95 2 6 CAP in varying compositions of ethanol ................................ ............................. 96 2 7 Calibrati on curves for CAP in three different solvents ................................ ........ 97 2 8 Stability CAP (200 µM) CAP prepared in 10% EtOH, as a function of time ........ 99 2 9 Stability CAP (350 µM) CAP prepared in 10% EtOH, as a function of time ...... 100 2 10 Stability CAP (500 µM) CAP prepared in 10% EtOH, as a function of time ...... 101 3 1 Diagram of pulmonary alveoli ................................ ................................ ........... 132 3 2 Diagram of molecules transferring from blood to exhaled breath ..................... 133 3 3 Molecular structures of flavorants and THC ................................ ..................... 135 3 4 Photographs of Intoximeters mouthpiece ................................ ......................... 139 3 5 Comparison of FAIMS cell s utilized in this research ................................ ......... 140 3 6 3D and 2D representations of the µFAIMS chip in the Lonestar ....................... 141 3 7 Photo of the breath simul ation system ................................ .............................. 142 3 8 Diagram of the breath simulation system on the LTQ ................................ ....... 143 3 9 ESI/MS spectra for flavorants on LTQ ................................ .............................. 144 3 10 APCI/MS spectra for flavorants on LTQ ................................ ........................... 145 3 11 Calibration curve for Car ................................ ................................ ................... 146 3 12 Calibration curve for MethAn ................................ ................................ ............ 147 3 13 Calibration curve for Van ................................ ................................ .................. 148 3 14 APCI/MS/MS spectra for flavorants on LTQ ................................ ..................... 150 3 15 APCI/MS spectra for THC on LTQ ................................ ................................ .... 151


12 3 16 Chromatograms demonstarting breath collection on LTQ ................................ 152 3 17 Exhaled breath vs simulated breath ................................ ................................ . 153 3 18 APCI of breath with and without use of auxiliary solvent ................................ .. 154 3 19 Simulation of breath with silanized of glassware ................................ .............. 155 3 20 Diagrams of neat standard and breath introduction into Lonestar .................... 156 3 21 Dispersion plots illustrating lack of effective data outside 25 to 75% DF .......... 157 3 22 Dispersion plots of flavorants on Lonestar ................................ ........................ 158 3 23 Dispersion plots of air vs. Van illustrating DF of optimal separation ................. 161 3 24 Plots of signal intensity vs. CV for flavorants at a specific DF .......................... 162 3 25 Real time monitoring of standards and mixtures by standalone FAIMS ........... 166 3 26 Real time detection of breath by standalone FAIMS ................................ ........ 167 3 27 Photographs of ESI and EESI/FAIMS/MS setups ................................ ............. 168 3 28 Detection of THC by ESI and EESI/FAIMS ................................ ...................... 169 4 1 Normal vs. mutated cell division ................................ ................................ ....... 202 4 2 Stage progression of melanoma ................................ ................................ ....... 203 4 3 5 and 10 year survival rate s after melanoma diagnosis ................................ .. 204 4 4 Visual interpretation of skin lesions ................................ ................................ .. 205 4 5 ................................ ................................ .............................. 206 4 6 PLS DA scores and loadings plots from UHPLC/HRMS of cell cultures ........... 211 4 7 PCA scores and loadings plots from UHPLC/HRMS of cell cultures ................ 212 4 8 PCA and PLS DA scores plots from UHPLC/HRMS of cell cultures ................. 21 3 4 9 HCA for UHPLC/HRMS of cell cultures using ANOVA/t test modeling ............. 214 4 10 HCA for UHPLC/HRMS of cell cultures using Random Forest modeling .......... 215 4 11 MDA plot for UHPLC/HRMS of cell cultures ................................ ..................... 216 4 12 VIP scores plot for UHPLC/HRMS analysis of cell cultures .............................. 217


13 4 13 T IC comparison of AN vs. XM ................................ ................................ .......... 218 4 14 PCA and PLS DA scores plots from APCI/MS of cell cultures .......................... 219 4 15 PLS DA loadings plot and MS2 spectra from APCI/MS of cell cultures ............ 220 4 16 PCA and PLS DA scores plots from DART/HRMS of cell cultures ................... 221 4 17 PLS DA scores and loadings plots from DART/HRMS of skin tissue ............... 222 4 18 Workflow for compound identification ................................ ............................... 223 5 1 DESI process diagram ................................ ................................ ...................... 233 5 2 LAESI process diagram ................................ ................................ .................... 234


14 Abstract of Dissertation Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy INVESTIGATION AND DEVELOP MENT OF NOVEL METHODOLOGI ES FOR DISEASE DIAGNOSIS AND DRU G MONITORING UTILIZING MASS SPEC T R OMETRY AND FIELD ASYMMETRIC ION MOBILITY SPECTROMETRY By Michael Thomas Costanzo August 2015 Chair: Richard A. Yost Major: Chemistry In a society of ever increasin g medical costs and invasive procedures, the development of quick, cheap, and painless diagnostic tests is crucial for not only the doctor, but also the patient. By utilizing ambient analytical techniques, medical professionals could directly analyze sampl es in a more efficient, cost effective manner, more applicable to the clinical se tting; i.e., no need to send off to the lab. By cutting down on analysis time and cost, and increasing accessibility of diagnostic results to non experts, visits to physicians would become quicker and more beneficial to patients who would no longer have to submit to a variety of disease panels and wait days or weeks for results. Mass spectrometry (MS) has long been used for the identification of compounds, as both a standalone instrument and coupled to any variety of supplemental techniques. This research employed several different types of mass spectrometers and numerous ionization methods (e.g., APCI, DART, EESI, ESI) to identify compounds of interest for multiple diagnostic applications. MS can prove especially valuable when coupled to ultra high performance liquid chromatography ( U HPLC) for global analysis methods.


15 Utilizing chromatographic separation in conjunction with mass spectrometric analysis can produce a list of thou sands of compounds ; multivariate data analysis can potentially reduce this list to only those that may be of importance to a specific study. Moreover, MS can be combined with high field asymmetric waveform ion mobility spectrometry (FAIMS) to perform targe ted analysis on select compounds of known significance. Standalone FAIMS has the added benefit of portability, providing an easy transition to the clinical se tting. With this in mind, FAIMS/ spectra that correlate t o specific analytes of interest, then MS could be decoupled and standalone FAIMS could provide rapid, noninvasive results in and outside of the lab. The primary focus of the work presented will be two related projects , comprehensive metabolomics of malign ant melanoma by U HPLC and direct analysis MS, and analysis of human breath by FAIMS and FAIMS/ MS. In addition, the use of UV spectroscopy to characterize capsaicin solutions will be described.


16 CHAPTER 1 A BRIEF INTRODUCTION TO A VARIETY OF BIOMEDICALLY APPLICABLE ANALYTICAL TECHNIQUES AND METHODOLOGIES Ultraviolet Visible Molecular Absorption Spectrometry Background Ultraviolet/Visible (UV/Vis) spectroscopy is used for analyte quantitation in a variety of scientific disciplines, including biology, and in organic and organic chemistry. 1 This wide applicability can be attributed to its sensitivity to the basic electr onic transitions of molecules. A bsorption band broadening due to vibrational, rotational, and translational interactions is a defining characteristic of UV/Vis. 2 It is this broadness and high interaction probability that simultaneously limits qualitative analysis, yet allows UV/Vis to excel at quantitation of chemical species. 2 In essence, UV/Vis is not ideal for identificati on of an unknown mixtur e, but is profic ient in accurately quantifying a targeted analyte. A portion of this research employed the use of UV/Vis molecular absorption spectrometry to measure the quantity of capsaicin present in solutions comprised of varying solvent systems. More accurately, the region of the electromagnetic spectrum monitored wa s exclusive to the UV range, 180 40 0 nm. 1 Principles of Molecular Absorption S pectrometry Fundamentally, molecular absorbance spectroscopy is based on a n equation 1 : A = log T = log = (1 1) Transmittance ( T ) and absorbance ( A ) are a function of the incident radiant power ( P 0 ) applied to a sample compared to the transmitted radian t power ( P ) detected after


17 attenuation through the sample. By measurement of the T or A of a solution contained within a transparent cuvette, of path length b , the concentration of an analyte ( c ) may be linearly calculated based on the molar absorptivity ( ). 3 Transmittance and absorbance are not measured absolutely due to significant reflection and scattering effects as a result of interferences from the container and solution molecules, as shown in Figure 1 1. In a typical cuvette with a square cross section, there are two air wall interfaces and two wall solution interfaces through w hich the light beam must pass. At each of these four stages, a substantial amount of incident light is lost by reflection. 1 Moreover, larger molecules present in the analyte solution may scatter the incident light beam, or some light can be absorbed by the container walls; both result in further beam attenuation. 1 Hence, there is a secondary path, called a reference path, for the incident beam to travel whereby the light is measured a fter transmittal through an identical cell containing only solvent (Figure 1 2). This is what is referred to as a double beam instrument, and provides automatic correction of the calculated T and A of an analyte. The Hewlett Packard 8450A UV/Vis spectropho tometer utilized in this work computes the absorbance of a sample according to following equation, itself a modified A = log B (1 2) where S is a measurement of the light reaching the detector after passing thr ough the sample, R is a measurement of the light after transmittance through the reference path, and D and B are corrections for the dark current and the degree of balance between the two paths. 2


18 Ambient Ionization Mass Spectrometry Ambient ionization involves the formation of ions in a source outside of the mass spectrometer, without the need for sample preparation or separation. 4 , 5 Typically, ambient ionization occurs at room temperature and pressure, but may involve application of a heated spray or gas stream. 6 8 Nevertheless, while analyte ionization occurs at atmospheric pressure, the mass spectrometer must be ma intained under high vacuum. Examples of ambient ionization include ubiquitous techniques such as atmospheric pressure chemical ionization (APCI) and electrospray ionization (ESI), in add ition to more novel methods such as direct analysis in real time (DART ), desorption electrospray ionization (DESI) and extractive electrospray ionization (EESI). All of these methods can be used on a variety of mass spectr ometers, each with its own inherent advantages to more efficiently ionize certain species, further expan ding the range of analytes that can be studied by mass spectrometry. These techniques, as well as some common mass analyzers, will be discussed further below. Ionization Techniques Atmospheric pressure chemical i o nization Atmospheric pressure ionization (A PI) was first reported in the early 1970s by Horning et al., 9 , 10 where they utilized both 63 Ni and corona di scharge to create ions for mass spectrometric analysis. Today, API is a blanket term encompassing numerous techniques, whereas corona discharge APCI is the term reserved for gas phase ion molecule chemical reactions at atmospheric pressure, and is currentl y one of the most common ionization techniques utilized for mass spectrometric analysis. More specifically, APCI is the process of gas phase chemical ionization (CI), whereby reagent ions are initially formed from ionization of air and solvent molecules,


19 t ypically using corona discharge. These reagent ions subsequently react with analyte molecules to produce analyte ions (Figure 1 3). 11 The underlying mechanism for positive ion mode is the result of a series of chemic al reactions , 12 beginning with electron ionization (EI): e + N 2 N 2 + + 2e (1 3) e + H 2 O H 2 O + + 2e (1 4) Nitroge n and water found in air far outweigh the relative abundance of solvent and analyte present; hence, most EI reactions involve these two species more than any other. Next, secondary ions form from interactions with the nitrogen and water reagent ions: N 2 + + H 2 O H 2 O + + N 2 (1 5) H 2 O + + H 2 O H 3 O + + OH (1 6) Again, the nitrogen and water cations could react with any other molecule present, including solvent and analyte. It is these secondary ion reactions that are responsible for the higher probabili ty of [M·] + ion produced in APCI over other ionization techniques. However, formation of [M+H] + drastically higher abundance compared to all other species present. Therefore, a third chemical reaction, proton transfer, usually takes place to ionize the analyte molecules: H 3 O + + M [M+H] + + H 2 O (1 7) Negative ion mode typically involves formation of the [M H] ion as a result of proton abstraction by OH . 12 fragmentation, and is often used to analyze less polar compounds of a low molecular


20 weight and high er thermal stability. Moreover, Figure 1 4 plots the applicable ranges of several API and vacuum pressure ionization methods as a function of analyte polarity and molecular weight. The best sensitivity is achieved at higher liquid flow rates than those typ ically used for ESI (i.e., 100 µL/min 1 mL/min). 12 Also unlike ESI, analyte ions do not need to exist in solution. Consequent ly, APCI is ideal for breath analysis. API was one of the first techniques used for breath analysis, and remains a popular choice today due to its robust nature. Unlike more solvent dependent methods, such as ESI, modifying or changing the solvent system d oes not have a profound effect on analyte sensitivity. 11 , 12 Instinctively, this may seem like a weakness since optimizing the solvent to improve limits of detection is often a precur sory step taken in MS analysis; however, the matrix of breath not only varies from human to human, but also from breath to breath in a single individual. 13 Therefore, having a method that is less susceptible to subtle solvent changes is a distinct advantage. APCI of breath has a similar set up to conventional APCI, but vaporized solvent is not required (Figure 1 5). Moreover, utilizing corona discharge APCI for breath detection adds portability not offered by any other technique, other than 63 Ni radiation. Electrospray i onization The most versatile of all ambi ent ionization techniques, ESI can ionize the widest range of analytes. Thermally labile and high molecular weight species, previously unsuitable for mass analysis, can be analyzed by ESI. As long as the analyte can exist as a preformed ion in solution, it can be ionized by ESI. 12 Truthfully, ESI is more of an ion extraction technique than an ion forming technique. To produce anal yte ions, a sample solution is first passed through a needle, held at a high voltage relative to a counter electrode (usually the entrance to the MS). The


21 sample solution is sprayed into a fine mist of droplets that are electrically charged at the surface. The net charge on the droplets depends on the polarity of the preformed ions in solution; i.e., acidic molecules form negative ions whereas basic molecules for positive ions. Changing the polarity on the needle and counter electrode allows different types of ions to be analyzed. As the solvent evaporates from the droplets, the charge density at the surface increases until reaching a critical point, known as the Rayleigh stability limit. 11 The coulombic repulsion at t he surface exceeds the surface tension of the droplet, causing the droplets to split and create smaller, more stable droplets. Desolvation and fission continue until very small, highly charged droplets are formed. From these extremely small droplets, analy te ions are ejected into the gas phase by further electrostatic repulsion. There are two key theories which explain this final stage of gas phase ion production: the ion evaporation model (IEM) and the charge residue model (CRM). In short, the IEM propose s field assisted ion desorption, implying that upon reaching a certain radius, the field strength at the droplet surface becomes large enough to promote desorption of the solvated ions. 14 Alternatively, the CRM states that the droplets continue to undergo evaporation and division until a single analyte ion remains to be desolvated. 15 analyte ion. Evi dence suggests that both are at least partially true, where small ions are liberated by the IEM and larger ions by the CRM. 16 , 17 The basic ESI process is detailed in Figure 1 6. Like APCI, ESI is considered a soft ionization technique. Unlike APCI, ESI affords the greatest sensitivity at low flow rates on the order of nL/min, also known as


22 nano ESI. Moreover, ESI has a wider dynamic mass range, which can be attributed to the production of multiply charged species. Realistically, mass spectrometers afford mass to charge analysis, rather than strictly mass analysis. Therefore, having more than one char ge on a higher mass ion allows that species to be monitored in the same m/z range as a singly charged lower molecular weight species. The number of charges carried by an ion depends on the structure of the analyte and solvent molecules. 11 Consequently, variability in droplet size, surface charge and tension, solvent volatility, and solvation strength are all a function of the analyte of interest and the carrier solvent. 12 Hence, ESI is drastically affected by a change in the solvent system or the levels of b uffers or electrolytic species as they cause ion suppression, a crucial factor in signal loss due to competition and interferences between analyte and background matrix. Likewise, modifying the geometry of the electrospray system has a profound effect on t he ionization mechanism and efficiency. A primary example of this is extractive electrospray ionization, an alternate ESI technique developed in 2006 by Cooks et al. 18 Rather than direct desolvation of a preformed ion, EESI relies on liquid liquid extraction of the analyte species from an auxiliary sample spray into the charged solvent microdroplets. The sample solution is nebulized (or aerosolized) separately f rom the ionized solvent, then the two sprays are directed toward each other and the MS inlet (F igure 1 optimized.


23 EESI offers extraction and ionization of analyte from raw samples typically 19 Samples such as blood, urine, and breath have complex matrices that often suppress i on signal. Whether from adduct formation, sample carry over, or instrument contamination there is a definitive loss in sensitivity. Dried blood spot testing and i ntroduction of the sample off axis to the MS inlet have addressed this problem for certain typ es of analysis, but not all, and performing the necessary pretreatment steps can be cumbersome. 18 Conversely, EESI allows continuous analysis of these unaltere d samples at a sensitivity equal to, if not greater than, ESI. 19 , 20 This is a result of the more selective na ture of EESI over traditional ESI. While the exact mechanism at work in EESI is unknown, studies suggest liquid phase interactions dominate. 18 , 21 , 22 If this hypothesis is correct, several processes may occur during the ion molecule interactions: bounce, disruption, fragmentation, and total coalescence (Figure 1 8). 23 , 24 During a bounce, the droplets do not make surface contact due to the presence of a thin gas layer between them that remains unbroken, 24 hence no analyte extraction should occur. However, during disruption and fragmentation, selective analyte extraction is expected, giving way to secondary solvated analyte ions which are then released from the droplets by the traditional ESI mechanisms. 22 Disruption and fragmentation consist of a partial coalescence of the droplets, where the kinetic energy is too high to permit total coalescence. They differ in that disruption reforms droplets similar in size and mass to those pre collision, while fragmentation involves catastrophic breakup into numerous smaller droplets. 24 The final


24 possibility is total coalescence, where both droplets combine into a single, larger droplet of mass equal to the sum of the original droplets. 23 Ideally, disruption and/or fragmentation would occur as these processes are what improve the extraction selectivity associated with EESI. Contrary to ESI, EESI would have a 2 fold dependence on analyte solubility, needing to account for both the ES I and sample spray solvents. Therefore, the solubility is significant in determining extraction efficiency from the sample spray. For effective liquid liquid extraction to occur, the compound of interest must be more soluble in the ESI solvent. 22 This implies that selective extraction occurs between charged ESI droplets and the sample spray, whereby the analyte could be easily drawn into the ESI spray while unwanted compounds remain unionized in the sample spray; therefore unable to disrupt analyte detection. Conversely, in total coalescence, no extraction transpires. Thus, the solvent miscibility between the two sprays and the subsequent surface tension between the droplets are critical to the extent of coalescence. 22 Mass A nalyzers While ionization may occur at ambient pressure, the region a fter the source must be evacuated to lower pressures in order to effectively perform mass spectrometry. 11 A vacuum system is necessary to prevent analyte ions from colliding and interacting with gas molecules, subseq uently altering their velo city or undergoing a reaction, preventing the analyte from reach ing the detector. Therefore, the longer the mean free path (mfp), the average distance a molecule may travel before striking another molecule, the less likely this wi ll occur. Ideally, the mfp is much, much larger than the distance from the source to the detector. Additionally, maintaining near vacuum pressures allows the


25 instrument to keep high voltages on several components without arcing, and minimizes spectral cont amination. Typically, this is achieved by way of rotary vane pumps which provide a drop in pressure to around 10 3 Torr over the length of the ion optics region. 12 Furthermore, this region is often comprised of a series of skimmers and flow restrictors to allow efficient ion transmission, while removing solvent vapors and neutral species. The use o f turbomolecular pump(s) are then employed to reach high vacuum in the mass analyzer region. The pressure of the analyzer region, and number of turbomolecular pumps necessary to reach them, is determined by the analyzer geometry, and is characteristically around 10 7 Torr, 25 but may be as low as 10 11 Torr (i.e., the orbitrap) 26 or as high as 10 5 Torr (i.e., the ion trap) 12 . W hile all mass analysis must be performed under vacuum, the analyzer geometries vary widely in the design, application, and mass analysis. Table 1 1 provides a comparative look at the advantages and disadvantages of using the following three geometries. Ion trap The primary mass spectrometer used in this work, a Thermo Scientific LTQ XL, employs a linear quadrupole ion trap (LIT) mass analyzer geometry. An LIT consists of an array of four electrodes utilized to trap ions, manipulate their trajectories, and s electively eject them by their m/z . Ionization of an analyte occurs outside the source interface, before the ions travel from this region toward the ion guides by way of a decreasing pressure gradient. A continued pressure differential, and a series of cha nges in potential, carry the ions along this interface and then through three ion guides, consisting of a succession of ion optics, quadrupoles, and an octapole, before reaching the linear ion trap (Figure 1 9). Upon reaching the trap, the ions are stored


26 until they are radially ejected by linearly ramping the main radio frequency (RF) voltage, to collect a mass spectrum. 12 The en tire population of ions in the trap may be scanned out, or a specified m/z range may be isolated and fragmented before being scanned out. This fragmentation can be applied to a single mass range (MS 2 ), or multiple masses consecutively (MS n ) due to the capa bilities as a result of the trapping region design . 27 Collision induced dissociation (CID) is the trigger for fragmentation in the ion trap. Once the ions are isolated in the trap, the main RF remains at the same frequency while an additional DC waveform is set to the resonance excitation frequency of the specified m/z , enhancing the ions motion in the radial direction as it gains kinetic energy. 12 As the energy increases, the num ber and intensity of collisions between the ions and helium damping gas also increases, until the ions gain enough internal energy to fragment and form product ions. After fragmentation, the ions are subjected to the same radial ejection as before by re ach ing the resonant ejection voltage, and a product ion mass spectrum is collected. 12 To better understand how ions may be trapped and ejected by a quadrupolar field, refer to the Mathieu equations. 27 In brief, the stability of an ion can be plotted as a function of two parameters, a and q , as defined by Equations 1 8 and 1 9. 11 This plot is commonly denoted the Mathieu stability diagram, the p ositive portion of which is displayed in Figure 1 10. The first parameter, a , relates to the constant potential (or direct current, DC) bias ( U ), while the second parameter, q , is as sociated with the RF potential ( V ). 27 Both are influenced by the electronic charge ( e ), mass ( m ), electrode radius ( r 0 ), and RF angular frequency ( RF ). 11 a = (1 8)


27 q = (1 9) Several conclusions can be drawn from consulting this diagram. First, any ions relating to coordinates found within the shaded region are set upon stable trajectories to reach the detector. Second, if the instrument is set to operate with a defined DC and RF , as represented by the operating line in the diagram, only those ions with m/z values c orresponding to the tip of the shaded space will be ejected and subsequently detected. 11 a = 0 and no DC field is applied, thus, the operating line lies alo ng the abscissa. 12 Orbitrap The second ion trapping mass spectrometer used in this work is also manufactured by Thermo Fisher S cientific, however it varies from the LTQ in many ways. The Thermo Scientific Q Exactive houses an orbitrap mass analyzer, which is the latest development in trapping devices utilized in MS research. An orbitrap is axially symmetrical and comprised of bell shaped outer electrodes that encompass an inner spindle shaped central electrode. Classically, ions analyzed by LIT and other trapping geometries are ejected from the trap and physically destroyed by contact with a conversion dynode in order to measure th eir signal. 27 An orbitrap utilizes Fast Fourier Tra nsformation (FFT) of the amplified image current produced by the trapped ions. 26 Hence, it is a less destructive tec hnique than conventional trapping methods, as well as one that offers extraordinary resolving power and dynamic range. Another key difference is the presence of a second, curved linear trap, called the C trap. Unlike the dynamic electric field afforded by an LIT, an orbitrap only maintains a static field; thus, trapping ions would be nigh impossible with the orbitrap alone. 11 Rather, such a device


28 opts for a specialized dynamic injection p ulse in conjunction with the C t rap. Moreover, the QE houses a quadrupole mass filter for ions to pass through before the orbitrap, thus, the orbitrap is a tandem in space instrument, distinguishing it from the L TQ which is tandem in time . Finally, an orbitrap also requires substantia lly lower pressures to op erate, much lower than all other ion traps, as well as most mass analyzers in general . A schematic of the Q Exactive can be seen in Figure 1 11. As before, an in depth exploration into the intricate mechanisms inv olved in orbitrap technology is easily obtained elsewhere, 11 , 26 and will only be discussed briefly here. Prior to entering the orbitrap, the ions ar e initially focused into the C t rap, wher e they are held (a.k.a. trapped) by the RF and DC voltages applied to the cell. 26 For ion extraction, the RF is scal ed down and the DC potential ramped up, electrodynamically squeezing the ions into discrete bundles and accelerating the packet out of the C t rap. 11 Due to a combination of several features: the curvature of the tra p, presence of subsequent lenses, and differential pumping; the ions are spatial ly focused into the orbitrap while also eliminating gas carryover. 26 The orbitrap is symmetrical, yet the ions are injected off center from the C t rap, in the direction of the center electrode. The ions arrive as a tiny packet with a kinetic energy equivalent to the opposing potential energy of the radial electric field between the two asymmetric surfaces. 11 The static field in the orbitrap sets the ions on a circular path between the inner and outer electrodes, the radius of which is defined by the bal ancing of centripetal forces acting on the ions, independent of mass. 11 Thus, the radius of orbit is the same for all ions, regardless of mass. Furthermore, due to the inhomogeneous yet symmetrical nature of the fiel d in the orbitrap, the ions naturally oscillate axially, along the length of the center electrode, 26 as


29 described by the following equation; where is the f requency of these oscillations and k is the instrumental constant : (1 10) As demonstrated by Equation 1 10, the oscillations are mass dependent, hence a mass spectrum can be collected based on the image current observed from frequency. As a side effect of this dependence the resolution is enhanced at lower masses, a result of the fact that smaller ions oscillate more than larger ions. Time of flight The third, and final, analyzer utilized in this research is a time o f flight (ToF) mass spectrometer. A ToF is inherently diverse from trapping instruments as it relies on measuring the time required for a group of ions to traverse a flight tube given the same initial kinetic energy, 11 thus differentiating ions by their separation in space rather than systematically ejecting ions simultaneously held in a trap. Furthermore, the flight tubes are typically meters long compared to the centimeter scale of most traps. There are sev eral sub types of ToF analyzers; the Agilent 6220 uses orthogonal acceleration ( oa ToF) to pulse a cloud of ions down the flight tube perpendicular to the direction they enter the mass analyzer. 25 Moreover, this instrument is a reflectron ( re ToF) mass spectrometer, utilizing an ion mirror to reflect ions with equivalent m/z values but dissimilar energies compelling th em to arrive at the detector simultaneously; thus, improving the resolving power. 11 , 25 Many commercial ToF instruments exploit both oa ToF and re ToF geometries concurrently. Figure 1 12 displays a schematic of the Agilent 6220 ToF system.


30 Going into more detail, ions are continually focused through the optics region of the instrument, however, upon reaching the analyzer region, only small packets of this ion beam are accelerated orthogonally toward the reflectron on the opposite end of the flight tube. As stated, all ions are initially imparted with the same kinetic energy. In addition, they retain a slight forward momentum in the direction of the original ion bea m to ensure the ions reach the detector (mounted adjacent to the pulser) as opposed to returning to the pulse region after reflection by the mirror. 25 It is their flight along the field free drift region where ions of different m/z begin to diverge. Since they all received the same energetic push down the region, any reduct ion in velocity correlates to that ion having a greater mass . The longer an ion takes to reach the detector, the larger its mass. Hence, the longer the drift tube, the better the resolution of the instrument. The reflectron effectively doubles the flight path in the same space, as well as p roviding a refocusing effect, both increasing the resolving power achievable. 26 Simply stated, ToF analyzers involve discrete bunches of ions sequentially striking a detector in order of increasing m/z value. 11 The relationship between the mass of an ion and its time through the flight tube can be illustrated by a series of fundam ental equations relating velocity and mass. The kinetic energy imparted upon an ion ( KE , Equation 1 11) is equal to the potential energy ( PE , Equation 1 12) of the pulser field through which the ion beam passes. KE m ) and ve locity ( v ), while PE z ), the charge of an electron ( e ), and applied voltage of the acceleration field ( V ): KE = ½ mv 2 (1 11) PE = zeV (1 12)


31 As the ions are accelerated, essentially all of the potential energy is con verted to kinetic energy, 11 thus, Equations 1 11 and 1 12 may be set equal to one another and rearranged to show the inverse relationship between velocity and m/z : (1 13) However, direct measurement of velocity is unfeasible, 11 therefore the time it takes for an ion to traverse the flight tube ( t ) is determined by its dependence on velocity and the tube length ( L ): (1 14) The geometry of a ToF analyzer is responsible for both its greatest advantage and biggest pitfall. Theoretically, there is no upper mass limit, and while an infinite mass analyzer is not achievable in practice, the ToF does offer the greatest mass range of any analyzer. Conversely, due to its reliance on drift tube length, the instrument s resolution relies on large, cumbersome instrumentation. The size can be a detriment for many reasons, but one major drawback is i capabilities; typically limited to one or two stages of mass analysis , achieved by coupling a second mass analyzer in series (typically a quadrupole mass filter or second TOF) . Compound Identification Mass spectrometry is one of the principle tools used for analyte identification, whether targeted or untargeted. As detailed above, a mass spectrum provides an to charge ratio, allowing identification and structural elucidation, depending on the type of mass analysis performed. Two primary strategies may be


32 used: high resolution mass spectrometry (HRMS) from accurate mass measurements, or tandem mass spectrometry (MS n ). Each possesses their own strengths and weaknesses, detailed below. A quick, overall summa ry of the two may be found at the end of the tandem MS section. Accurate mass HRMS involves utilizing an instrument capable of collecting accurate mass data. That is, mass spectral peaks which provide information about the mass defect of an ion, thus, affo rding calculation of the empirical formula. The mass defect is the difference between the exact mass (the mass with isotopic consideration) and the integer mass of a compound. 11 ToF and FT mass analyzers often employ mass accuracy in their detection, whereas conventional ion traps do not offer high enough r esolving power , instead centering on MS n capabilities. As mentioned, information about the empirical formula can be gleaned from HRMS, but it is not adequate for pr oviding structural explication. Therefore, accurate mass is incredibly beneficial for untargeted analysis as it begins the arduous process of identifying potential formulas when presented with countless unknown components in a sample. Tandem mass spectrome try On the other end of the identification spectrum from accurate mass is MS n , where structural elucidation is a fundamental attribute. Instruments only capable of nominal mass analysis, such as the LTQ, tend to utilize MS n to offset the lack of mass spect ral isotopic and defect information; hence ion traps are the most common instruments for MS n analysis. However, tandem MS may be performed on HRMS instruments, but other factors tend to limit these instruments to two stages at most (MS 2 ). By breaking up an ion into its product components, information relating to its structure is revealed based on


33 key fragmentation patterns. There are numerous online databases which allow comprehensive comparison to various fragment patterns for thousands of ions. 28 30 Therefore, MS n is most useful for targeted analysis. It should be noted that a combination of HRMS and MS n would provide the most comprehensive analysis possible when conducting large scale compound identifica tion of unknown species (e.g., hybrid quad orbitrap and quad TOF geometries both afford MS 2 ) . HRMS could initially deliver potential empirical formulas, which may then be validated by MS n . Moreover, if the list of potential compounds and formulas is too in clusive, MS n could narrow down the list, while the complimentary accurate mass data could validate. In this work, MS 2 was used in small part in Chapter 3 to distinguish compounds in breath, while both methods were utilized extensively in Chapter 4 as untar geted compound identific ation is the foundation of the m etabolomics research presented. High Field Asymmetric Waveform Ion Mobility Spectrometry FAIMS Principles of IMS and FAIMS This research utilizes high field asymmetric waveform ion mobility spectrom etry (FAIMS), also called differential mobility spectrometry (DMS), a filtration technique that uses two uniformly distanced plates to impart an asymmetric electric field onto analyte ility under high and low electric fields. As an analytical technique, FAIMS is able to separate ions in the gas phase on a time scale dramatically shorter than can be achieved by conventional chromatography. Nevertheless, in order to appreciate how FAIMS o perates, it is crucial to first understand the concepts of ion separation associated with ion mobility spectrometry (IMS).


34 Ion mobility spectrometry, as with FAIMS, separates gas phase ions based on their inherent mobility through a carrier gas. 31 The brainchild of Earl McDaniel, IMS was developed through a series of studies on ion molecule reactions in gases during the 1950s and 60s. 32 35 In traditional IMS, ions are pulsed through a drift tube by way of an applied electric field i n the presence of a buffer gas which opposes the ions motion (Figure 1 13). As the field is held constant, the velocity, and therefore the mobility of an ion , is dependent upon its mass, charge, size, and shape as it traverses the drift tube and collides w ith the buffer gas molecules. 36 This is illustrated by the following equation, where ions that are subjected to an electric field (of strength, E ) in the presence of a drift gas will exhibit a velocity ( V ) tha K ) through the gas: V = KE (1 15) Hence, ions of differing size and shape may be separated, as the velocities of the ions through the drift tube will differ, as long as they are subjected to the same electri c field . 37 However, this concept is only true in low field applications. At low fields, the mobility consta nt is independent of the applied electric field. 38 However, as the elect ric field increases , the velocity does not increase proportionally with field strength, therefore the mobility of an ion becomes field dependent. 36 This nonlinear relationship between mobility and electric field intensity was first de scribed by Buryakov et al. , 36 and is the basis for the operational principals of FAIMS. A high field is typically reached at, or above, approximately 10,000 V/cm, compared to a characteristic low field strengt h of about 200 V/cm. 39 The following equation represents the mobility of an ion in a high field ( K h ):


35 (1 16) where K represents the mobility of the ion at a low field, and are compound sp ecific N is the gas number density of the carrier gas. FAIMS is a measure of the difference in the low field and high field mobility of an ion. This difference is demonstrated b y the application of an asymmetric waveform to ions as they drift through the FAIMS cell (Figure 1 14). Rather than the drift tube used in IMS, FAIMS employs two uniformly spaced electrodes (which can take the form of several distinct geometries, discussed in more detail later). One electrode is grounded (or near ground), while an asymmetric waveform is applied to the opposite electrode. The waveform is described by the application of a dispersion voltage (DV) to create a high field environment for a short amount of time, t high . The opposite polarity is then applied, creating a low field environment for a greater amount of time, t low , than for high field. 37 Furthermore, the waveform is applied orthogonally to the plates; contrary to the application of an electric field in parallel with the drift tube to propel the ions down the length of the tube, as seen in IMS. 37 For instance, in Figure 1 14, the ions have a greater mobility during t high , causing them to drift faster toward one of the plates. When the field is switched (t low mobility decreases, so they begin to drift toward the opposite plate. Due to the asymmetry of the waveform, there is a net displacement of the ions after one period of t he field as they will not traverse the same vertical distance when in low field as they did under high field. 37 With each continuous period, the ions will move closer and closer to one of the electrodes until they make contact and are annihilated. To compensate for


36 the net displacement and to allow ions of interest to reach the detector, a secondary, DC voltage is applied; this is called the compensation voltage (CV). Figure 1 15 illustrates how CV affects ions imparted with an asymmetric FAIMS waveform. Therefore, it is the oscillation of the ions under the asymmetric waveform that affords the fundamenta l separation in FAIMS. Conversely, IMS utilizes the buffer gas to oppose the ion motion, thus creating a difference in velocity (and mobility) as a function of cross sectional area. FAIMS also employs the use of a carrier gas, however, it is utilized to dr ive the ions through the cell in addition to bombardment. There are three types of ions, which differ in their mobility as a function of electric field. 38 K h /K , as a function of field strength, these difference can be illustrated (Figure 1 16). Type A and C ions, such as the red and green ions shown in Figure 1 15 , respectively, demonstrate opposing relationships between mobility and electric field strength. For type A ions, as the field strength increases so does the mobility. Whereas for type C ions, the mobility decreases with increasing electric field. For type B ions, as the field strength increases, the mobility increases initially before eventually decreasing. In truth, all type A ions will eventually become type B ions if at a high enough electric field. Modifications to C onventional FAIMS Cell geometry As m entioned above, FAIMS utilizes two uniformly distanced plates, however the geometry of these plates may vary. 37 , 39 41 Figure 1 17 illustrates the three main geometries, including planar, cylindrical, and spherical (or hemispherical). Planar cells, consisting of two parallel plates, provide relatively better resolving power and selectivity than the other geometri es. 40 However, low transmission of i ons


37 through the cell is the greatest limitati on. For a curved cell geometry, e.g., cylindrical or spherical, the electrodes remain uniformly separated, however, they are no longer flat, parallel plates. The electrostatic focusing effect afforded by a curve d geometry allows for improved transmission. 40 42 Whereas a cylindr ical cell can focus ions in one to two dimension s , a spherical c ell affords focusing in all dime nsions. Miniature FAIMS cells and micromachined FAIMS chip The greatest advantages afforded by FAIMS as an analytical technique are its quickness and portability. These attributes are what will allow breath analysis to become a biomedical force in the clin ical arena, as well as an essential tool for law enforcement. With this in mind, the greatest obstacle to overcome in terms of portability is the size of the power supply and waveform generator; both are responsible for application of DV and CV to the FAIM S cell. As the cell increases in size (i.e., the analytical gap increases), the voltage requirement to separate ions on that cell also increases. Subsequently, the power needed to apply the waveform will similarly rise, hence the larger the supply and gene rator necessary. To combat this power consumption, the voltage applied across the cell should be as low as possible. Therefore, the analytical gap needs to be shrunk to a size that allows for adequate separation with low voltage requirements. This is poss ible because as the voltage is decreased, the frequency of the waveform may be increased to compensate (Figure 1 18). This theoretically allows for the residence time of the ions to remain the same while simultaneously lowering the power requirements, thus decreasing the size of the FAIMS device. Resolving is directly proportional to the residence time of an ion power in the cell . 37 , 43 Therefore, a reduction of time spent applying the waveform to the ions equates to a reduction in the resolving power of the


38 ions. In practice, it is not so simple, but st rides are being made toward an ideal device that marries selectivity with portability. 43 46 For exampl e, discussing the FAIMS cells utilized in this work, this process can be described in more detail (Table 1 2). The home built planar FAIMS cell is a more conventional size and has an analytical gap of 2 mm. The power supply used, a Thermo waveform generato r, provides a DV of up to 5,000 V at a typical frequency of 750 kHz. The cell affords resolution on par with chromatography, but lacks portable convenience since the generator has a base of around 13 ft 2 and height of nearly 2 ft. (Figure 1 19A). Another F AIMS cell used is part of a larger device called the Owlstone Lonestar portable gas analyzer. This cell is more accurately defined as a micromachined FAIMS chip, and has an analytical gap of only 0.035 mm. As such, the maximum applied waveform is a fractio n of the full size FAIMS cell at 250 V, but the frequency is up to 28.6 MHz. Due to the limited voltage requirements, the entire device is less than 9 ft 2 , and weighs considerably less than the Thermo generator. (Figure 1 19B). A side by side comparison of the two devices can be seen in Figure 1 20. They have a nearly equal width and length, yet the Lonestar is half the height of the generator. More remarkably, the Lonestar is a complete device, including the FAIMS chip, ionization source, detector, and com puter readout all in one. When using the full size FAIMS cell, from the generator to the FAIMS cell, and the generator currently requires a mass spectrometer for ion detection; an instrument notorious for its size. Presently, the Lonestar is the only commercially available device capable of real time, portable FAIMS analysis, however the resolution and sensitivity are generally


39 inadequate for biomedical use. As it sta nds now, creation of a miniature FAIMS device, Furthermore, this device would hopefully be coupled to a generator that is more compact, a task that is also being undertaken c urrently. If a compromise can be met, a FAIMS device could be created which affords adequate resolving power in quick, concise, and portable manner. Addition of solvent v apor No matter how small or large the FAIMS device, FAIMS as a separation method has i nherently limited resolving power compared to the more common techniques, such as LC and GC. 45 , 47 Ty pical resolving powers of up to approximately 30 m ay achieved by dry F AIMS, 39 which pale in comparison to those seen in chromatographic separation. However, recent work by Rorrer et al. 39 has shown that these number can be improved upon. Typically, when performing FAIMS, the cell is kept extremely d ry, under a constant flow of nitrogen gas. Yet, if solvent vapor were to be added to the carrier gas of the FAIMS cell in controlled amounts, a significant increase in resolving power would result. 45 Rorrer et al. demonstrated that alteration of the high to low field mobility ratio, K h /K , was likely caused by clustering effects due to interactions between the analyte ions and solvent vapor molecules in the cell. 39 When in a low field environment, in addition to having intri nsically lower energy, the ions may undergo a drag effect due to dipole alignment with the orthogonal electric field; thus, reducing their speed and mobility. In the presence of solvated carrier gas, partial gas sol vation of the analyte occurs as the velocity decreases, increasing the cross section of the analyte molecule which further reduces its mobility. Conversely, in a high field environment, the ions have


40 a much higher kinetic energy, and thus a greater velocit y through the solvent vapor. This affords minimal to no solvation and dipole alignment, resulting in lower ion cross sections and greater mobility. This greater change in mobility between high and low fields resulting from the solvation spheres produced by addition of a solvent vapor to the carrier gas is what accounts for the dramatic increase in resolving power from under 20 to over 300. 39 It should be noted that this effect can be saturated as a steady state of solvation will ine vitably be r eached . Therefore, unlimited addition of solvent does not equate to infinite incr ease in resolving power. Additionally, the size of the solvent molecule affect s the increase in resolving power achieved. 39 The bigger the solvent molecule, the l arger the solvation sphere formed, the greater the increase in resolving power. The application of solvent to the carrier gas has proven to improve resolution in a conventional sized FAIMS cell, but further experiments need to be performed into the size of the cell as a function of solvation effects attainable. Some work utilizing a micromachined FAIMS chip similar to that mentioned earlier, 45 , 48 has shown promising results. However, no commercialized microchip has yet to provide resolution remotely close to that achieved on the conventional sized cell. Application of a square wa veform Practically all FAIMS devices utilize a sinusoidal function (Figure 1 21A) to create the asymmetric waveform. Generally, this is a bisinusoidal wave ( Figure 1 21B and C) as a single sine wave is symmetrical and would create no separation. However, t his choice of waveform is inherently poor for mobility separation since it involves a gradual transition between high and low fields, and restricts the user to a single duty cycle, 33% for bisinusoidal. Research performed by Prieto et al. has demonstrated that application


41 of a square waveform (Figure 1 21D) and modification of the duty cycle results in an improvement in resolving power. 44 The reason square waveforms are not ubiquitous already is due to the stress such a dra matic field transition puts on the power supply. Besides the strain of switching from an extremely high field to low (and vice versa), the capacitance and power required to perform such a maneuver is greater than with sinusoidal switching. Therefore, the o nly cells on which square wave FAIMS is practicable are smaller than most conventional sized cells. 44 Square wave separation intrinsically provides better resolution as a product of these instantaneous transition between f ields. Furthermore, no longer hindered by the sinusoidal function, duty cycles outside of 33% could be applied. Thus, square wave application is the direction FAIMS separation should be heading toward as research continues forward. High Performance Liquid Chromatography HPLC Apart from FAIMS, this research exploits one other separation technique known as liquid chromatography (LC). Contrary to the electronic separation imparted upon ions in FAIMS analysis, LC separates a sample into individual components based on physical and chemical interactions with the stationary (SP) and mobile phases (MP). LC is the most widely used analytical separations technique due to its sensitivity, applicability, and ease of automation. 1 LC can be split into many sub types, with the broadest divide encompassing how the physical states of the phases relate to each other. Various combinations of stationary and mobile phases can be employed, but only one will be discussed in this work. Liquid s olid chromatography is the most popular chromatography technique, and involves a liquid mobile phase carrying an analyte


42 through a solid stationary phase. 1 Additionally, more than one device may be used to detect separa ted component, this research will focus strictly on LC/MS analysis. Normal Phase vs. Reversed Phase Chromatography The effectiveness of LC relies on the different affinities that each compound, or component, in a mixture has for the stationary and mobile p hases. Components of different polarities will either be more drawn to the stationary phase causing them to slow their elution through the column, or will drive through the column faster by remaining bonded to the mobile phase. This difference in elution t ime is what creates the characteristic chromatogram produced by LC/MS. The type of adsorbent material used as the stationary phase and the proper solvent chosen for the mobile phase will play a vital role in the separation of the mixture. However, the anal yte of interest, whether a handful of species or all compounds present in a mixture, is just as important in determining which type of chromatography will provide the desired separation as the selection of mobile phases. There are two main types of liquid solid chromatography that may be performed that differ in the polarity of the phases. 1 Normal phase involves the use of a polar stationary phase and less polar mobile phase. The opposite is true for reversed phase (as t he name might suggest), whereby the stationary phase is nonpolar and the mobile phase is of a higher polarity. Generally speaking, i f the analytes are polar, nor mal phase chromatography is the method of choice since the polar components of interest will ad here to the stationary phase while the less polar components will elute quicker. As these polar analytes stick to the column, they are more efficiently separated. In reversed phase, the more polar components elute first, with compounds of increasing polari ty following suit. Recent innovations in stationary phases utilized in reversed phase chromatography have


43 yielded columns that can perform sufficient separation for compounds of both polarities, without needing to switch the column (e.g., pentafluorophenyl or PFP). In many experiments, a gradient elution is performed rather than an isocratic method . This improves the resolution by gradually increasing/decreasing the polarity of the mobile phase, hence, slowly migrating the polarity of the eluate. 49 Other common ways to improve separation efficiencies , and provide more reproducible and rapid chromatography is to increase the pressure of the column, as well as decrease its length and pore size. High Performance Liquid Chromat ography By performing LC at high pressures and decreasing the diameter of the packing diameter, scientists discovered early on that major increases in efficiency could be achieved over traditional gravity flow methods; 1 now commonly referred to as high pe rformance liquid chromatography , or HPLC. The basic rel ationship between retention time and column packing can be best expressed as a series of equations 49 detailed below: t R = t S + t M (1 17) (1 18) (1 19) In Equation 1 17, t M corresponds to the time it takes the unretained species to reach the detector (a.k.a. the void time), t R represents the time it takes the analyte peak to reach the detector (th e retention time), and t S is the time the analyte is retained by the stationary phase; a conjectural example of which is visualized in Figure 1 22. In


44 Equations 1 18 and 1 19, L is the length of column packing, is the mean rate of solute migration, an d u is the average linear velocity of the mobile phase molecules. The void time, t M , is an important parameter in identifying analyte species as it is a measure of the migration rate of the mobile phase, 1 in essence pro viding a blank to reference the analyte peaks against , or to gauge how long it takes an injection to flow through the column . However, to obtain a more quantitative measure of column efficiency one looks to the relationship between packing column length, a nd the height ( H ) and number ( N ) of theoretical plates (Equation 1 20). 49 The efficiency of the column increases as the number of plates increase and as the height of each plate decreases. 50 Dozens of expressions have been proposed to define the calculation of plate height over the last several decades, and while none are perfect, a useful expression for explaining how to improve column efficiency as a function of several path and diffusion effects through the packing parameters of the column is revealed in Equation 1 21, equivalent to the well known van Deemter equation. 49 The first term, A , represents multiple path effects, known as eddy diffusion, responsible for slight peak broadening due to the spontanei ty with which a molecule will travel through unevenly distributed and varyingly sized particles. The second term, B/u , is the longitudinal diffusion coeffic ient, meaning how the molecule is affected by moving from a region of higher flow (center of column) to that of a more dilute flow (sides of the column). The last two terms correspond to the mass transfer coefficients for the stationary ( C S u ) and mobile ( C M u ) phases. (1 20)


45 H = A + + C S u + C M u (1 21) While the mass transfer processes that occur in the mobile phase are too complex to provide a complete quantitative description, a good qualitative understanding of the variables that affect zon e broadening from this effect are better understood. As revealed in Equation 1 22, 1 there is proportionality between this quantity and particle size ( d p ), where D M is the diffusion coefficient of the analyte in the mobi le phase and k is equal to the retention factor. The take home message from a ll of these equations is that C M is directly related to the diameter of the packing particles, thus, efficiency of LC can be improved by decreasing particle size. (1 22) LC and HPLC are used almost interchangeably today, since the popularity of pressurized columns has supplanted the use of gravimetric columns that have typically reserved a high performance liquid chromatography, is becoming increasingly commonplace, providing the new benchmark for most efficient chromatography available. UHPLC is defined by the use of packing material with a pore size under 2 µm, complimentary to the use o f high pressure flow through a column. UHPLC is the method utilized in this work, as detailed further Chapter 4 of this manuscript. Multivariate Data Analysis The past decade has given rise to an increasing number of mass analyzers capable of performing hi gh mass resolution. Moreover, the absolute resolving power attainable by a given instrument has shown a similar increase as the technology of both


46 the hardware and software become more efficient and precise. An improved ability to differentiate a greater n umber of analytes is invariably linked to an increase in complexity and size of samples. Consequently, the time it takes to analyze and mine the pertinent information from these extraordinarily large datasets continues to grow. This is seen no better than cations in recent years, i.e., m etabolomics. As a result, multivariate data methodologies are becoming more and more commonplace in the typical workflow of an analytical chemist. Having the capacity to reduce and filter potentially insignificant data in a rapid and controlled fashion provides a format for scientists to analyze data in a practical manner. 51 , 52 The time consuming nature of fully interpreting numerous high resolution LC/MS datasets, which can often nu mber in the hundreds of sample injection s, would prove impossible to manually analyze without statistical support . The most common multiv ariate data analysis techniques, and two of the several utilized in this research, are partial least squares discriminant analysis (PLS DA) and principal component analysis (PCA). These techniques serve to reduce the data by determining the axis of greates t variance, also known as the first component, through the multidimensional space of a dataset. Orthogonal to this first component, a second axis of variance may be derived, followed by consecutive axes of variance (components) for as many variables as the re are in the data set. 52 , 53 These variables are a result of disparities in the features of the analytes that make up a particular dataset. For example, the signal intensity at a sp ecific m/z produced by a single compound makes up one feature, while the elution time of that compound (if one were


47 performing LC/MS) would define a second feature. All of these features are taken into account by the multidimensional analysis of the method (PCA, PLS DA, etc.), and subsequent components are created at the greatest variances between sample types. The greater the contribution of a specific variable to the variance within a dataset, the larger the relative weighting factor (loadings) given to t hat variable. 51 A sample can then be assigned a score, representing the summation of the product of a variable and its corresponding loadings coefficient. With the scores assigned, the variance can be visualized in two (or three) dimensions by plotting the scores of one component aga inst the scores of another component, called a scores p lot. If significant correlation exist s within the dataset, then those that contain high numbers of variables can be sufficiently reduced to a handful of components, where the first few components will account for a large percentage of the variation. 51 Principal Component A nalysis While it may seem like a straightforward task to merely input data into statistical software and run PCA, the underlying principles are not so simple. Data first needs to be scaled and normalized in order to represent an accurate depiction of the inherent similarities and differences. Most often, the data is autoscaled by the software to mean center each value around an origin value. This effectively reduces the dataset to two dimensions, above and below t he origin. The data is also commonly normalized to ensure accurate correlation of the data. There are many ways to implement normalization, quantifying the samples relative to one another is ideal, but data may be normalized by other means if this is not o btainable. Once the data is transformed, PCA can be performed. Transforming and analyzing data with only a couple variables is trivial, however, going beyond two or three variables (i.e., real world samples)


48 introduces complexities that are impossible to g rasp and scrutinize with the human mind alone. The advent of modern computers allows the processing of this multifaceted data and modeling, which provides the driving force for multivariate analysis. As stated earlier, PCA organizes the dataset by samples with the greatest variance, and in an unsupervised fashion. A simple test case, adapted from Menger, 53 may better illustrate this concept. Figure 1 23A displays a data set comprising hypothetical data regarding the length (cm) and weight (kg) of two sample groupings: data are scaled by calculating the average value for each data set (i.e., length and weight) and centering the data about the mean. This data is then plotted in 2D space by using Cartesian coordinates where one axis represents the variation amongst length (x axis) and the other, weight (y axis), as seen in Figure 1 23B. By plotting the data in two dimensions, it becomes clear that the axis of greatest variance is a line passing through the center of each grouping, as represented by the red line. This axis is the first principal component (PC1), while the axis orthogonal to this line (the purple line) is denoted the second principal component (PC2), which generally depicts variability amongst an individual sample grouping. The plot may then be shifted to pro ject the PCs onto the x and y axes generating the aforementioned scores plot, which is the most common display of PCA (Figure 1 23C). The more variables present in the data set, the greater the degree of dimensionality associated with the plot. For instan ce, you could add more species of dolphins for a larger number of groupings, and then add more features to distinguish them, such as number of teeth, tooth diameter, length of snout, size of dorsal fin, etc. Moreover, as you increase the number of variable s, the complexity surpasses 3D space (and human


49 conceptualization), thus, requiring computer modeling to project this continual orthogonality in comprehendible Euclidian space. Partial Least Squares Discriminant A nalysis PLS DA as a methodology is similar to PCA whereby it organizes datasets by their maximum variance. However, rather than analyzing the dataset as a collection of individual samples, it oversees the analysis from a group sampling perspective. In this way, PLS DA is classified as a supervised method, 52 as it takes into account the origin of a sample, and biases the variance based off thi s intrinsic knowledge. PLS DA helps to reduce and eliminate variation unrelated to the sampling grouping itself (e.g., sample preparation and biological variability), consequently, provi ng advantageous for building and developing classification models. Applying Statistical Analysis to Biomarker D etection Biomedical datasets are often comprised of enormous amount of data that would be irrational for a scientist to manually analyze without computer aided methods. PCA and PLS DA, in addition to other statistical analysis methods such as Random Forest (RdF) and Analysis of Variance (ANOVA), help to reduce the dimensionality of this information, and attempt to extract pertinent information base d off of obvious as well as subtle variations inherent to the dataset. By applying these methods to particular types of mass spectra collected on specialized mass spectrometers, compounds of interest may be identified. This work in volves a comprehensive m e tabolomics study of melanoma samples, a primary goal of which is to identify potential biomarkers. Employing multivariate analysis techniques in the search of these markers is made possible by feature extraction from the accurate mass data yielded by high resolution mass spectrometers,


50 in addition to fragmentation analysis provided by spectral patterns produced by tandem mass spectrometry. Statistical analysis is key to mining the data for information of importance, and running those exact masses, empirical formulas, and fragmentation patterns through open source databases to identify putative biomarkers for melanoma. Scope of the Dissertation This dissertation covers a broad scope of biomedical applications utilizing a combination of conventional and novel analytical techniques. The first chapter has provided the necessary basics to understanding all of the methodologies and instrumentation employed in this work. Chapter 2 details a study into the effects of modifying the solvent system used to create soluti ons of capsaicin for tussigenic challenges. UV s pectrometry was used to quantify capsaicin concentrations for investigations into solubility and stability of the solutions. Switching gears, Chapters 3 and 4 explore the role of mass spectrometry in clinical analysis. Chapter 3 evaluates the use of MS, FAIMS/MS, and standalone FAIMS for online analysis of human breath. Chapter 4 involves a metabolomics study of melanoma and normal skin samples by a variety of ambient ionization MS techniques. Finally, Chapter 5 provides a brief summary of the aforementioned research, as well as a future outlook into directions the research could pursue.


51 Figure 1 1. Transmittance interferences caused by an a lyte container . Incident radiant power ( P 0 ) loses intensity as sample absorbs in UV/Vis range producing transmitted radiant power ( P ). Losses also occur as a result of reflection at surface boundaries (e.g., air glass and solution glass), scattering caused by interaction with other solution molecules, and absorption by the walls of the container.


52 Figure 1 2. Simplified d iagram of the double beam geometry with single photodetector found in the HP 8450A spectrophotometer.


53 Figure 1 3. APCI process in positive polarity mode. Solvent containing the analyte is nebul ized and vaporized by the APCI probe and directed toward the corona discharge region around the needle, where reagent ions react with the uncharged solvent and analyte mole cules to convert them to ions Discharge region adapted from LTQ Hardware Manual . 12


54 Figure 1 4. Plot displaying applicable range of several atmospheric and vacuum pressure ionization methods.


55 Figure 1 5 . APCI process for breath sampling in positive polarity mode. APCI of breath may be performed with (A) or without (B) addition of vaporized solvent. In (A), the solvent and breath are at similar abundance, so either could feasibly be ionized by the reagen t ions. In (B), the signal to noise increases as only the aerosol and gas phase molecules in breath are ionized without solvent interference.


56 Figure 1 6. ESI process where preformed analyte ions in a solvent are sprayed through a charged needle to for m aerosol droplets (A) be ions (B).


57 Figure 1 7. EESI process where pure solvent is ionized and nebulized by a charged needle and directed toward an auxiliary sample spray. The charged solvent droplets extract the analyte from the sample spray droplets, then angles between the sample spray/inlet ( ) and two sprays ( ), as well as the distances between the intersection of the sprays and the inlet ( a ) and the two sprays ( b ) Adapted from Law et al. 22


58 Figure 1 8. Four types of liquid liquid droplet interactions. From top to bottom: bounce, disruption, fragmentati on, and tota l coalescence Adapted from Orme . 24


59 Table 1 1. Advantag es and disadvantages between LTQ, QE, and TOF instrumen ts . Metric LTQ QE TOF Mass analyzer pressure 10 5 10 11 10 7 Mass resolution Unit 150,000 12,000 m/z range Limited , low mass cutoff Dynamic Dynamic , no upper mass limit Tandem MS MS n capabilities MS 2 capabilities Single stage of MS (w/out auxiliary an alyzer) Size Small Small Large Throughput Continuous operation suitable for high throughput sampling Continuous operation suitable for high throughput sampling Pulsed operation suitable for coupling to MALDI, IMS, etc.


6 0 Figure 1 9. Schematic of Thermo Finn igan LTQ XL Adapted from LTQ Hardware Manual . 12


61 Figure 1 10. Positive portion of a stability diagram for ions in a transmission quadrupole mass filter.


62 Figure 1 11. Schematic of Thermo Scientific Q Exactive Adapted from Exactive Operating Manual . 26


63 Figure 1 12. Schematic of Agilent 6220 TOF Adapted from Agilent Concepts Guide . 25


64 Figure 1 13. Drift tube ion mobility spectrometry. Packets of ions are pulsed down the drift region where they collide and interact with gas molecules, decreasing their initial velocity as a function of their cross sectional a rea.


65 Figure 1 14. Field asymmetric ion mobility spectrometry (FAIMS) . An asymmetric waveform (top) is applied to the upper electrode while the lower is grounded. Under high field (red) the ion moves faster toward the lower plate than it moves to ward the upper plate during low field (green), creating a vertical displacement toward the lower plate. With each period of the waveform, the ion moves closer and closer to the lower plate until it makes contact and it annihilated.


66 Figure 1 15. The asy mmetric FAIMS waveform causes ions to drifts toward one of the electrodes, destroying the ion when it makes contact (red and green). To counteract this displacement, an auxiliary voltage (CV) is applied to the cell to allow specific ions to reach the detec tor (blue) .


67 Figure 1 16. FAIMS ion classification. Ions can be classified as type A, B, or C depending on their change in mobility as a funct ion of electric field strength Adapted from Buryakov et al . 36


68 Figure 1 17. 3D representations of planar, cylindrical, and hemi sp herical FAIMS cell geometries. T he blue arrow demonstrates flow of ions. Planar utilizes two parallel plates through which ions pass, no intrinsic focusing. Cylindrical houses a rod for ions to pass over or under between two grounded electrodes, providing vertical focusing . Hemispherical involves using a partial spheroid electrode in the center of a chamber (seen transparent above) which ions drift around, affording radial focusing ( a greater dimensionality than seen with cylindrical) .


69 Figure 1 18. Relationship of frequency and amplitude of FAIMS waveform. The top waveform is 3 times the amplitude but 1/3 the frequency of the bottom waveform (left). Ions acted on by both wavef orms will undergo the same vertical displacement in the FAIMS cell during the same period of time (right).


70 Table 1 2. Comparison of FAIMS cell features. Home built planar FAIMS Owlstone Lonestar µFAIMS chip Number of channels 1 47 Ionization Amb ient (APCI/ESI) 63 Ni decay Maximum applied dispersion voltage 5,000 V 250 V Waveform frequency 750 kHz 28,600 kHz Field strength 25 kV/cm 72 kV/cm Number of oscillations (@ 1.5 L/min) Thousands Hundreds


71 Figure 1 19. A Thermo waveform generator used for the full si ze FAIMS cell (A) and an Owlstone Lonestar portable gas analyzer standalone FAIMS device (B ) p hoto s courtesy of author .


72 Figure 1 20. Side by side comparisons of a Thermo waveform generator and an Owlstone Lonestar device photo s cou rtesy of author .


73 Figure 1 21. Various FAIMS waveforms: sinusoidal (A), bisinusoidal (B), true Lonestar waveform (C), 43 and square (D).


74 Figure 1 22. Conceptual chromatographic peaks produced by LC, illustrating the void volume elution at t M , the analyte peak elution at t R , and the relationship of the two ( t S ) Adapted from Skoog et al . 1


75 Figure 1 23. Hypothetical example of PC A applied to 2 variable dataset. Table of data corresponding to two dolphin centered scatter plot of data in relation to variable axes (B), and this plot transformed into a scores plot (C) Adapted from Menger . 53


76 CHAPTER 2 STANDARDIZED METHOD FOR SOLUBILITY AND STORAGE OF CAPSAICIN BASED SOLUTIONS FOR COUGH INDUCTION Introduction The volatile compound capsaicin (CAP , Figure 2 1 ) can be utilized to investigate human respiratory reflexes, often to assess successful treatment of many chronic and ac ute respiratory ailments. 54 56 However, the greatest challenges in using CAP for such tests are its low solubility in water and storage instability. Therefore, while a pure water solvent system would be ideal for such solutions, as they involve human consumption or inhalation, water alone cannot be used since the capsaicin does not dissolve in sufficient quantities to afford meaningful testing. Hence, use of orga nic solvent systems to dissolve the capsaicin into solution is required. This presents a safety hazard as a large number of organic solvents are toxic to humans, while many more are fatal. Fortunately, CAP has been shown to be completely soluble in various organic solvents considered safe at moderate levels, with ethanol (EtOH) providing the greatest solubility. 57 , 58 Further research has revealed that dissolving CAP in a mixture of EtOH and polysorbate 80 ( Tween 80, Figure 2 2 ) allows for a higher solubility than EtOH without Tween. 59 62 For this reason, use of Tween solvent systems has become ubiquitous among many of the reported studies on the subject. Previous experiments (results not shown) suggests Tween 20 is an equivalent replacement solvent for Tween 80; thus, Twee n 20 was utilized in this research. *Portions of this chapter were reprinted with permission from Costanzo, M. T.; Yost, R. A.; and Davenport, P. W. Cough . 2014


77 CAP solutions containing Tween 20 have been prepared and tested on subjects with well documented dose response relationships for cough reflex testing, also called a tussigenic challenge. 63 66 Regrettably, there was a common complaint of a bad taste detection and awareness as the adverse stimulus . 67 69 This taste, often described as soapy and unpleasant, can be attributed to the presence of the surfactant Tween 20, or Tween 80, in solution. Since subject s are required to inhale CAP solutions for an extended period of time, the taste of the solutions becomes a primary cause for discomfort. Although previous studies reported quantitative comparisons of CAP in solutions with Tween versus solutions without Tw een, 59 , 60 there has yet to be a study that directly compares the solubility of a broad range of concentrations of CAP in 10% EtOH alone to CAP in a Tween solution. Furthermore, a comparison of CAP solubility in solutions of varying percentages of EtOH has yet to be reported. The purpose of such a comparison would be to produce CAP solutions in an EtOH based solvent system, equivalent to those deemed fit for use in the aforem entioned cough thres hold tests, 54 56 but without Tween. Determining the optimal method for capsaicin detection is difficult as an English speaker as all reviews on the subject are in Chinese, but analysis of the English literature, and titles of foreign language articles, indicates ultraviolet/visible absorption spectrophotometry (UV/Vis) and high performance liquid chromatography (HPLC) are the analytical tools of choice for quanti tation of capsaicin, whether initially in solution or solid form. HPLC and UV/Vis have been utilized as early as 1979 to perform quantitative studies o f capsaicin containing samples. More recent publications related to


78 this work also utilize HPLC to separa te the capsaicin from the extraneous components of their solutions directly prior to quantitative measurement, by mass spec trometry and spectrophotometry. However, since the CAP solutions involved in this work were comprised of a limited number of componen ts, UV/Vis detection was performed without prior HPLC separation to decrease analysis time and complexity. In this study, we sought to determine (1) the percentage of EtOH yielding CAP solubility comparable to a desired concentration of CAP in Tween 20 sol ution, (2) the concentration of CAP dissolved in 10% EtOH required to produce analogous solubility to a desired concentration of CAP in solution containing Tween 20, (3) if the ratio of CAP present in EtOH solution to CAP present in Tween 20 solution chang es with variations in CAP concentration, and (4) the shelf life of CAP in 10% EtOH. Based upon these results, we developed and outlined a standardized method for optimal preparation of CAP solutions for use in tussigenic challenges . Experimental Chemicals and R eagents Capsaicin (CAP), pharmaceutical grade, was purchased from Formosa labs (Taoyean, Taiwan) and stored at 20ºC until use. 190 proof ethyl alcohol (95% EtOH) was purchased from Decon labs (King of Prussia, PA). HPLC grade water was pur chased fro m Fish er Scientific (Fair Lawn, NJ). Tween 20 was purchased from MP Biomedicals (Solon, OH). All solutions were stored at room temperature, unless otherwise stated. Solubility of CAP in Different Solvent S ystems Prior to each study, a stock solution of 25 mM CAP was prepared. The stock solution was prepared by dissolving 76 mg (250 µmol) of solid CAP in 10 mL of 95%


79 EtOH. Preparing such a high concentration of CAP for the stock solution allowed any error associated with variation of volume to be considered negligible. In the research presented, solutions with EtOH levels at 5, 10, 15, 20, 25, 35, 50, 65, 75, and 95% EtOH in H 2 O were prepared. Additionally, solutions of CAP were prepared at concentrations of 0, 200, 350, and 500 µM for all percentages of EtOH mentioned. Finally, a much broader range of concentrations (0, 50, 100, 150, 200, 250, 300, 350, 400, 450, and 500 µM) were prepared in 10% EtOH for more in depth analysis. All samples were diluted to a total volume of 5 mL in triplicate. To prepare the Tween solutions, the CAP powder was initially dissolved directly in a solution made up of 80:10:10 H 2 O:Tween 20:95% EtOH (v/v/v, hereafter referred to concentrations of CAP. Instead, the 25 mM stock solution in 95% EtOH provided the CAP, while a solution of 80:10:10 of H 2 O:Tween 20:95% EtOH (v/v/v) was prepared separately. To this Tween solution, a specified volume of 25 mM stock CAP was added to achieve the desired concentra tion of CAP. Solutions of CAP were prepared at concentrations of 0, 50, 150, 250, 350, and 450 µM, with a total volume of 5 mL. Similar to the CAP solutions in EtOH, the Tween solutions were prepared in triplicate. Stability of CAP in 10% EtOH S olutions To study the stability of CAP in the optimized solvent system, solutions at concentrations of 0, 200, 350, and 500 µM CAP in 10% EtOH/H 2 O (without Tween 20) were prepared by diluting 95% EtOH down to 10% and then spiking in the desired amount of 25 mM stock CAP in 95% EtOH. These solutions were stored in one of four different environments: 1) room temperature and exposed to light, 2) room temperature and protected from light, 3) approximately 3 °C and protected from light, and 4)


80 approximately 20 °C and pro tected from light. All solutions were kept in glass vials, and placed in a cardboard box to shelter them from light, except one set of vials (3, above) which were directly exposed to room light for the purposes of the study. For solutions stored below room temperature (3 °C and 20 °C), the solutions were allowed to warm up to room temperature for two hours and then vortexed, before analysis. Visual inspections of the solutions prior to analysis indicated no turbidity. All samples were prepared in triplicat e to a final volume of 5 mL. Determination of CAP C oncentration The concentration of CAP in the solutions was measured by ultraviolet (UV) absorption spectrophotometry. A Hewlett Packard 8450A UV/Vis spectrophotometer was utilized to monitor the absorbance from 200 to 300 nm, as well as focus exclusively on 281 nm. 58 Although the instrument offers automatic blank subtraction due to its double beam geometry, blanks of the EtOH so lvent were analyzed separately from the samples, and manual blank subtraction was performed as the EtOH signal was negligible. However, as Tween 20 exhibited appreciable absorbance in the spectral region of interest (Figure 2 3 B ), the double beam capabilit ies of the instrument were exploited to perform blank subtraction automatically. To ensure removal of anomalous P = 0.05 was performed on any data suspected to be a statistical outlier. All outliers were s ubsequently removed from the data set . Results and Discussion S tudies reported by Kopec et al. 59 , 60 provided the groundwork for this study of CAP solutions involving the use of Tween as a solvent. Yet, this report invo lves a more comprehensive comparison of Tween and EtOH based solutions, in terms of both


81 solubility and stability. The primary purpose of our study was to determine if preparing capsaicin solutions in similar fashion to Kopec benefitted from Tween addition enough that its use would be recommended. As the reported preparation of the Tween solutions did not involve any separation step, we wanted to analyze the solutions as they would be prepared clinically. In this respect, any aspect of dissolution that may result in suspension/dispersion of CAP would accurately reflect clinical implications when conducting the test. Looking at the results, it happens that no turbidity was present for any solutions upon analysis, as discussed in more detail below. Preliminary Analysis of CAP As mentioned, HPLC was excluded prior to UV detection. Absence of a chromatographic step decreases both analysis time and analysis complexity. For example, the HPLC analysis described by Kopec et al. required 12 minutes per sample, 60 whereas an average sample analy sis time with the method described in this work was significantly faster (approximately three minutes). However, while skipping separation of the samples allowed for quicker analyses, there were initial concerns over the solubility of the capsaicin. Partic les large enough to scatter light were observed in solutions stored below room temperature during the time course study; however, these particles all redissolved over the course of the two hour warming period and vortexing of samples before analysis. Visua l inspections of the solutions prior to analysis indicated no turbidity. Furthermore, by removing HPLC separation, analysis is streamlined, but all components will be present and detected at once, potentially increasing the complexity of the sample. Howev er, if the sample is simple to begin with, there may not be a need for separation. To e xplain in more detail, each CAP solution was comprised of only a


82 few components (i.e., CAP, H 2 O, EtOH, and Tween 20), and each component has a different maximum absorban ce in the UV spectrum (Figures 2 3A and 2 3 B) . Consequently, a singular component may be monitored for its absorbance at a specified wavelength that is limited to only that component. Instances where more than one component absorbs at a given wavelength we re limited to one observed wavelength producing absorbance of two components. The use of a double beam spectrophotometer allowed automatic blank subtraction of the second component, affording the ability to monitor a single component of choice (Figure 2 3 C ). It should be noted that while exclusion of a separation method prior to UV/Vis detection provides the described benefits , it also lowers the accuracy and precision of the measurements. However, this method is still efficient enough to produce meaningful data and reproducibility (Figure 2 4 ). This not only takes into account systematic error any human error in measurement of the analytes and inconsistencies produced by infrequent calibration of the micropipettes used. For all of these reasons , UV/Vis detection was performed without prior HPLC separation, unlike previous studies . The use of a spectroscopic technique permits non destructive and rapid analysis of the sam ples, but affords less selectivity than would be achieved with a supplementary HPLC separation step. 1 Future studies utilizing HPLC separation or other analytical techniques could provide further confirmation of these r esults, but were not performed here. Selection of Wavelength for UV Spectroscopic A nalysis As the Merck index states that the molecule capsaicin has two peaks of maximum absorbance in the UV range, 227 and 281, 58 a range of 200 300 nm was


83 observed for all of the blanks and solutions. There were indeed two distinct maxima; however, the peaks produced at 281 nm were more defined and better resolved than those at 227 nm (Figure 2 4 and 2 5 ). Thus , 281 was chosen as the wavelength to monitor for the remainder of the study. Solubility of CAP in Different Solvent S ystems To begin, the solubility of CAP in varying concentrations of EtOH in H 2 O was evaluated to determine how decreasing the amount of solvent (while simultaneously increasing the concentration of water) would affect the dissolution. Figure 2 6 displays the blank corrected absorbance at 281 nm of CAP as a function of EtOH percentage present in the solvent. As expected, the solubility of CAP increases with increasing EtOH percentage; ramping the concentration of ethanol from 5% to 75% results in a 10 to 15% increase in solubility. The effect of a mixed solvent system containing both Tween 20 and EtOH was also investigated. Fi gure 2 7 details the CAP absorbance response curve for three different solvent systems, 10% EtOH, 95% EtOH, and Tween solution . CAP dissolved in all three solvents exhibited similar absorbance, suggesting that the solubility of CAP is similar in each solve nt system tested . Despite the similarity, CAP appeared to have the greatest solubility in 95% EtOH, as evidenced by the greater slope of the response curve. As CAP is completely soluble in pure EtOH, 58 the absorbance measurements obtained from CAP in the 95% EtOH solvent system were used to calculate the correct (or predicted) CAP concentrations. The average ratio of the concentration experimentally determined for CAP in 10% EtO H to the concentration of CAP in 95% EtOH ( E/P ) is around 91% (Table 2 1). The average ratio of the concentrations of CAP in 10% EtOH to Tween solution ( E/T ) is nearly 99%.


84 From the results of this study, it can be concluded that CAP solutions that are sim ilar in solubility stability (detailed further in the next section) to those with Tween 20 can be prepared using purely EtOH based solvents, yet lack the foul taste asso ciated with Tween. Kopec et al. 59 , 60 determined th at utilization of a solvent system of Tween 80, ethanol, and water for preparation of CAP solutions yields better solubility and stability than a solvent system comprised of ethanol and water without Tween. However, the concentration of ethanol in their re ported solutions was only around 1%. The present work suggests that Tween 20 yields a less than 2% increase in solubility over a solvent system of ethanol and water, where the ethanol concentration is 10% rather than 1% by volume (Table 2 1). Furthermore, regardless of the inclusion or exclusion of Tween 20, solvent systems safe for human consumption (i.e., less than 20% ethanol by volume) generally demonstrate CAP concentrations below the predicted concentration; therefore, neither allows complete solubili ty. As noted in Figure 2 6 , there is a modest increase in solubility as the composition of EtOH is increased from 5 to 75%; however, this difference can be easily corrected for by dissolving additional CAP into the solution, as described in more detail lat er. Thus, lower percentages of EtOH are recommended to allow safe human inhalation, approximately 20% or less EtOH. Therefore, we have chosen to use 10% EtOH solutions to decrease the amount of EtOH solvent in the inhaled vapor. Furthermore, the concentrat ion of CAP used for the tussigenic challenge does not typically exceed 500 µM; 65 therefore, an increase in variation betw een high and low EtOH compositions at higher concentrations of CAP does not become a factor.


85 For solutions in 10% EtOH, there is a 10% decrease in solubility relative to the predicted solubility (Table 2 1); again, predicted concentration is determined fro m absorbance measured for CAP dissolved in the 95% EtOH solvent system. This can be overcome by increasing, by 10%, the amount of CAP initially dissolved in solution. Accordingly, a suggested standardized sample preparation for CAP solutions in an EtOH ba sed solvent system is detailed below. Recommendations for Preparation of CAP Solutions in EtOH Solvent S ystem In this study, we used 95% EtOH and accounted for the impurity in our dilutions; however, the following guideline uses 100% EtOH to allow for easi er calculation. Note: The ethanol used must be safe for human consumption and the capsaicin pharmaceutical grade. Step 1. Dissolve solid CAP in 100% anhydrous EtOH to create a concentrated CAP stock solution. Dissolving 0.7635 g of solid CAP (2.5 mmol) i n 100 mL of 100% EtOH will produce a 25 mM stock solution of CAP. Step 2. Prepare the solvent for the inhaled CAP solutions by diluting 100% EtOH with H2O. To create 10% EtOH solutions, combine 100 mL of 100% EtOH with 900 mL of H2O. Step 3. Add the CAP stock to the 10% EtOH to produce the desired concentration of CAP in 10% EtOH, using the following equation: ( 2 1) Where Cstock and Vstock are the concentration and volume of the CAP stoc k solution, and Ccap and Vcap are the concentration and volume, respectively, of the desired CAP solution in 10% EtOH. Recall that the initial amount of CAP added needs to be increased by 10% to account for the decrease in solubility associated with a 10% EtOH


86 solvent. Equation 2 1 may be utilized to prepare any desired CAP solution, or Table 2 2 may provide a quick reference for specific concentratio ns. Stability of CAP in 10% EtOH S olutions The stability of solutions of CAP prepared with Tween 80 have bee n previously reported, hence, we did not repeat a similar investigation. Instead, we focused solely on the stability of CAP solutions prepared in 10% EtOH, which were diluted in HPLC grade water rather than saline solution. To determine the stability of CA P in 10% EtOH under different environmental conditions, a time course study was conducted by monitoring the absorbance of CAP solutions. The data for each of the three concentrations monitored, 200, 350, and 500 µM CAP, are shown in Figures 2 8, 2 9, and 2 10 , respectively. Of all the environments, CAP stored around 3 °C and sheltered from light yielded the most consistent absorbance measurements for all three concentrations tested. Furthermore, all three concentrations tested yielded absorbance measurement s within error relative to the initial time point for the entirety of the seven month study. CAP solutions stored at room temperature (~23 °C), while protected from light, demonstrated a consistency in absorbance similar to when stored at 3 °C. Within erro r, there was no decrease in absorbance, and therefore no decrease in CAP solubility from 0 to 27 weeks, for all three of the concentrations tested. The one exception to this trend was the 12 week measurement for the 500 µM concentration. This anomalous dat a point demonstrated a decrease in intensity that is not within error of the observed trend. data point being considered an outlier ( P = 0.05).


87 CAP solutions stor ed at room temperature, and exposed to light, exhibit no decrease in absorbance throughout the entire seven months for concentrations of 200 and 350 µM. Conversely, for the highest concentration of 500 µM, there is a significant decrease in absorbance of C AP starting after only 2 months of storage. A single measurement for the 500 µM solution taken at the sixth week of storage (Figure 2 10 ) shows a decrease in absorbance, before returning to an absorbance similar to the week prior, and then gradually decrea sing to lower absorbance as the further weeks considered an outlier ( P = 0.05). The data shown for CAP stored at 20 °C, while protected from light, proved to be the lea st consistent, in terms of both stability and error. The 200 µM solution demonstrates little to no decrease in absorbance for the first four months of storage before beginning to slowly diminish in intensity (Figure 2 8 ). The 350 and 500 µM solutions revea led a much sooner drop off in intensity, as the absorbance began to significantly decrease after only a month (Figures 4 and 5, respectively ). For the concentrations of 200 and 350 µM, the drop in absorbance intensity correlates with a dramatic increase in standard deviation of the data, rendering the data questionable at best. The measurements taken for 500 µM contain much less associated error, and can be considered more referential The most pertinent findings from this study show that the greatest stabil ity may be achieved by storing the specified solutions of CAP in a light free environment, and at a temperature of approximately 3 °C. Under such conditions, the solutions of CAP in 10% EtOH remained stable, over a range of concentrations, for a minimum of 30 weeks


88 (nearly 8 months, the duration of the storag e study), as seen in Figures 2 8 2 10 . Additionally, the highly concentrated stock solutions of CAP prepared in pure EtOH remained stable for at least a year (the maximum time tested) under the same c onditions. This increase in stability of non Tween CAP solution compared to t hat shown in the previous study 59 , 70 may be attributed to either the use of distilled water as the dilution s olvent rather than saline solution, or the increase in EtOH composition. Use of saline may destabilize the CAP solutions causing them to degrade quicker. Alternatively, the presence of a higher percentage of EtOH in solution may provide greater stabilizati on of CAP. The drastic drop in solubility of solutions stored at 20 °C might be due to the low temperature causing some CAP to precipitate out of solution and adsorb to the walls of the container, and then not redissolve once the solution is brought back up to room temperature and vo rtexed . Conclusions In this study, it was demonstrated that CAP solutions prepared in a 10% EtOH solvent system are nearly as soluble as CAP solutions prepared in a Tween incorporated solvent system. Although neither solvent sy stem allows complete solubility of CAP, the concentrations that can be achieved are quite adequate for tussigenic challenges. Therefore, to avoid the foul taste of Tween and still efficiently prepare soluble CAP solutions in 10% EtOH, a simple equation may be followed (Equation 2 1). This research provides the groundwork for creating a standardized approach to preparing CAP solutions for use in tussigenic challenges. Until further refined methods are established, the preceding preparation steps should be t aken to achieve maximum solubility of CAP in solution. Similarly, for optimal storage, keep solutions shielded from light and at a temperature of around 3 °C.


89 Prior to publication of this research, clinical studies were performed utilizing the described me thods (unpublished results). 50 500 µM capsaicin solutions in 90% physiological saline and 10% ethanol were stored for 1 4 months, either refrigerated or at room temperature. No differences as a function of vehicle solution (90% physiological saline, 10% e thanol vs 80% physiological saline, 10% Tween 20, and 10% ethanol), storage time, or storage temperature were observed for cough threshold, cough number, Urge to Cough Threshold, and Urge to Cough sensit ivity compared to prior studies 66 , 71 , 72 using equal concentrations of capsaicin in 80% physiological saline, 10% Tween 20, and 10% EtOH solutions. There appears to be fewer vehicle elicited coughs with 90% physiological saline and 10% ethanol solutions compared to 80% physiological saline, 10% Tween 20, and 10% EtOH solutions. Future studies are required to systemat ically compare the cough response to capsaicin using these vehicle (90% physiological saline, 10% ethanol vs 80% physiological saline, 10% Tw een 20, and 10% EtOH) solutions.


90 Figure 2 1. Molecular structure for CAP.


91 Figure 2 2 . Molecular structur es of Polysorbate 80 and 20. Polysorbate (Tween) compounds contain a total of 20 oxyethylene groups ( CH 2 CH 2 O fatty acid associate with the polyoxyethylene sorbitan part of t he molecule (i.e., 20 monolaurate and 80 monooleate).


92 Figure 2 3 . Abs orbance of 95% EtOH [A ], Tween solution (10% EtOH/10% Tween/80% H 2 O) [B ], and CAP (50 500 µM, 50 µM increments) dissolved in the Tween solution [C ] from 200 300 nm. The EtOH and Tween solution spectra are produced without prior blank subtraction, CAP produces a maximum absorbance peak at =281 nm after automatic blank subtraction. Tween also absorbs at 281 making visualization of CAP difficult without its preceding subtractio n.


93 Figure 2 4 . Absorbance of CAP (50 500 µM, 50 µM increments) in 95% EtOH from 200 300 nm ; taken over the course of several days to demonstrate the reproducibility of the method and instrument. CAP produced maxima at 227 and 281 nm. The peaks at 281 nm are better defined and resolved than 227 nm .


94 Table 2 1. CAP concentration in EtOH vs. Tween based solvent systems . Experimental ( E ) vs Predicted ( P ) Concentrations (µM) Absorbance at 281 nm ± 1 std. dev. (3 replicates ) 95% EtOH 10% EtOH Tween Solution Ratio ( E/P ) 50.0 ± 2.3 45.1 ± 3.3 43.1 ± 2.7 0.90 ± 0.08 150.0 ± 0.1 135.8 ± 4.6 137.7 ± 0.8 0.91 ± 0.03 250.0 ± 0.3 226.5 ± 8.8 232.3 ± 2.9 0.91 ± 0.04 350.0 ± 0.1 317.2 ± 3.5 326.9 ± 4.4 0.91 ± 0.01 450.0 ± 0.1 40 7.9 ± 20.1 421.5 ± 8.9 0.91 ± 0.05 0.91 ± 0.02


95 Figure 2 5 . Absorbance of 200 µM CAP solutions in varying concentration s of EtOH in H 2 O (5 20%, 5% increments). CAP produced maxima at 227 and 281 nm. The peaks at 281 nm are better defined and reso lved than 227 nm .


96 Figure 2 6 . The blank rying compositions of ethanol. Error bars correspond to ± 1 standard deviation .


97 Figure 2 7 . Calibration curves for CAP in three different so lvents: 10% EtOH, 80:10:10 H 2 O:EtOH:Twee n 20 (Tween), and 95% ethanol. Error bars correspond to ± 1 standard deviation .


98 Table 2 2. Recommended preparation of 1L CAP solutions at a range of useful (tussigenic) concentrations . Desired concentration of caps aicin in solution (µM) Volume of 25 mM capsaicin stock (mL) Volume of 10% ethanol (mL) 5 0.22 999.78 10 0.44 999.56 20 0.88 999.12 40 1.76 998.24 100 4.40 995.60 150 6.60 993.40 250 11.00 989.00 375 16.50 983.50 500 22.00 978.00


99 Figure 2 8 . as a function of time, for four different storage environments: 20°C (dark), 3°C (da rk), 23°C (dark), 23°C (light). Error bars correspond to ± 1 standar d deviation .

PAGE 100

100 Figure 2 9 . as a function of time, for four different storage environments: 20°C (dark), 3°C (dar k), 23°C (dark), 23°C (light). Error bar s correspond to ± 1 standard deviation.

PAGE 101

101 Figure 2 10 . as a function of time, for four different storage environments: 20°C (dark), 3°C (dark), 23°C (da rk), 23°C (li ght). Error bars correspond to ± 1 standard deviation .

PAGE 102

102 CHAPTER 3 DEVELOPMENT OF NOVEL METHODOLOGIES FOR ANALYSIS OF HUMAN BREATH BY HIGH FIELD ASYMMETRIC WAVEFORM ION MOBILITY SPECTROMETRY (FAIMS) AND MASS SPECTROMETRY (MS) Introduction Analys is of human breath is currently used in law enforcement to assess blood alcohol content, but what if the same quickness and reliability could be used to identify someone who has recently used cocaine, marijuana, or other illicit substances? Furthermore, wh at if a similar test could be used to detect respiratory disease or cancer? Although breath analysis has the potential to conduct these types of analyses, current techniques do not provide such capabilities in a rapid and cost effective clinical setting. 73 76 Today, most diagnostic tests for drug monitoring and disease detection require drawing blood and conducting several tests on the blood sample. Most tests take a m inimum of several hours, and blood requires special storage conditions. 73 , 77 All of this equates to high cost from both a financial and time of analysis standpoint. Breath analysis is noninvasive, providing a safer and more rapid alternative to the typical barrage of tests conducted for many diseases. Breath also has a less complex matrix than urine and blood. This implies that, alt hough the detection technique may not be as sensitive as those for urine or blood, the less complex background noise could increase the signal to noise and overcome any inadequate sensitivity of the detection method. 73 Analysis of br eath has even been shown to provide sufficient separation of blood borne biomarkers, allowing possible identification of drugs, alcohol, and cancer. 20 , 78 , 79 Breath analysis could also offer real time assessment for disease and drug abuse, proving an and law enforcement vehicle.

PAGE 103

103 Finally, breath analysis could save millions of dollars in medical and labor costs, and, more importantly, may have the potential to save lives. Although there are many volatile compounds observed naturally in exhaled breath, 80 few of proven clinical importance are widely accepted by the medical field. Those that are deemed relevant, such as nitric oxide (NO), 77 , 81 have well established protocols for detection, thus, creating a new method of analysis that caters specifically to these would not be prudent. Still, a method that is capa ble of detecting established compounds, in addition to novel biomarkers and illicit substances would prove incredibly beneficial. Furthermore, many volatile organic compounds (VOCs) native to breath 78 , 82 are easily altered by endogenous and exogenous factors , 75 resulting in hesitance in relying solely on volatile markers for accurate detection of specified diseases. Therefore, while detection of those volatile compounds is vital, having the capability to provide rapid, real time detection of supplemental nonvolatile compo unds would be of greater significance. Contrary to what may be believed about the make up of exhaled breath, it is not limited to VOCs. In reality, many compounds of higher mass and less volatility can be found due to the blood breath interface in the lung s, located in the pulmonary alveoli (Figure 3 1). 78 The alveolar membrane is the gas exchange surface through which mole cules in the blood may diffuse. 83 If the molecule can diffuse from the blood in the capillary, through the mem brane, and into the alveolar chamber , it moves from a liquid to a gas environment. Then, if the kinetic motion of the molecule and pressure inside the alveoli are high enou gh, the molecule will be expelled from the lungs and present in the exhaled breath (Figure 3 2). 76 , 84

PAGE 104

104 Several types of respiratory ailments, including ast hma and chronic obstructive pulmonary disease (COPD), exhibit the potential for clinical diagnosis via a wide variety of volatile and nonvolatile biomarkers present in the breath. 77 , 81 , 85 88 Similarly, diabetes, lung cancer, and diseases of the liver could prove diagnostically amenable to detection in breath. 76 , 78 , 79 , 83 Nevertheless, the availability of these less volatile compounds is not as easily obtainable as it seems. Breath testing itself, dates back to antiquity, where physicians in ancient Greece knew the illnesses. 76 Modern breath analysis can be traced to the early 1970s, where Linus Pauling analyzed human breath using gas chromatography (GC), 80 determining over 200 components that compromise human breath. Over the past few decades, many advancements have been made in collection, separation, and identification of compounds in breath, but there has yet to be a single technique that can provide both sensitive and selective analysis over a wide mass range, and in real time (Table 3 1). 73 Furthermore, all of these requirements are desired in an instrument amenable to the clinical setting. The methods used for breath collection and introduction are vital for determining the type of molecules ab le to be detected and may account for potential loss, dilution, or contamination of analytes. A single exhaled breath is not homogenous. 75 The first one third of This comes from the upper airway where no gaseous exchange between the blood and breath can occur. That last two thirds (approximately 350 mL) that follows is called alveolar, or end expired, air. 73 This comes from the lungs where the aforementioned

PAGE 105

105 blood/breath exch ange may occur. A handf ul of breath testing techniques are contingent on upper airway sampling , 74 , 86 and indeed, many more useful analyses may be performed utilizing the upper airway component of breath; for instance, measurement of certain bacteria exclusive to the upper airway. However, when serves to dilute the biologically significant compounds found in the alveolar portion , therefore finding a method which allows separate collection would be beneficial . 73 , 75 Avoiding dilution and contamination of end expired air is one of the prominent advantages to on line breath sampling. Since the entire breath is analyzed in real time, the dead space air can either not be collected at all, or ignor ed during data analysis as it will not have a chance to mix with the alveolar air. This is not true with batch sampling methods, which involve collection of samples prior to analysis. The entire breath, and more often a collection of many breaths, are coll ected and stored in a single container. 73 , 89 , 90 The most common method for analyzing breath remains GC. 73 It is often coupled to a secondary method such as mass spectrometry (MS), to allow identification of compounds of interest. Due to the part per billion/trillion (ppb/ppt) concentrations in which trace analytes are excreted in breath, preconcentrati on is often necessary. The most widely used method is adsorptive trapping, such as solid phase microextraction (SPME). 73 . As these are batch sampling techniques, they afford no real time detection, and may involve dramatic dilution of analyte during collection steps with the use of exhaled breath condensate (EBC), 89 or loss of specific reagents over time (e.g., tedlar bags). 90 While certain recent techniques, such as proton transfer reaction mass

PAGE 106

106 spectrometry (PTR/MS) and selected ion flow tube mass spectrometry (SIFT/MS), allow real time analysis in the ppb range, 73 , 79 , 91 they lack the adequate separation necessary to distinguish closely related chemical species as it is impossible to account for any mass overlap (whether from isomers, ion fragmentation, or cluster formation) with only molecular weight information being determi ned. 92 In addition, PTR and SIFT are only capable of volatile detection, and many potential markers of biological significance tend to have low volatility. Very recently, Zenobi et al. demonstrated direct detection of breath samples in real time with the use of extractive electrospray ionization/tandem mass spectrometry (EESI/MS/MS). 20 , 21 With this setup, bo th volatile and nonvolatile compounds in breath can be detected, quantified, and calibrated. 20 introduction of a heated breath sample directly into an electrospray plume, wh ere the sample undergoes ionization. As the analyte ions are formed, they are directed into the no prior separation step; thus, isobaric compounds will be indistinguisha ble. Additionally, portability would be an issue as a mass spectrometer is a primary component. A potential solution to the analytical problems posed by all of these techniques is ion mobility spectrometry (IMS), and more specifically high field asymmetric waveform ion mobility spectrometry (FAIMS). IMS and FAIMS both separate ions by moving them through a collision gas by way of an electric field. Separation of ions is based on their mobility, which is determined by their mass, size, shape, and charge. 37 , 93 The principle difference between these two techniques being that IMS separates ions by their mobility

PAGE 107

107 in a low field environment, while FAIMS alternates between high and low electric field strengths, therefore separating ions based on the difference in their high and low field mobility. There have been several successful reports of conducting breath analysis by IMS and FAIMS , 94 97 but while the problems of adequate separation and real time analysis are solved, the sensitivity of the technique remains poor. The limiting factor for resolution with IMS is the drift tube length, therefore portability will always be a concern since the drift tube can only become so small before even satisfactory separation is compromised. In contrast, the separation lost by miniaturizing a F AIMS cell could be overcome by other means. Shortening the analytical gap between the FAIMS electrodes subsequently lowers the voltage required to separate ions, hence a smaller waveform generator could be used. Moreover, increasing the waveform frequency permits a greater number of ion oscillations per unit of length, which affords higher resolution. Therefore, the FAIMS cell is not required to be as long, and the overall device can be manufactured on a much smaller scale. Micromachined FAIMS chips, with a nalytical gaps less than a millimeter wide, have been proven to effectively separate ions. Moreover, a standalone FAIMS device complete with FAIMS chip, power supply, waveform generator, detector, and computer for data readout is already commercially avail able, and occupies as much space as a conventional sized computer monitor . 43 , 98 The performance of FAIMS separation could also be improved by cha nging the drift gas. While nitrogen is the common choice, helium and carbon dioxide could also serve as the drift gas. Recent advancements in FAIMS involving the addition of modified ent in the

PAGE 108

108 resolution and signal intensity over conventional FAIMS methodologies. 39 Therefore, the breath in real time, and in a point of care setting. Loss of ions in the physical distance from the source to the FAIMS cell could be a concern when applying FAIMS to detection of low abundance analytes. However, in combination with a novel device that applies a high frequency oscillation (HFO) to the airway dur ing breath collection, similar or better detection limits than those published by Zenobi et al. 20 , 21 , 99 may be achievable. HFO has proven to increase the amount of substances exhaled in breath. 84 By coupling FAIMS to MS, and with the use of HFO for sampling, sensitivity and selectivity on par with the best breath analysis techniques is possible, but in r eal time and with the prospect of peak identification and characterization. This work reports the development of novel methodologies for the on line collection and analysis of exhaled human breath and simulated breath by MS, FAIMS/MS, and standalone FAIMS. However, the primary aim of this research was to evaluate the potential for FAIMS as a viable breath analyzer, permitting efficient, quick, and reliable detection of exogenous and endogenous compounds in human exhaled breath. Initially, methods for breath introduction and detection were compared and optimized for this research. Following that, standards of volatile dopants and an illicit substance were analyzed in solution to create reference spectra, as well as to optimize ionization and mass analysis par ameters. Finally, the same standards were diluted and spiked into human breath or a simulated breath system, and analyzed directly by MS, FAIMS/MS, and standalone FAIMS.

PAGE 109

109 Experimental Chemicals and R eagents HPLC grade formic acid (FA), methanol (MeOH), and water were purchased from Fischer Scientific (Fair Lawn, NJ). Ethyl alcohol, 190 proof (EtOH) was purchased from Decon labs (King of Prussia, PA). Four food grade flavorant compounds were purchased from Sigma Aldrich (St. Louis, MO): methyl anthranilate ( MethAn), methyl salicylate (MethSal), (R) ( ) carvone (Car), and vanillin (Van). THC, ( ) trans 9 tetrahydrocannabinol, was purchased from Cerilliant (Round Rock, TX). The molecular structures of these five compounds can be seen in Figure 3 3. All reagents were stored at room temperature, unless otherwise stated. Instrumentation All standalone mass spectrometric experiments were performed on a Thermo Scientific LTQ XL linear ion trap mass spectrometer (LTQ). The LTQ was equipped with an IonMax ionization chamber capable of electrospray ionization (ESI) or corona discharge atmospheric pressure chemic al ionization (APCI). The housing was modified to allow introduction of breath (or modeled breath) and ionization by extractive electrospray ionization (EESI) and APCI. Table 3 2 outlines the MS ionization and optics parameters utilized on the LTQ for the various experiments, unless otherwise stated. Experiments conducting FAIMS/MS analysis were performed on a Thermo Finnigan LCQ Classic 3D ion trap mass spectrometer (LCQ) utilizing ESI or EESI. Table 3 3 outlines the MS and FAIMS/MS ionization and optics p arameters utilized on the LCQ, unless otherwise stated. Introduction of human breath on all instrumentation and devices involved the use of an Intoximeters mouthpiece (Figure 3 4), unless otherwise specified. For MS

PAGE 110

110 analysis, breath was blown into the mout hpiece, passing through the one way check valve and into the IonMax chamber. The valve allowed breath to pass unhindered into the ionization region, but did not permit air to flow in the opposite direction, e.g., due to back pressure from the instrument. T he mouthpiece was held in place by an acrylic face plate with a hole drilled into the center to allow the piece to fit snuggly inside. The face plate is a replica of the original glass face plate that is typically used on the front of an IonMax chamber, an d was machined by the departmental Machine Shop. For FAIMS analysis, breath was blow into the mouthpiece and directed toward the instrument inlet while the mouthpiece was held around 6 10 cm away. FAIMS separation was executed on two distinct FAIMS cells, param eters and specs of which are detailed in Table 3 4. The first was a home built planar FAIMS cell utilizing a Thermo Scientific waveform generator and coupled to the LCQ. This cell is 65 mm in length, 20 mm wide, and has an analytical gap of 2mm (Figur e 3 5A). Analytes are pneumatically and electronically pushed through the device and into the MS. The second was actually a standalone FAIMS device, the Owlstone Lonestar portable gas analyzer, complete with ionization source, FAIMS cell, detector, and out put interface. This device utilized a microfabricated FAIMS chip (Figure 3 5B) made up of a boron doped silicon interdigitated FAIMS filter. Unlike conventional FAIMS cells, which constitute a single FAIMS channel through which the ions pass, the chip is m ade up of a 115 mm serpentine conduit which effectively creates 47 individual planar channels to filter the ions. Each channel has a path length of 0.300 mm and analytical gap of 0.035 mm. Ions only pass through one channel before reaching the detector, wh ere a single channel will provide identical separation to the rest. The filter is gold plated on both

PAGE 111

111 faces constituting four individual filter electrodes, with two electrodes at each face, creating four individually controllable potentials (Figure 3 6). T hus, a difference in potential can be applied top to bottom, across the silicon filter, to generate a longitudinal field to drive the ions through the chip, while an asymmetric waveform is simultaneously applied between the digits to perform the ion separa tion. In addition to being electronically driven through the chip, ions may also be pulled through the device by an external pump attached to the exhaust port. The standalone device also includes a deflector plate, ionization source ( 63 Ni radiation), and d etector. Ions enter the ionization radiation from the 63 Ni source (a cylindrical foil of 63 Ni located between the deflector plate and filter), before the bias on the deflector plate directs the ions toward the FAIMS filter. The ions pass through the filter, and are subsequently detected by the charge collector, a gold hexagonal grid. 43 Flavorant A nalysis Standard solutions were prepared by diluting neat flavorants in EtOH to the desired concentration, which varied over the course of the different studies conducted. EtOH was the solvent of choice since the flavorants would eventually be spiked into human breath, therefore, the solvent needed be safe for human ingest ion. MS and FAIMS/MS analysis involved the study of diluted standards, while standalone FAIMS analysis entailed the study of both neat flavorants and diluted standards. MS and MS n spectra were acquired by ESI or APCI for direct infusion of standards, and E ESI or APCI for breath sampling. To analyze flavorants in breath, a 10 µL drop of standard solution was placed directly onto the tongue around 10 seconds prior to sampling.

PAGE 112

112 THC A nalysis THC was purchased at standard concentrations of 1 mg/mL in MeOH. To cr eate solutions with lower levels of THC, these standards were further diluted in MeOH to desired concentrations. MeOH was chosen for dilution since the THC standards came already dissolved in MeOH. Since MeOH provided adequate ionization, there was no need to extend the preparation steps by reconstituting the THC in a new solvent. THC was analyzed by MS and FAIMS/MS, but not standalone FAIMS as the device no longer produced meaningful data during the time this study was being performed. The FAIMS chip in th e Lonestar device had a practical life of around 1 year before it began to show signs of degradation. This analysis took place 4 years after acquisition of the device, therefore it was decided to not collect inadequate data. MS and MS 2 spectra were acquire sampling. In the state of Florida, it is currently illegal to ingest THC for any reason, whether medical, recreational, or for research purposes, 100 therefore, a breath simulation sy (Figure 3 7). Results and Discussion Development of Breath Sampling M ethodologies On line introduction and collection of breath In order to analyze human breath in real time, an apparatus allowing introduction of breath samples directly into the MS instrument and FAIMS device needed to be developed. There are a few published techniques for collecting breath by on line monitoring involving the use of face masks, 86 , 101 breathing tubes, 102 or nasal airway collectors. 103 , 104 As there is no standardized procedure for breath collection at this time,

PAGE 113

113 further study would be needed to determine which method is optimal. In this research, two methodologies were creat ed and compared. The first consisted of a 6 cm long piece of Teflon tubing fitted into the auxiliary port of the APCI manifold of a Thermo Finnigan TSQ 7000 mass spectrometer. While analytes of interest were indeed detected in breath by APCI/MS, the sensit ivity of the method and reliability of the instrumental software were poor. However, the primary issue with this set up was the difficulty involved with introducing the breath through the tube itself. The diameter of the tube, as well as the opening in the manifold for the breath sample to flow through, were both too narrow to provide adequate sample introduction to the source. Furthermore, the back pressure from the instrument into the source chamber was so much that air would come out of the tube in the o pposite direction the breath was being introduced. In addition, a large amount of condensation would accumulate inside the tube with each consecutive breath. This buildup of condensed breath droplets could cross contaminate subsequent samples, as well as c ause dilution of the analyte as they may adsorb into the droplets. Due to the increased availability and lack of software issues compared to the TSQ, collection was instead performed on the LTQ or LCQ for all future experiments. Moreover, the Intoximeters mouthpiece was chosen for the purposes of expanding the diameter through which to breath, as well as inhibiting any air flow out of the breath tube from the pressure build up in the ionization chamber. This allowed more efficient and reproducible breath si gnals, in addition to more native breath samples from the ability to p erform relaxed, or tidal breathing, as there was no backflow to oppose the breath introduction.

PAGE 114

114 Accumulation of condensation was still seen with the new mouthpiece, however it was not as abundant as with the Teflon tubing. To combat this condensation, every few runs the mouthpiece was removed from the face plate, and the inside wiped with a Kim wipe to remove any liquid. Additionally, condensation could be discouraged by heating the tube during collection. 105 Simulation of breath To create a system by which to introduce illicit substances (i .e., THC) for analysis in a manner similar to breath, a simulation apparatus was created. This system was designed to allow known concentrations of a standards to be spiked into solution, then volatilized, and the headspace directed toward the inlet of the instrument for subsequent analysis (Figure 3 8). Nitrogen (N 2 ) from a pressurized tank was bubbled through a solution of MeOH residing in a sealed HPLC pressure safe bottle, first flowing through a controller to adjust the flow rate into the bottle. As th e N 2 continued to flow into the sealed bottle, the solution begins to volatilize and form a headspace that rose to the top of the bottle. When the headspace built up to a substantial level, it blew out of the second valve in the cap of the bottle. Attached to this valve is a length of open ended Teflon tubing. The headspace flowed through this tubing and out of the opposite end. The tip of the tubing where the headspace streamed out was held around 5.75 cm from the inlet of the MS, or a few centimeters away from the FAIMS cell (depending on whether MS or FAIMS/MS was being performed).. A known amount of standard was spiked into the MeOH in the bottle. This analyte was subsequently volatilized into the headspace and directed toward the MS

PAGE 115

115 inlet as before. Thi s created a gaseous form of the analyte to be detected, allowing sampling similar to human breathing, but in a more reproducible manner. Flavorant and THC A nalysis First and foremost, the flavorants and THC were selected for analysis due to their volatile nature. Flavorants were chosen over other types of compounds since they were safe for human consumption, making development of breath sampling methodologies easier since the samples could be spiked directly into breath. Those specific compounds were chosen due to their similar mass and/or shape to test the capabilities of FAIMS. THC was chosen as a more real world example, while still providing a high volatility. MS d irect infusion of standards Initially, ESI and APCI/MS experiments were conducted on the four flavorant volatile organic compounds (VOCs) detailed above. These experiments were performed by direct infusion of standard solutions into the LTQ. APCI/MS data provided more prolific results, since the standalone FAIMS device uses 63 decay, and APCI is more similar than ESI to this ionization mechanism. Table 3 2 outlines MS parameters utilized for direct infusion of flavorants by ESI and APCI/MS. All data reported was taken in positive ion mode, unless otherwise stated. First, a blank (EtOH) and all four VOCs (0.25 M) were analyzed individually by ESI/MS. The most abundant peak for EtOH was observed at m/z 93, corresponding to the [2M+H] + peak (Figure 3 9E). The most abundant ion for all of the VOCs was observed at the correspondi ng [M+H] + peak (Figure 3 9A D). Other prominent peaks include source fragments and i on adducts for most of the VOCs: m/z 65 corresponds to a reagent ion from the ESI process (as it is decreases when switching to APCI), m/z

PAGE 116

116 211 is a contamination peak, [M C 3 H 5 ] + ( m/z 109) for Car, [M+H MeOH] + ( m/z 120) for MethAn , [M+K] + ( m/z 181) for Van, and the dimer, [2M+H] + , for both Car and Van. Next, APCI/MS analysis was conducted on similar standards. The heater was set low to create an environment resembling that of standalone FAIMS, allowing the data produced by both methods to better correlate. Data were again collected by direct infusion for an EtOH blank followed by the VOCs (250 ppm); however only three of the four flavorants were analyzed. MethSal was contami nated between studies, and therefore was not analyzed by APCI/MS. Similar to ESI/MS, the most abundant peak observed for EtOH was m/z 93, [2M+H] + , and the corresponding [M+H] + ion for all of the VOCs (Figure 3 10). Additionally, the dimer ([2M+H] + ) can be observed for Car. The secondary peaks associated with adducts and fragments are noticeably less intense in the APCI spectra compared to ESI. In fact, you cannot see many of these peaks in the spectra presented here, their presence was only seen upon compre hensive inspection of the mass spectra, but they are indeed present above the baseline in APCI and will be discussed in more detail below. However, it should be noted that the most likely reason for their lower abundance with APCI is the much lower concent ration analyzed. Standard solutions were created at a range of concentrations from 1 250 ppm, then an optimal concentration was determined by monitoring the [M+H] + peak for each VOC. This was done to allow creation of calibration curves (Figures 3 11, 3 12, and 3 13), as well as to not saturate the sample capillary, an issue that frequented ESI/MS analysis at much greater concentrations. Table 3 5 gives the limit of detection (LOD) and limit of quantitation (LOQ) for Car, MethAn, and Van, based on measure ment of the EtOH blank. The calibratio n curve for Van demonstrated an exponential increase in

PAGE 117

117 concentration rather than a linear rise. One explanation could be the fact that Van was the only compound where the neat sample was a solid rather than a liquid. This may have caused calculation of variable concentrations as the standards were measured by mass rather than by volume. However, it is more likely an issue of solubility of the compound. The single stock solution was prepared from the solid, and all stan dard dilutions were created from this, therefore any error associated with measurement of the solid should carry through linearly. Conversely, if the solubility fluctuates with the concentration, non linear quantification would be an expected result. The calibration curves allowed quantification of the compounds in the forthcoming breath studies. By direct infusion, Car and MethAn generated signal above the baseline at a concentration of 1 ppm, while Van was not detected until a concentration of 5 ppm was reached. The sharp increase in relative abundance of Car and MethAn at 1ppm could infer that low ppb levels for these compounds could be perceived if measured on the LTQ. For all compounds, c ollecting measurements at such lower concentrations would also im prove the accuracy of the calibration curves as well, since the sudden jump to 1 ppm seems to bias the curve toward higher intercept values. Prior to conducting breath analysis, e ach flavorant was analyzed by single ion monitoring (SIM) at their respective [M+H] + masses to ensure adequate signal of the most prominent peak for detection in breath . Moreover, MS 2 was performed on each [M+H] + ion to record characteristic fragmentation patterns for each VOC, also facilitating identification of more peaks present in the spectra. These fragmentation spectra are shown in Figure 3 14. Carvone pro duced four main fragment peaks for m/z 151

PAGE 118

118 observed at m/z 133, 123, 109 , and 93. These correspond to C 10 H 13 , C 8 H 11 O, C 7 H 9 O, and C 7 H 9 , or neutral losses of H 2 O, C 2 H 4 , C 3 H 6 , a nd C 3 H 6 O, respectively . 28 MethAn produced a sing le prominent peak from fragmentation of m/z 152 corresponding to C 7 H 6 NO ( m/z 120), likely from loss of MeOH. Van produced two primary fragment pe aks for m/z 153 observed at m/z 125 and 93. These correspond to C 7 H 9 O 2 and C 6 H 5 O , or neutral losses of CO and C 2 H 4 O 2 , respectively 28 Incidentally, having acce ss to the fragment ion patterns could prove useful if fragmentation were to occur when the samples were analyzed in breath, the compounds would still be able to be identified. Once flavorant analysis was complete, THC standards were analyzed by APCI/MS. Ch aracterization of THC is pivotal to FAIMS/MS experimentation to allow identification of primary, adduct, and fragmentation peaks prior to FAIMS separation being applied. A single concentration at 10 ppm was chosen for proof of concept experimentation via d irect infusion, as well as simulated breath analysis detailed in the following section. THC gave numerous adduct and frag mentation peaks (Figure 3 15). While not as significantly prominent as seen in the flavorant spectra, the [M+H] + peak at m/z 315 was ch osen to monitor THC in simulated breath. MS b reath sampling Analysis of breath was performed in conjunction with analysis by direct infusion experiments. Table 3 2 may be consulted for instrumental parameters. While EESI/MS of breath was indeed conducted , the data were similar to APCI/MS and will not be detailed in this work. Again, APCI is preferred to EESI as it is more comparable to the ionization found in the standalone FAIMS unit. The main distinction between the two methods was that for APCI, EtOH w ithout formic acid was used and injected at a flow rate of 20 µL/min rather than 10 µL/min. The increase in flow rate can be attributed to

PAGE 119

119 the fact that ESI (and hence EESI) affords better sensitivity at lower flow rates. Whereas the use of formic acid is only needed for EESI as a reagent ion since in APCI, a corona discharge needle is used to generate reagent ions; thus, the formic acid can be omitted. However, the mass spectral results, as well as many of the instrumental parameters were consistent. As wi th direct infusion, the [M+H] + ion of each VOC was monitored in SIM mode. Breath samples were doped at a range of concentrations of the VOC standards. This allowed determination of the minimum concentration of flavorant needed to dope breath and still obta in an observable signal. Knowing the signal would be much less intense for detection in breath, flavorants were spiked in breath starting only as low as 250 ppm. After each collection, if adequate signal was not observed, the concentration would be raised. This would continue until a signal was observed, in one instance reaching as high as 30,000 ppm. Most notably, there was visual distinction of each individual breath observed in the total and extracted ion chromatograms (TICs and EICs) displayed in Figure 3 16. The initial breath correlated with a sharp increase in the corresponding [M+H] + peak of each compound, and with each consecutive breath after the initial doping, the signal intensity at of the [M+H] + peak gradually decreased as the flavorant was qui ckly dissipated. On the other hand, the TIC produced a dip in intensity with each breath. This may seem counterintuitive to see a decrease in signal with each breath, but upon further experimentation the issue was resolved. The MS scan range started at m/z 50, however, many compounds in breath produce ions at lower masses. This was evident when a low mass scan was run for breath, revealed in Figure 3 17A. When setting the

PAGE 120

120 scan range to begin at m/z 20, the TIC demonstrates peak increases with each breath, r ather than the previous decreases. Both Car and Van allowed observation of the [M+H] + peak above the baseline at a minimum concentration of 250 ppm, while Van did not generate signal above the baseline until reaching a concentration of 30,000 ppm. Though, it should be noted that Van may well produce observable analyte signal at a concentration of 250 ppm, however, there seems to be a peak at m/z 153 produced by raw breath interfering with ed that the peak rose above the breath baseline, confirming Van was the present. Use of FAIMS prior to at lower concentrations. Utilizing the calibration curves with the trendline set to a y intercept of 0, the detected concentration of the compounds in breath was de termined. A concentration of 200 ± 10 ppb for Car, and 10 ± 0.4 ppb for MethAn was calculated from doping 250 ppm in breath . A concentration of 5,000 ± 80 ppb for Van was calculated from doping 30,000 ppm in breath. Predictably, the concentration detected in breath was much lower than the concentration of the standard used to dope the sample. Additionally, certain compounds seem to metabolize/degrade in the bre ath at different rates. MethAn produced the peaks of greatest abundance relative to the other compounds, yet it produced a peak in breath much lower than Car, even though both compounds were doped at the same concentration. One final remark on flavorant br eath analysis concerns the use of solvent with APCI for analysis. Experiments were conducted to examine whether addition of solvent

PAGE 121

121 through the APCI probe would be useful, or if ionization strictly by corona discharge of the air in the ionization chamber w as adequate as is. As demonstrated in Figure 3 18, utilization of a solvent (in this case MeOH) hinders ionization of the breath sample. Not only are the individual breaths less distinct, but the signal decreased by two orders of magnitude with solvent. Ad dition of 0.1 % formic acid to the MeOH solvent would likely counteract this ionization suppression. However, APCI of breath benefits more from ionization of the breath directly since a lack of solvent would imply greater portability. Therefore, portable b reath analysis is still a viable option as solvent is not needed, given a sufficient ionization technique. As before, THC was analyzed after completion of flavorant analysis. Described in detail above, a simulated breath system was utilized to introduce TH C to the MS in a gaseous form. Initially EESI/MS was performed (Table 3 2 outlines the MS parameters), demonstrating that the simulation system can achieve introduction of sample similar to human breathing, as evidenced in Figure 3 17. Further analysis was performed by APCI/MS. With both EESI and APCI, the bubbler system seemed to produce a distinct drop in THC signal in the minutes immediately following spiking of the standard. THC, along been known to adsorb to the surface of hydrophilic containers, such as glass due to reactive silicate and silanol groups on the surface . To mask these groups, thus decreasing the hydrophilicity, various reactive silanes may be used to coat the surface with a short polymer chai n or oil in a process called silanization. 106 Assuming this was the problem, a new set of glass bottles was obtained in order to test if silanization was necessary, and which procedure may be

PAGE 122

122 optimal, as there were two forms to choose from (gas and liquid phase) . All were ordered f rom the same manufacturer, in the same lot, and made of the same material; therefore , should be identical. Following the liquid and gas silanization procedures outlined by Seed, 106 and leaving one bottle unsilanized, three test bottles were created. To each bottle, 50 mL of MeOH w as added, and the breath simulation system implemented, first without any THC added. After 5 minutes of collecting system blanks, 0.5 mL 1 mg/mL THC standard was added to create a solution of around 10 ppm. This headspace was immediately, and continually c ollected for the next 10 minutes. The bubbler was switched off and the THC solution allowed to rest in the bottles for one hour. The solutions were then bubbled through with N 2 once again, to detect the THC present after an hour. The system was once more s witched off, and the bottles left capped overnight. The next day, 24 hrs after initial testing, the solutions in each of the three bottles were bubbled through one final time to detect THC. Results from this experiment are shown in Figure 3 19. Contrary to what was observed during initial testing, no loss of THC was found. For the initial tests, a clean, but used, HPLC pressure bottle was utilized in the simulation system. These later tests were performed on brand new HPLC pressure bottles. While the exact cause will remain a mystery, it can be inferred that a reaction occurred between the THC and some modified surface of the older glass bottle. Whether from some residual compound that was not removed by cleaning, or a more reactive glass surface from years of laboratory use and interaction with different solvent systems, it seems that the initial result of THC adsorption is the outlier. Therefore, silanization is no longer necessary. It should also be noted that the silanization process itself produced trace peaks in the

PAGE 123

123 background of the mass spectra. They are not observed visually in the provided spectra, but statistical analysis of the data (not detailed here) revealed a significant difference in the spectra between silanized and untreated glass. Standalon e FAIMS d irect detection of standards Upon characterizing the analytes of interest by MS, the potential for analysis by standalone FAIMS was evaluated. Initially, neat standards were detected with the ach compound. Table 3.4 outlines the FAIMS parameters set on the Lonestar device. As the device sustains a closed interface, only 63 Ni ionization could be utilized. A Varian rotary vane pump was attached to the exhaust port to pull the volatile samples int o the device and through the FAIMS chip, while the neat standard sat several centimeters away from the inlet ( Figure 3 20A). A rotameter between the pump and exhaust was used to control the flow, whereas the flow rate through the FAIMS cell was determined internally. The Lonestar provides calibration versus the differential pressure as a function of temperature, with sensors for temperature and pressure located immediately after the FAIMS chip in the instrument. 43 Due to their highly volatile nature, the neat standards were placed in an uncapped vial, allowing volatilization into the open air, before being pulled toward the device without preconcentration. Lastly, data on the Lon e star device was collected in both positive and negative ion modes simultaneously. The four flavorant compounds are all similar to one another in molecular weight, and distinction by a single stage of MS proved difficult for some of the species (i.e., MethSal and Van produce [M+H] + peaks at the same m/z value of 153). However, FAIMS has been shown to separate isomers, 39 , 45 , 47 , 48 and should allow adequate separation as the analytes vary in size and shape, and would therefore have differences in their mobilities.

PAGE 124

124 At the outset of any experimentation, the flow parameter needed to be optimized as altering the flow would greatly affect the data produced. This was determined by maintaining the minimum flow rate where adequate signal intensity was s till observed. All data shown was collected at a gas flow rate of 1.50 L/min. No carrier gas was added outside of the ambient air present, and all other parameters were kept constant. For analysis of standards, the neat compounds were used without any prio r dilution. Plots of dispersion voltage (DV) vs. compensation voltage (CV) vs. ion current (IC) were collected for the ambient air (as a blank) and each of the four VOCs. To create these plots, the device would set a DV to a single value and scan through t he CV ( 6 to 6 V). Upon completing the CV scan, the DV would be increased to a higher value, and the CV scanned again. This process was repeated over the selected range of DV, whereas CV could not be selected and must be run from 6 to 6 V. When selecting the DV range to be scanned, the value was specified as a percentage of the maximum field attainable on the device (250 V), or percent dispersion field (DF). For example, much of the data from these experiments was collected from 25 75% DF instead of 0 100%. This is due to the fact that all of the data produced below 25% DF did not produce any separation of the sample components since the field was not high enough to yield meaningful data ( Figure 3 21). Similarly, once a certain field strength was reach, typically around 70 80% DF, the energy of the field was too high, and fragmentation began to exponentially decay the sample. The number of steps of the voltage within the selected range could also be designated. Continuing with the previous example, a r ange of 25 to 75% DF might be chosen with a step size of 51; therefore, the DF will be incremented in such a way as to include 51 scans of the CV from 25 to 75% DF.

PAGE 125

125 Essentially, the DF will increase by a percent with every step and scan the CV. The greater the number of steps, the better the resolution of the peaks. All of the data presented here utilized 251 steps of the DV range. Once collected, the data could then be analyzed using the software provided by compound based on the 3D plot created. Figure 3 22 displays the dispersion plots collected from each of the four VOCs. Once a characteristic pattern is established, a specific DV value is chosen that would allow optimal separation of one, or several, of th e compounds from the background, or each other ( Figure 3 23). This was done visually, however, a more laborious process was explored by exporting the raw data as a .txt file and importing the file into Microsoft Excel. Then, the signal intensities at a sin gle DV could be plotted as a function of CV, illustrating a more detailed plot of separation. Figure 3 24 displays plots of signal intensity vs CV at DF values of 25, 35, 45, 55, 65, and 75%. Note that optimal separation of Car and MethAn is achieved aroun d 55%. Van and MethSal are better separated in negative ion mode as they correlate very close to ambient air in positive mode. If a similar plot were created for negative mode, there would be baseline separation of a Van and MethSal from ambient air, and a dequate separation between each other. Nevertheless, once a DV value is chosen, it could then be monitored as a function of time, where a plot of CV vs. time vs. IC could be collected at the specified DV. This type of plot allows monitoring of samples in r eal time for direct detection of neat standards, as well as analysis of standards in breath (detailed in the following section). Figure 3 25A displays three such plots demonstrating detection of individual

PAGE 126

126 standards, as well as distinct separation of a mix ture of two standards, all in real time. The right half of the figure, Figure 3 25B, shows a 2D cross sectional representation of a single point in time of the signal vs CV for each of the three plots. Knowing that the neat flavorants may be separated in r eal time, the next step was utilizing this same technique for analyzing the flavorants in breath. Standalone FAIMS b reath sampling All of the parameters were held constant from the previous section involving neat standards detection. No full dispersion p lots were collected for breath as they were unnecessary, the hope being the analytes would show up at the same voltages in breath. Additionally, it takes a minimum of 3 minutes to collect the most rudimentary dispersion plot, and breathing for that long, l et alone at a consistent flow rate, was not feasible. Therefore, breath analysis was solely performed utilizing the second type of plot, CV vs IC as a function of time. Similar to analysis of breath by MS, an Intoximeters mouthpiece was used to direct the breath toward the inlet. In this case, no housing was present, thus the mouthpiece was held aloft orthogonal to, and a few centimeters above, the Lonestar inlet ( Figure 3 4C). As breath was blown toward the inlet, the air flow from the pump attached to the exhaust pulled the sample through the device, same as with the neat standards. Once again, 10 µL of dilute standard was placed on the tongue prior to breath collection. When collecting breath data, the monitored DV was selected based on previous experimen ts with the neat standards. Figure 3 26 displays the culmination of this work by demonstrating real time detection of a flavorant in breath, and correlating it to the neat standard during a single analysis. Methyl anthranilate was doped in breath and blow toward the inlet twice, followed by detection of a neat standard of the same

PAGE 127

127 flavorant, before one final doped breath was collected, all in under three minutes. The peaks produced by each introduced sample, breath or neat, line up with what was seen by dir ect analysis of the neat standard. FAIMS/MS d irect infusion of standards After assessment of both standalone techniques, FAIMS and MS were paired to allow identification of the separated FAIMS peaks. These final two sections detailing FAIMS/MS analysis a re only preliminary, as the timing of the study would not allow a more in depth experiment to be performed. Thus, only THC standards were analyzed; no volatile flavorants were run. Table 3 3 outlines the instrumental parameters utilized, all experiments in this and the following section were run on the LCQ MS, standards were directly injected and ionized by ESI, the home built planar FAIMS cell was coupled to the front end of the MS, and all data were collected in positive mode ( Figure 3 27A). First, 10 ppm THC was run without FAIMS, producing a mass spectra moderately similar to th at seen from APCI/MS on the LTQ, where m/z 315 from the [M+H] + ion remains prominent. Once several peaks were chosen to monitor ( m/z 315, 343, and 371) , FAIMS separation was perfo rmed. The scan speed and frequency of the full size cell is much lower than that of the standalone device, thus a full CV/DV plot was not obtained. Instead, a DV was selected and monitored, and the CV was scanned from 5 to 5 V over a period of 2 minutes. An initial DV of 2,000 V was chosen and then increased in increments of 500 V up to 4,000 V. The data seemed to indicate that an optimal DV could be found between 3,000 and 3,500 V, so an incremental ramp was subsequently performed between these voltages a t steps of 100 V. At a DV of 3,300 V, optimal resolving power and signal intensity was achieved; accordingly, 3,300 V was chosen for a more detailed CV scan of 10 minutes ( Figure 3 28 A). Apart from the large

PAGE 128

128 peak around 0 CV, numerous smaller peaks were ob served by FAIMS/MS of the THC standard. Each of these peaks can be attributed to fragment and adduct ions formed from the THC solution, as demonstrated earlier. The large peak itself can also be divided into distinct ions by looking at EIC instead of the T IC. This indicates that as long as specific FAIMS parameters are maintained, one or several ions specific to THC may be monitored at a set DV and CV. However, the goal would be to monitor THC in breath, thus the simulation system was utilized once more to introduce THC headspace into FAIMS to assess its capabilities further. FAIMS/MS As mentioned above, this section details preliminary exploration of FAIMS/MS analysis of THC standards. MS and FAIMS instrumental parameters are listed in T able 3 4, the home built planar FAIMS cell was utilized on the LCQ, and the breath simulation system was employed to created THC spiked headspace to be sampled by EESI in positive mode. THC was spiked into the bubbler system (10 ppm) and the headspace was directed toward an auxiliary ESI spray of pure MeOH (5 µL/min) and the FAIMS inlet, as illustrated in Figure 3 27B. As previously, the DV was held constant while the CV was scanned from 5 to 5 V over a period of 2 minutes. Then, the DV would be stepped up , and the process repeated. For EESI/FAIMS/MS of THC, the optimal DV was found at 4,000V ( Figure 3 28 B). There is an evident CV shift compared to results from ESI/FAIMS/MS. Indeed, altering the DV will change the CV at which an ion will be detected; howeve r, the primary reason for the drastic shift here is the change in pressure and flow rate from switching to a headspace sample over a nebulized solution. When directly injecting the standard solution by ESI, there was no carrier gas in the

PAGE 129

129 FAIMS cell outsid e of the ambient air. For analysis of headspace by the breath simulation system, N 2 is used to pump the sample toward the FAIMS cell, and subsequently through it. This addition of N 2 as a buffer gas radically affects the ions in the FAIMS field and their e nsuing separation. Incidentally, the m/z peaks present remain essentially the same as from direct infusion. Thus, confirming the ability to analyze gaseous THC by FAIMS/MS. Further analysis involving addition of breath background is needed to fully assess the capabilities of this methodology, but proof of concept has been established from these results. Conclusion In this work, novel and conventional methodologies for breath introduction, collection, and analysis were developed and compared. Additionally, M S and FAIMS analyses were performed utilizing a variety of ionization techniques including APCI, ESI, EESI and 63 Ni. Experiments were conducted by MS, FAIMS/MS, and standalone FAIMS. The research presented involves the first use of a µFAIMS device to analy ze human breath, as well as the first instance of EESI/FAIMS/MS. Initially, breath introduction methods were developed and optimized, along with assessment of existing methods. A simple method of direct, on line breath introduction was established utilizin g an Intoximeters mouthpiece. The mouthpiece was deemed a better choice over Teflon tubing hooked up to the auxiliary port. The back pressure of the chamber is often too much to overcome by way of the narrow Teflon tubing so the breaths would be effectivel y stopped from even entering the chamber, or only a small portion of the breath would be analyzed. Also, the level of condensation permitted by the length of the tubing may impede the flow of molecules of interest in breath from exiting the tube. Still, ev en with the success of the mouthpiece, the efficiency of the

PAGE 130

130 method may not be optimal and other methods should be developed to supersede this one. In this work, the mouthpiece was either attached to source housing, or held aloft, in both cases directing t he breath toward the instrument inlet for detection. Where human breath sampling could not be implemented due to legal restrictions, a breath simulation system was created. This headspace generator allowed analysis of the illicit substance THC in a state m ore analogous to breath than delivered by direct infusion of standard solution. The first study entailed analysis of standard solutions of all compounds of interest by MS to identify characteristic analyte peaks. Identification of said peaks affords monito ring of these compounds in exhaled breath. APCI and ESI/MS provided the necessary characterization of the selected compounds. The most prominent peak in all of the analyte spectra, and those eventually monitored in SIM mode all correlated with the [M+H] + i on of the respective compounds. Moreover, establishment of LODs and LOQs were made possible by creation of calibration c urves derived from the MS data, also providing quantitation of the identified analyte peaks in breath. The same standards were then meas ured by APCI and EESI/MS, but this time in exhaled breath. The peaks identified by analysis of standards were monitored in breath to determine if real time detection was permitted. Indeed, direct detection of volatile analytes doped into exhaled breath was achieved. Again, SIM mode was utilized, monitoring the respective [M+H] + masses of each analyte. The breath simulation system Next, the use of a standalone FAIMS devi ce revealed that direct detection of volatile standards in both neat form and breath is feasible in real time. Standards could

PAGE 131

131 be distinguished by the specific chemical fingerprint produced by collecting a plot of DV vs. CV vs. IC. Furthermore, by monitori ng a single CV at a specified DV, various analytes in a mixture could be differentiated from each other. Not only does this work demonstrate separation of neat standards, but it displays a direct comparison of a breath sample doped with a standard and the neat standards itself. The result of this comparison revealed that both sample types produced peaks at the same CV and DV. Leading to the most important conclusion of this work, that detection of VOCs in breath was possible on a micromachined FAIMS chip, d emonstrating portable analysis in real time. Finally, FAIMS/MS analysis was performed on THC by both direct infusion of standards and direct introduction of analyte spiked headspace. ESI and EESI were utilized depending on the sample type, but both provide d similar results. THC and its adducts can be separated by direct, real time FAIMS analysis, and those peaks may be subsequently identified by MS. Overall, FAIMS has proven to be a viable technique for breath analysis. By itself, FAIMS offers the ability t o produce characteristic fingerprint spectra which may be used to identify compounds of interest. Besides, when coupled to MS, those characteristic peaks may be subsequently be identified. It is the only technique that can provide rapid analysis in a porta ble manner. However, the sensitivity and selectivity of the method remain an issue as they both see a sharp decline as the device is made smaller and more portable. Yet, if a compromise between absolute miniaturization and adequate resolution may be found, such a device will revolutionize the biomedical and diagnostic fields.

PAGE 132

132 Figure 3 1. Diagram of pulmonary alveoli. Pulmonary alveoli are chambers of air found in the lungs with capillaries tightly wound around them. Molecules in the blood may diffuse th rough the capillary membrane into the open air of the alveolar sac and be expelled fr om the lungs in exhaled breath adapted and reprinted by courtesy of Encyclopædia Britannica, Inc., copyright 2006 ; used with permission . 107

PAGE 133

133 Figure 3 2. Diagram of how molecules in the blood are exhaled in breath. Molecules that pass through the capillaries tightly wound around alveoli (A) may diffuse through the wall membrane and into the open air of the sac (B) . Then, i f the kinetic motion of the molecule and the pressure inside the alveoli are high enough, the molecule will be expelled from the lungs into the breath.

PAGE 134

134 Table 3 1. Comparison of select analytical methods for use in breath analysis . Advantage GC/MS PTR/MS & SIFT/MS EESI/MS FAIMS Sensitivity Selectivity * Nonvolatile analytes Real time Portability

PAGE 135

135 Figure 3 3. Molecular structures of methyl salicylate, methyl anthranilate, (R) carvone, vanillin, and ( ) trans 9 tetrahydrocannabinol .

PAGE 136

136 Table 3 2. MS parameters for LTQ experimentation on human exhaled breath . Parameter ESI APCI THC EESI and APCI Full Scan Range ( m/z ) 50 500 50 350 50 650 Infusion Flow Rate (µL/min) 7 20 5 10 Sheath Gas (a.u.) 65 20 10 Spray Voltage/Current 6.0 kV 2.0 µA 4.5 kV Capillary Temperature (ºC) 47 35 35 Capillary Voltage (V) 100 200 200 Tube Lens (V) 70 60 60 Vaporizer Temperature (ºC) N/A 50 0

PAGE 137

137 Table 3 3. MS and FAIMS/MS parameters for LCQ experimentation . Parameter ESI and E ESI FAIMS/MS Full Scan Range ( m/z ) 150 400 Sheath Gas (a.u.) 0 Spray Voltage/Current 6.0 kV Capillary Temperature (ºC) 300 Capillary Voltage (V) 3 Tube Lens (V) 10 FAIMS N 2 Flow Rate (L/min) 0.4 FAIMS Frequency (kHz) 750 FAIMS Max. App. DV (V) 5,000

PAGE 138

138 Table 3 4. FAIMS cell dimensions and parameters. Parameter Home built Planar FAIMS Lonestar µFAIMS chip Analytical Gap (mm) 2 0.035 Path Length (mm) 65 0.300 Entrance Width (mm) 20 2.465 Maximum Applied Dispersion Voltage (V) 5,00 0 250 Compensation Voltage Range (V) 5 to 5 6 to 6 Waveform Frequency (kHz) 750 28,600 Gas Flow Rate (L/min) 0.4 1.5 Cell Temperature (ºC) 55 50

PAGE 139

139 Figure 3 4. Intoximeters mouthpiece (A) fitted into the IonMax ionization chamber on the LTQ for MS analysis (B), and held around 6 cm from the Lonestar inlet for FAIMS analysis (C ) p hotos courtesy of author .

PAGE 140

140 Figure 3 5. Home built planar FAIMS cell (A) and µFAIMS chip in the Lonestar device (B). Picture of each cel l (top left), picture for size comparison where the home built cell fits in the palm of a hand and the chip fits on the tip of a finger (top right), and 3D rendering of each displaying the dimensions of the cell where the blue arrow represents the flow of ions through the cell (bottom). p hotos of µFAIMS chip in (B) reprinted with permission from Shvartsburg et al . 43 and Owlstone Nanotech Inc , 98 photo of planar cell in (A ) courtesy of author .

PAGE 141

141 Figure 3 6. Displays a 3D (A) and 2D (B) representation of the µFAIMS chip in the Lonestar. In (A) The green top plate is the deflector plate, the blue and orange represent gold plating that creates four different electrodes (orange, light orange, blue, and light blue) from the potential applied across the chip (blue to orange) and the potential bias through the chip (difference in color shading), and the bot tom black plate is the detector. In (B), the dark grey region indicates the interdigitated silicon filter, the light grey region indicates the etched substrate used to properly space the FAIMS region from the detector plate, the dotted blue line represents the flow of ions through the chip, and the other colors are the same as in (A). Note this is not drawn to scale, and the number of digits was reduced to allow an easier interpretation.

PAGE 142

142 Figure 3 7. Photo of the breath simulation system u sed to analyze THC in headspace. N 2 flows through the restrictor, into the bottle, and out through the bubbler. The continual addition of N 2 begins to create a headspace gas spiked with the analyte in the solution through which the gas is bubbled. The pres sure build up is alleviated by the headspace flowing out of the bottle through the second tube, the tip of which is directed toward the MS inlet photo courtesy of author .

PAGE 143

143 Figure 3 8. Diagram of the breath simulation system on the LTQ MS. N 2 gas was bubbled through methanol spiked with a known concentration of THC creating a THC saturated headspace (bottom). The headspace was directed toward the inlet of the MS to be ionized by EESI (inset, top).

PAGE 144

144 Figure 3 9. ESI/MS spectra for 250 ppm Car (A), MethAn (B), MethSal (C), and Van (D) in EtOH, and pure EtOH (E) on the LTQ. Average over approximately 100 scans. [M+H] + ion is most prominent in VOCs, [2M+H] + is most prominent in EtOH, m/z 65 corresponds to a reagent ion, and m/z 211 is a c ontamination peak.

PAGE 145

145 Figure 3 10. APCI/MS spectra for 250 ppm Car (A), MethAn (B), Van (C) in EtOH, and pure EtOH (D) on the LTQ. Average over approximately 100 scans. [M+H] + ion is most prominent in VOCs and [2M+H] + is most prominent in EtOH.

PAGE 146

146 Figure 3 11. Calibr ation curve for Car. Each point is the average of three measurements, the mselves the average of 15 scans. Error bars correspond to ± 1 standard deviation.

PAGE 147

147 Figure 3 12. Calibration curve for MethAn. Each bullet is the average o f three measurements, the mselves the average of 15 scans . Error bars correspond to ± 1 standard deviation.

PAGE 148

148 Figure 3 13. Calibration curve for Van. Eac h bullet is the average of 3 measurements, each the average of 15 scans. Error bars correspond to ± 1 standard deviation.

PAGE 149

149 Table 3 5. LOD and LOQ for Car, MethAn, and Van by APCI/MS on the LTQ . Compound m/z Limit of Detection (ppb ) Limit of Quantitation (ppb ) (R) Carvone 151 200 ± 30 200 ± 90 Methyl Anthranilate 152 40 ± 7 40 ± 20 Vanillin 153 30 ± 3 30 ± 10

PAGE 150

150 Figure 3 14. APCI/MS/MS spectra for 250 ppm Car: NCE = 35%, m/z 151 (A); MethAn: NCE = 31%, m/z 152 (B); and Van: NCE = 48%, m/z 153 (C) in EtOH, on the LTQ. Average over approximately 150 scans.

PAGE 151

151 Figure 3 15. APCI/MS spectra for 10 ppm THC (MW = 314) in MeOH, on the LTQ. Average over 250 scans . Peak of note is observed at m/z 315 ([M+H] + ), this was chosen to monitor for future studies.

PAGE 152

152 Figure 3 16. APCI/MS chromatograms and spectra collected by spraying EtOH solvent and breathing into the LTQ at specific intervals Each decrease in intensity in the chromatograms (A) corresponds to single breath. Peak at m/z 151 (Car [M+H] + peak) is initially insignificant ((B), top), but upon introduction of breath m/z 151 increases by two orders of magnitude ((B), bottom) .

PAGE 153

153 Figure 3 17. TICs of unaltered human breath (A) and simulated breath (B) from direct analysis by APCI/MS on the LTQ. All of the peaks from breath sampling (A) were produced by sam pling raw exhaled breath. For simulated breath (B), pure MeOH headspace was detected for 5 minutes, then 10 ppm THC was spiked into the MeOH and detected for the remainder of the time .

PAGE 154

154 Figure 3 18. APCI/MS TICs for exhaled human breath with no so lvent (A) and with MeOH as solvent (B) on the LTQ. Starting around 0.3 min, three raw breaths were detected followed by three breaths doped with analysis of breath with no auxiliary solvent produces more intense and define peaks than observ ed for solvated APCI analysis.

PAGE 155

155 Figure 3 19. E IC for m/ z 315 from APCI/MS analysis of three simulated breaths of pure MeOH followed by simulated breath spiked with THC revealing immediate drop in signal after THC added at 5 min mark (A), perfor med in old bottle, on the LTQ. TIC from APCI/MS analysis of three simulated breaths of pure MeOH followed by simulated breath spiked with THC revealing a gradual increase in signal after initial addition of THC at 4.5 min (B), performed in newly purchased bottle, on the LTQ. TIC from APCI/MS analysis of simulated breath an hour after addition of THC (C), and 24 hrs after addition of THC (D) in new bottle from (B) on the LTQ .

PAGE 156

156 Figure 3 20. Diagrams of sample introduction into Lonestar. In (A), n eat st andard placed in open vial where its volatile nature allows it to flow into the open air before being pulled into the device by a pump on the exhaust of the Owlstone Lonestar . In (B), b reath directed toward Lonesta r inlet by an Intoximeters mouthpiece, hel d orthogonal to the in let about 3 5 cm away. A Varian pump was attached to exhaust port, pulling air through the system at 1.5 L/min.

PAGE 157

157 Figure 3 21. Displays 3D dispersion plots of DF vs CV as a function of signal intensity for ambient air in positive ion mode (left) and negative ion mode (right). Illustrates how halting data collection below 25% and above 75% DF does not affect collection of pertinent data. Below 25% DF, there is always a single reagent ion peak. Above 75% DF there is rarely any useful signal as it becomes too scattered and low in intensity from the high energy of the field.

PAGE 158

158 Figure 3 22 . 3D dispersion plots of DF vs CV as a function of si gnal intensity for Van (A), MethSal (B), Car (C), and MethAn (D) in positive ion mode (le ft) and neg ative ion mode (right). Data were collected from 25 to 75% DF, with 251 DV steps and a single CV scan from 6 to 6V, at 55 °C.

PAGE 159


PAGE 160


PAGE 161

161 Figure 3 23. 3D dispersion plots of DF vs CV as a function of signal intensity, for Ambient Air (A) and Van (B) in negative ion mode. Data were collected from 25 to 75% DF, with 251 DV steps and a single CV scan from 6 to 6V, at 55 °C. 53% DF demonstrates best separation where Van peak can be efficiently distinguished from the background air (yellow arrow).

PAGE 162

162 Figure 3 24. Plots of ion current (signal intensity) vs CV for ambient air and four VOCs at the specified DF %.

PAGE 163


PAGE 164


PAGE 165


PAGE 166

166 Figure 3 25. (A) 3D plots of CV and time as a function of signal intensity (DV = 146V or 53% DF) for neat standards of methyl anthranilate (top), (R) carvone (middle), and a mixture of the two (bottom). (B) Cross section taken around 35 seconds and displayed in 2D as intensity (IC) vs CV of the same neat standards.

PAGE 167

167 Figure 3 26. (A) 3D plot of CV a nd time as a function of signal intensity (DV = 146V or 53% DF) demonstrating real time detection of methyl anthranilate, both as a neat standard and as a standard solution doped into human breath. (B) Cross sections taken from each sample (one for each br eath and one for the neat standard) displayed in 2D as intensity (IC) vs CV to illustrate CV correlation.

PAGE 168

168 Figure 3 27. ESI (A) and EESI (B) FAIMS/MS on the LCQ . ESI involved spraying analyte in solvent directly into a home built planar FAIMS cell in parall el with the MS inlet. EESI involved pure solvent sprayed from ESI needle toward a home built planar FAI MS cell (~45°) while analyte headspace was directed toward the ESI spray in parallel with the FAIMS cell and MS inlet photos courtesy of author .

PAGE 169

169 Figu re 3 28 . Plots of a bundance vs CV which demonstrate detection of 10 ppm THC in MeOH by direct injection ESI/FAIMS/MS (A) and headspace EESI/FAIMS/MS (B) on the LCQ. Shift of peak for EESI is primarily a result of addition of N 2 carrier gas vs none for ESI, while resolution change can be attributed to scanning CV range over 10 min (ESI) vs 2 min (EESI). ESI: 3,300 DV while scanning CV from 5 to 5 V over 10 minutes. EESI: 4,000 DV while scanning CV from 5 to 5 V over 2 minutes.

PAGE 170

170 CHA PTER 4 METABOLOMICS OF MELANOMA: IDENTIFYING AND CHARACTERIZING POTENIAL BIOMARKERS OF DISEASE BY MASS SPECTROMETRY Introduction Arguably the most researched medical topic in the world, cancer is the subject of myriad debates over the variety of potential causes and treatments. While the term itself is ubiquitous, the processes that lead to its formation in the human body are still widely unknown. 108 , 109 In fact, there is not a single, solitary cause of cancer. Cancer is a broad term used to define over 100 different diseases, all of which have different causes, originate in different tissues, and develop for vastly different reasons. 108 Therefore, each type of cancer requires its own unique treatment. The body is made up of many kinds of cells. These cells gr ow and divide in a controlled manner to produce more cells as they are needed to keep the body healthy. When cells become old and damaged, they die and are replaced with new cells (Figure 4 1). 109 Sometimes however, this orderly process goes wrong. DNA of cells can become damaged, producing mutations that a ffect normal cell growth and division. 109 When this happens, the cells no longer die when they should, and new cells form when the body time, these cancerous cells then split off into lymph channels of the blood stream durin g a process called metastasis. 108 , 109 Left unchecked, the disease will rapidly invade the rest of the body, which leads to debilitating illness, and in many cases, death. Cancer is often named for the or gan or type of cell in which it originates. The nomenclature is split up into many broad categories, and from there, further divided into numerous sub types. 108 , 109 Some well known examples include leuke mia present in blood forming tissues, such as bone marrow; sarcoma malignancy of the bone,

PAGE 171

171 cartilage, fat, muscle, and other connective tissues; and lymphoma affects the lymphatic system, including the immune system and white bloods cells. This resea rch will focus on carcinoma , cancer that begins in the epithelial tissues that line many internal organs and the largest external organ, skin. 110 More specifically, this work involves the analysis of melanoma , one of a number of sub types comprising carcinoma (e.g., basal cell, squamous cell, and transitional cell carcinomas). 111 Melanoma originates in skin cells called melanocytes, responsibl e for producing the protective skin darkening pigment melanin. 110 As with many types of cancer, melanoma starts with an abnormal pro liferation of cells (Figure 4 2); y et, it is still unclear how or why this initial process of abnormal growth begins. Melanoma typically radiates from the manifestation site up to a size of around 4 mm, and can eventually lead to skin ulceration and breaka ge. 108 Moreover, according to the American Academy of De rmatology , melanomas can reach sizes of 6 mm or greater in the latest stages. 111 While propagation along the surface is cause for concern, it is when the cells begin to descend vertically down into the lower layers of the epidermis that melanoma becomes truly worr isome. Each stage of melanoma is defined by the size of the tumor and the spread of cancerous cells throughout the body. 110 Stage 0 is defined by the initial abnormal proliferation of cells. Stages 1 and 2 are characterized by radial and vertical tumor growth. Stage 3 is reached once malignant cells spread to local lymph nodes. Upon breaking through the basement membrane and invading t he underlying subcutaneous layers to reach the blood stream, the melanoma has reached Stage 4, metastasis. This is the final stage of melanoma development, as the malignant cells disperse throughout

PAGE 172

172 the body. Survival rates observed by the American Cancer Society 108 reveal a dramatic decrease in five and ten y ear life expectancies with each progressive stage of melanoma (Figure 4 3). The main drawback with current melanoma detection methods is the reliance on crude visual interpretations of moles (Figure 4 4). 110 , 111 If a physician or derma tologist thinks a patient might have a malignant tumor, they biopsy the lesion and send it to a pathologist for further analysis. Multiple screenings and assays are performed on the e present, and what stage may have been reached. If melanoma is found, the dermatologist will excise the lesion to try to remove all of the cancerous cells. However, there is no way of knowing if all of the cancer has been completely removed. Additionally, until the 3, and there is no indication other than blood tests for the cancer having metastasized. 111 Moreover, biopsies often leave a highly visible scar, and are incredibly invas ive. The absence of a quick, noninvasive screening for melanoma is what provided the rationale for this research. By affording medical professionals definitive justification prior to excision of a potentially cancerous lesion, the research presented here could alleviate unnecessary biopsies (hence preventing unwanted scarring), while more notably combating the late discovery and lack of specificity associated with current detection schemes. Early discovery of unique biological markers is a priority in all aspects of cancer related research and clinical care. Earlier detection methods and more accurate markers result in lower mortality rates and higher quality of life for the patients. 112 This

PAGE 173

173 research specifically focuses on identifying and characterizing metabolic biomarkers for melanoma. There are several reasons for utilizing metabolomics over other common methods for disease classification, such as proteomics or genomics . The primary advantages of analyzing the metabolome involve (1) a close relation to the phenotype, (2) greater specificity, and (3) less volatile markers (Figure 4 5). More than any oth er compound class, metabolites are intimately linked to phenot ype, hence they should provide markers with the greatest specify, and hopefully a better understanding of the melanoma pathology. Furthermore, due to metabolites inherently low molecular weight, the small molecules should prove to have a higher volatility, meaning better correlation to less invasive testing. Cancer detection based on volatile anal ysis can be traced back to 1989 , 113 where a medical lett er was published detailing a case where a dog noticed the presence of a detection has only seen a resurgence in the past decade. Most studies still focus on the use of ani mals 114 118 and devices 119 122 ; however, utiliz ation of ambient ionization and separation methods will allow real time, direct analysis of biomarkers for melanoma, in addition to identification of new markers. Recent studies observed an apparent difference between the volatile metabolome of melanoma (c ancer) and melanocyte (normal) cells, but only after a series of sample preparation steps. 112 , 117 , 123 125 Furthermore, analyzing metabolites in a more native environment (i.e., diseased tissue samples and swabs of potentially cancerous skin lesions) could lead to a greater number of, and more accurate, bioma rkers. Therefore, the identification of additional biomarkers, as well as the

PAGE 174

174 development of a real time method incorporating little or no sample preparation, could revolutionize current melanoma detection schemes. Initially, the research plan was to achie ve the following: (1) improve current methodologies for biomarker detection, (2) validate current potential markers for melanoma as well as identify new ones, (3) develop a noninvasive diagnostic test for melanoma, and (4) employ this technique through a clinically applicable device. Time and sample constraints have limited investigation to the first two goals, as well as strides into the third, all of which will be presented here. However, future directions of this research and ideas for continuation of t his work as related to those aims left unaccomplished will also be outlined. Experimental Chemicals, Reagents, and Biological Samples Solvents and standards Ammonium formate (AF) solid and HPLC grade methanol (MeOH) and water were purchased from Fischer Sc ientific (Fair Lawn, NJ). Optima LC/MS grade acetonitrile (ACN) and 0.1% formic acid (FA) in water were also purchased from Fischer Scientific (Fair Lawn, NJ), and utilized as mobile phase solvents. An internal standard metabolite mix (IS) was purchased fr om C/D/N Isotopes Inc. (Pointe Claire, Quebec, CA); consisting of deuterated isotopes of creatine, L leucine, L tryptophan, and caffeine. All reagents were stored at room temperature, except for the IS ( 2 0 °C) unless otherwise stated. Mammalian cell cultu res and media All frozen cell cultures and media, as well as phosphate buffered saline (PBS) and fetal bovine serum (FBS) were purchased from ATCC (Manassas, VA), unless

PAGE 175

175 spec ifications. Two primary epidermal melanocyte cell lines, PCS 200 013 and PCS 200 012 (consisting of human adult and neonatal cells) were grown and maintained on Dermal Cell Basal Medium (DCBM) supplemented with a melanocyte growth kit specific to each cell type. These lines were used as controls to distinguish from the cancerous cells. Two human melanoma cell lines (SK MEL 28 and A 375) were grown and went through ten passag es, a third cell line comprised of half SK MEL 28 and half A 375 was created to minimize cellular variance between cell lines. In addition, while EMEM is recommended for these cancer lines, they were also grown and maintained on DCBM supplemented with the adult melanocyte growth kit to provide a growth environment as close as possible to that of the normal melanocyte cell lines. This helped to diminish the number of distinct features between melanoma and normal cells found in the LC/MS spectra, thus reducin g erroneous biomarkers. All four cell lines were initially optimized and also maintained on Isotopic Ratio Outlier Analysis (IROA) medium consisting of either 5% 13 C/ 12 C or 95% 13 C/ 12 C to identify putative pathways of metabolic dysregulation in melanoma. T his media was purchased from IROA Technologies (Bolton, MA). Cell lines were received and immediately stored under liquid N 2 , while media was stored at 20 °C or below until use. Cell cultures were grown in accordance to the respective handling, initiation , and subculture procedures as outlined by ATCC. 126 To initiate the first cell culture, the frozen cells were r emoved from liquid N 2 , and thawed in a 37 °C water bath for 2 5 minutes. The corresponding media (12 mL) was added to a

PAGE 176

176 culture dish for each cell line, and the dishes placed in an incubator (37 °C , 5 % CO 2 , 95% humidity) for 30 minutes. The thawed cells we re combined with 1 mL of culture media, then transferred to the warm culture dish. Cells were then incubated for at least 24 hours. Afterward, the dishes were observed under a microscope to determine percent cellular confluence. If 80 90% confluency was no t reached, the top layer of media was removed, without disturbing the cells, and 13 mL of fresh media was added. Confluence determination and media exchange were repeated until cells reached 80 90% confluence, at which point the cells were subcultured. Sub culturing procedure differed between the melanoma lines and the melanocyte lines. For melanocytes, media was warmed to 37 °C in a warm water bath, while cells were examined for % confluence. If confluent, the media was removed and the adherent cells rinsed twice with 2.5 mL PBS. To detach the monolayer of adherent cells from the culture dish, 0.05 M trypsin EDTA (5 mL) was added to each dish. The culture dish was then placed in an incubator ( 37 °C, 5 % CO 2 , 95% humidity) for 10 20 minutes to assist the detac hment process. Cells were observed under a microscope to ensure complete detachment. Once detached, trypsin neutralizing solution (5 mL) was added to each dish, and the entire cell solution was transferred to a conical tube. Each dish was again rinsed twic e with 2.5 mL PBS, to ensure complete removal of all cells, with the rinse solution being added to the conical tube with the cell solution. All conical tubes were centrifuged at 1500 rpm for 5 minutes. The subsequent supernatant was removed and the pellet re suspended in fresh growth media. To subculture the cells, a 1:2 split was performed, where half the media/cell solution was transferred to a new dish and diluted up to 13 mL with a new portion of media. The other half of the split solution was placed

PAGE 177

177 ba ck in the centrifuge for 5 7 min (1500 rpm), the supernatant was removed, and the subsequent pellet was stored at 80 °C until analysis. The subculturing procedure for the melanoma lines was similar, but simpler. Once again, the media was warmed to 37 °C i n a warm water bath while cells were examined for % confluence. If confluent, the media was removed and 3 mL (or 13 mL for later passages in a larger culture flask) trypsin EDTA was added to the dish to ensure detachment. The cells and trypsin were placed in an incubator (37 °C, 5 % CO 2 , 95% humidity) for at least 45 minutes. After incubation, cells were examined for detachment from the dish. If not detached, they were placed back in the incubator until fully detached . The cell solution was then centrifuged (1500 rpm, 5 minutes) and the supernatant removed. The cells were re suspended in new media and split 1:2 as detailed above. Half was used for a new passage (diluted to either 13 mL or 26 mL with fresh media depending on the size of the culture flask), and half was pelleted via centrifugation and stored at 80 °C for future analysis. Biopsied t issue Human skin tissue biopsies were collected by the UF CTSI Biorepository and de identified. Ten of these skin tissue samples were obtained from the Biorepository based on sex, age, and disease state of the tissue, five cancerous and five normal tissue specimens were selected, three male and two female for each. Tissue samples were stored at 80 ºC until analysis. Protein quantitation Qubit Protein Assay Kit consist ing of concentrated assay reagent, dilution buffer, and three pre diluted protein standards. The reagent was diluted with the buffer and the analyte to be quantified was added at the volume recommended by the accompanying

PAGE 178

178 procedure manual. Each standard wa s at a different concentration, and all were measured in turn , allowing calibration of the fluorometer along a dynamic concentration range. The reagent/analyte mix was then measured and quantified based on Equation 4 1; where QF value is the concentration given directly by the fluorometer and x is the µL of analyte added to the reagent tube. Dilution of the analyte needs to be corrected, hence utilizing the equation to calculate the true concentration. True concentration = QF value (4 1) Instru mentation Data were collected on three instrumental setups: 1) Thermo Scientific Q Exactive MS (QE) applying heated electrospray ionization (HESI), see Table 4 1 for QE MS parameters as well as the chromatographic parameters and gradient. The QE allowed da ta to be collected in high resolution, as well as polarity switching mode , which affords the alternating acquisition of both positive and negative polarity data. This was coupled with a Thermo Dionex UltiMate 3000 pump using an ACE Excel 2 C18 PFP column ( 100 x 2.1 mm) for UHPLC separation. An autosampler was used for sample injection and was held at 4 °C for all experiments. 2) Thermo Scientific LTQ XL MS (LTQ) employing atmospheric pressure chemical ionization (APCI), see Table 4 2 for LTQ MS parameters. 3) Agilent 6220 time of flight MS ( TOF) utilizing direct analysis in real time (DART). The TOF also allowed acquisition of high resolution mass spectral data, see Table 4 3 for TOF MS parameters.

PAGE 179

179 The Qubit Protein Assay was performed on a Qubit 3.0 Fluorom eter to quantify the protein concentration by sample fluorescence compared to that of select standards, allowing normalization of the mammalian cell culture data. Throughout all cell culturing, a Thermo Scientific Sorvall Legend RT Plus centrifuge was util ized. Biological Sample Preparation Mammalian cell metabolite isolation To prepare the cell cultures for MS analysis, pellets stored at 80 ºC were thawed on ice. Excess supernatant was removed if present and the pellets were washed three times with 1 mL i ce cold AF. For the first two washes, AF was added on top of the pellet without vortexing, the pellet and solution were centrifuged at 1500 rpm for 5 minutes, and the supernatant was removed. For the third wash, AF was added to the pellet, and the solution vortexed. At this point, several passages of a given sample were pooled together and vortexed again. An aliquot of this combined solution was removed (10 µL) and saved for protein quantitation, before the combined solution was aliquoted into equivalent pr edetermined portions . Centrifugation (1500 rpm, 5 minutes) was again performed on the solutions, then the supernatant was removed, and 10 µL IS was added to each pellet. After, 2 mL of ice cold 80% MeOH/H 2 O was added to lyse the cells and isolate the metab olites. The lysate supernatant and precipitate were separated by centrifugation (1500 rpm, 5 minutes), and both fractions (precipitate and supernatant) were collected for further analysis . The supernatant was dried under N 2 at 32 ºC and reconstituted in 30 µL 0.1% FA/H 2 O for UHPLC/MS or 250 µL 80% MeOH/H 2 O for APCI/MS and DART/ MS.

PAGE 180

180 Biopsied tissue preparation Tissue samples were removed from 80 ºC storage and placed directly into a 20ºC environment for sectioning. DART is an open air surface ionization te chnique so the thickness of each section and consistency of size from sample to sample were not of the utmost importance. Therefore, no special equipment was used for sectioning, only a sterilized straight razor to cut the tissue and sterilized tweezers to hold the tissue in place. Samples around 1 2 mm 3 were sectioned from the whole tissue sample. Once sectioned, the whole biopsied tissue was placed back into storage at 80 ºC, while the sample sections were analyzed by DART/ MS without any pretreatment or extraction. For analysis, sample sections were held in the DART stream with sterilized tweezers for around 3 7 seconds at a time, multiple times per run. Multivariate Data Analysis To determine significant differences (features) between normal and cancer s amples, several methodologies for statistical analysis were applied. The two primary methods were a supervised method, partial least squares discriminant analysis (PLS DA), and a second similar, yet unsupervised method called principal component analysis ( PCA). Other methods utilized include analysis of variance (ANOVA), hierarchical clustering analysis (HCA), random f orest (RdF), and t test. All methodologies require the use of specialized software to perform feature identification and reduction for large data files, 51 this research made use of the XCMS package in RStudio, 127 129 XCMS provided feature reduction for all complex LC/MS data. Once processed, multivariate analysis was performed on the subsequent data using MetaboAnalyst web server. 52 , 130 , 131 For data collected on the LTQ and TOF instruments

PAGE 181

181 where no LC was present, the data were submitted directly to MetaboAnalyst without prior use of XCMS since no r.t. alignment and peak picking algorithms were needed . Several databases were consulted for com pound identification of select m/z values: Human Metabolome Database (HMDB), 132 134 mzCloud, 28 and Scripps Center for Metabolomics metabolite database (METLIN). 135 Chromatographic statistical analysis workflow For data run on the QE instrument, where UHPLC was implemented, each individual sample undergoing statistical analysis consi sted of a minimum of three Convert, each replicate was converted from a .RAW file produced by the QE to .mzXML so it could be read by R Studio . The converted files for all sampl e types to be ana lyzed were uploaded into RStudio and run through the XCMS R script for feature isolation and reduction, and retention time (r.t.) alignment. By assigning a specified ppm error to the calculation in XCMS, the program peak picks m/z values t hat are within error from replicate to replicate, and correlates them with each other replicate from every sample. This creates a single Microsoft Excel file that produces correlating columns of median m/z value, median r.t., and peak area for each individ ual replicate of every sample uploaded, in addition to many more columns of features. All other columns were ignored at this juncture, except for these three (for each replicate), and the m/z and r.t. columns were combined into a single column containing b oth features. Each replicate was assigned into its proper sample group class, and then the file was saved as a .CSV to be run in MetaboAnalyst. This single file containing m/ z, r.t., and peak area data for every replicate of every sample type to be compare d was uploaded into the MetaboAnalyst web server for

PAGE 182

182 statistical analysis. MetaboAnalyst offers a wide range of options for normalization and further processing of the uploaded data. As the data were peak matched by XCMS, no further matching was required; however, the data were filtered by identifying and removing variables that were unlikely to be of use when modeling the data. This was done using a robust interquartile range (IQR) estimate which, due to the extremely high number of variables, filtered aro und 40% of the data. 131 Following this, signal intensities were normalized to account for signal variabil ity due to differences in the size of the sample (e.g., cell pellets could vary significantly in the number of cells contained within each). Intensities were normalized based on their total protein count calculated from the Qubit Protein Assay. A log trans formation was also utilized for some of the data analyses in order to further normalize the signal intensities by assigning the same weighting to high and low level data. Finally, all of the data were autoscaled (the values are mean centered and divided b y the standard deviation of each variable). Once fully processed, the data were run through PLS DA and PCA algorithms, amongst others, to produce statistically differentiated data. Non chromatographic statistical analysis workflow For data run on the LTQ a nd TOF instruments where no chromatographic separation was implemented, each individual sample undergoing statistical analysis consisted of a minimum of three replicates (three separate MS runs of the same sample). Each replicate was created by exporting a n average of at least 20 MS scans from the .RAW file, and importing the data into Microsoft Excel, producing two columns ( m/z and intensity). Replicates of a given sample type where saved as individual .CSV files before being placed into separate folders f or each different sample or sample type, depending on the extent of comparison being performed. Once every sample type had

PAGE 183

183 its own folder compiled of at least three replicates each, all of these were placed into a single folder and compressed to a .ZIP fol der. In some cases, if the data had a high abundance of 0 value intensities, all of the 0 intensities and their corresponding m/ z values were removed as their presence would cause an error when run through MetaboAnalyst. This folder was then uploaded to th e MetaboAnalyst web server for statistical analysis. As previously mentioned, the data may be further processed and normalized prior to statistical comparison. For this data, mass tolerances of 0.1 m/z for HR data ( TOF ) and 0.5 m/ z for nominal mass data (L TQ) were utilized to counteract possible mass shifts from run to run or due to space charge effects. Following peak matching, the IQR estimate was again used to filter the data further. For HR data, a percentage similar to chromatographic data were filtere d, however, due to the low number of variables in the nominal mass data, only 5 10% was filtered. Following this final filtering, signal intensities were again normalized to account for signal variability. Intensities from HR data were normalized based on their total protein count, while nominal mass was normalized by sum. Similar to chromatographic data, a l og transformation was utilized for some of the data, while all of the data were autoscaled, prior to being run through PLS DA and PCA algorithms (among st others) to produce statistically differentiated data. Results and Discussion All data were collected on one of three distinct instrumental setups. Each instrumental method afforded discrete advantages, as outlined in Table 4 4. LC/MS was utilized first as the current gold standard for metabolomics analysis. This allowed validation of more novel techniques used later on, and provided identification for the first

PAGE 184

184 round of putative biomarkers. As explored in more detail later in this chapter, compound ident ification of certain m/z values proved difficult due to the overlap of numerous potential compounds. In these instances, the use of the second instrumental technique, MS n , would be more fruitful as the ion trap geometry of the LTQ provided fragmentation an alysis to further elucidate compound identification. Additionally, the MS 2 capabilities afforded by the QE (due to the hybrid quad orbitrap geometry) could be utilized for fragmentation analysis, albeit to a lesser degree than offered by the ion trap; howe ver, no research using such will be presented here. Finally, DART/MS permitted direct analysis of samples without pretreatment or separation prior to mass analysis. This allowed analysis of samples in a more native en vironment, which is beneficial for more accurate and specific biomarkers for melanoma. High Performance Liquid Chromatography, High Resolution Mass Spectrometry First cell culture iteration Initially, mammalian cell cultures were grown and maintained by a third party , located in the Biomedical Sciences Department of the university . Four distinct cell lines were cultured: A 375 (female melanoma, FM), SK MEL 28 (male melanoma, MM), PCS 200 013 (adult melanocytes, AN), and PCS 200 012 (neonatal melanocytes, NN). FM and MM were maintained on EMEM, AN and NN were maintained on DCBM. Cell pellets from several passages were pooled together for each individual sample type (e.g., passages 22, 23, and 24 from MM), and the metabolites isolated for analysis by UHPLC/HRMS. Pooling samples prior to analysis r educed sample to sample (or biological) variance. Additionally, a fifth sample was created by pooling an equal

PAGE 185

185 All five samples were analyzed in triplicate by UHPLC/HRMS in positive mode, along with b lanks of each media and the internal standard , serving as a quality control for instrument performance . Upon collection, all data were then analyzed by PCA and PLS DA. Scores and loadings plots from PLS DA and PCA are displayed in Figure 4 6 and 4 7, respe ctively. This data provided conclusive visual proof that there is an undeniable distinction between the metabolites present in malignant melanoma cells vs normal melanocyte cells. Unfortunately, the foremost issue of variant media components being a primar y contribution to the separation still remained. Moreover, the data were only run in positive mode as there were not enough cell pellets collected to run replicates in both positive and negative mode, and the size of each pellet we were given was too small to afford protein quantification. Between lacking the ability to run the data under both polarities, questionable variance contributions, and normalization that could be improved upon, this initial set of data can be seen as confirmation that there exists compounds to be identified that produce drastic fold change differences between disease and control samples, but not as a useful tool for the identification itself. Second cell culture iteration Learning from the events of the first iteration, cells were now cultured by the author to maintain more control over the procedure and efficiency of their growth. The same four lines were initiated as before (FM and MM on EMEM, AN and NN on DCBM), however NN failed to grow in a stable and reproducible fashion, ther efore very little data created by combing the male and female melanoma lines at an approximate 1:1 ratio (v:v) after ten passages of each of the two lines. This was done to create a more broadly representative melanoma line by decreasing potential gender variance from the

PAGE 186

186 FM and MM lines, with the hope that it was be more demonstrative of generally defined growth characteristics than seen when the pellets were combined af ter growth, as before. More importantly, the melanoma lines were switched over to DCBM after 15 passages of stable growth on EMEM. Any statistical partitioning seen from these melanoma cells compared to normal melanocyte cells should originate strictly fro m cellular variation rather than differences in the media. Again, cell pellets from several passages were pooled together for each individual sample type (e.g., passages 22, 23, and 24 for SK MEL 28) to minimize sample to sample variance. All samples were analyzed in triplicate by UHPLC/HRMS with polarity switching mode (both positive and negative spectra collected consecutively), along with blanks of each media and the internal standard. Isolated metabolites from cell pellets of the melanoma lines grown on each media were both included in analysis. Upon collection, all data were then analyzed by PCA and PLS DA, as well as ANOVA, HCA, RdF, and t test, all through MetaboAnalyst . 131 The most pertinent conclusion was once again seeing first component separation of melanoma and normal samples. Both PCA and PLS DA produced similar scores plots revealing the greates t variance being between diseases and control (Figure 4 8). The second component of variance seemed to indicate a distinction between male and female melanoma cells, FM and MM. Furthermore, even though HM started with 50:50 FM to MM, the FM portion of the HM line proliferated to a greater extent than the MM line causing the PCA and PLS DA plots to reveal a separation between HM/FM and MM. The female line, A 375, was described as having an innately higher cell viability and confluence than the male line, SK MEL 28; hence, this gradual shift to a higher

PAGE 187

187 percentage of female cells in HM is not surprising. Even so, this statistical difference produced by multivariate analysis may not be as significant as it is perceived. There was a single male and a single fema le line, and while the cell lines are supposedly far removed from having human bias, there still may be individual predisposition. Therefore, this separation may be due more to variation associated from one human to another rather than solely from one sex to another. Further research analyzing more than one line of each gender would be required to verify. Besides PCA and PLS DA, other statistical techniques were performed on the data as well. These methods include: ANOVA and t test, statistical models whic h analyze the variance between two or more sample groups; 136 hierarchical clustering a nalysis, a strategy for grouping sets of similar features into clusters and arranging them in a hierarchy based on these observations; 137 and random f orest, an ensemble learning method that operates by constructing a multitude of decision trees and outpu tting a plot based on the mean predic tion regression. 138 Given that there several distinct methods, each with an intricate set of calculations and theories, the des criptions will remain brief. For more detailed explanations and depictions, please consult the referenced works. 136 138 Many of these statistical methods are use d in a complimentary fashion. For example, the models used for ANOVA/t test and RdF can first be used to produce distinction between sample groupings, whereby HCA could then utilize this modeled data to yield a broad visual representation of the fold chang es observed between groupings , as well as provide an idea of which metabolites may be related due to their branching . Such examples are exhibited in Figures 4 9 and 4 10. Likewise, there are

PAGE 188

188 many other plots from mean decreased accuracy (Figure 4 11) , to V IP scores (Figure 4 12) , to fold change (or volcano) which utilize these and other statistical models with the intention of producing the best representations of sample separation. These plots can then be further mined for data features, like mass and rete ntion time, to identify compounds of interest that differ between sample groupings. This will be discussed in more detail later. Chromatography column improvements Considerable time was spent optimizing the instrumental parameters at which the LC was condu cted (e.g., MS temperatures and voltages, LC run time per sample, LC solvent gradient), but will not be detailed further. However, efficiency of the column used will be quickly reviewed. This research utilized a C18 PFP column consisting of a typical C18 s tationary phase bonded to a PFP layer. The most prominent column in LC analysis today is C18, due to a combination of its hydrophobic, stable, and low bleed characteristics. 139 The PFP bonded phase affords an alternate selectivity to improve separations that may prove problematic on simple C18 phase columns, while maintaining similar hydr ophobic qualities. Indeed, this column is a workhorse and provides stellar separation for most sample types , and proved adequate for the mammalian cell cultures analyzed here. Yet, insufficient column retention was observed for many of the compounds that w ere extracted by mult ivariate statistics (Figure 4 13 ). Most of the analytes of interest eluted within the first few minutes of the 22 minutes separation. This would suggest that a majority of the pertinent compounds are polar in nature. Therefore, a hydro philic column that retains polar compounds more efficiently could produce significantly improved separation of these more essential compounds.

PAGE 189

189 Atmospheric Pressure Chemical Ionization, Tandem Mass Spectrometry Only the first iteration of cell cultures were run by APCI/MS. As mentioned, the four cell lines (FM, MM, AN, NN) were initially grown on two different media, depending on whether they were malignant melanoma or normal melanocytes. APCI/MS was performed on these samples to assess the ability to analyz e complex biological samples by ambient ionization without prior separation. Moreover, MS n was utilized with the desire of using fragmentation analysis to expand identification of potential biomarkers beyond molecular feature classification. As a consequen ce of growing the different cell types on different media, a major contributing factor in the axis of greatest variance amongst sample groupings was this difference in media components. This discrepancy, as well as others noted above, tends to dampen any s ignificant impact that could be ascertained from the data produced by these samples. However, this exercise was not completely futile, as it indeed demonstrated the usefulness of tandem MS and the quick analysis only possible without chromatography. The fi rst merit of note was the rapid time of analysis achieved by APCI/MS. UHPLC is rooted in separation efficiency, and the greater the efficiency, the longer the run time . A compromise is often found between run time and effective peak resolution , but inheren tly it can only be cut so short. That is to say, LC will always limit sample throughput because it will never be achievable on a timescale similar to that of MS, i.e., seconds as opposed to minutes or hours. In this work, the LC run time of a single sample was 22 minutes. Running every samples once, let alone in triplicate, takes over an hour, and that does not account for blanks and quality control measurements. Compare that to MS with no chromatographic separation where samples could be

PAGE 190

190 analyzed for as sh ort of a period of time as necessary to acquire the desired data, sometimes down to one minute. The quickness with which samples could be analyzed was not the only advantage. In fact, rapid sampling was just one benefit of the larger concept of improved ad aptability. LC runs are set ahead of time, and even without an autosampler to conduct the entire set of runs, a researcher must wait until the end of the LC/MS run in order to evaluate its success. Conversely, MS runs can be monitored in real time without chromatography, hence, if a problem arises, it can be swiftly dealt with. On the same token, even if there is no problem, parameters can be optimized and altered at will. This leads to the final advantage MS has over LC/MS, unconstrained tandem MS capabili ties. Tandem MS can be performed during an LC/MS run, but the window of opportunity is incredibly short (depending on the peak width of the compound), and you have to wait until that exact point in the run every time to execute it. Without chromatography, tandem MS can be performed ad infinitum, on any detectable mass, to any degree the instrument allows. This allows the most in depth fragmentation analysis. While many runs were conducted, see Figure 4 14 for PCA and PLS DA scores plots, their inevitable us efulness in creation of the table of potential biomarkers was essentially nonexistent due to the inefficient samples that were run. The main use in this work was for proof of concept of rapid MS n analysis; an example of w hich can be found in Figure 4 15 . F irstly, the number of viable peaks could be reduced by determining if some peaks are fragments or adducts of others, meaning they come from the same compound. Secondly, a compound identified by other means (i.e., accurate mass ,

PAGE 191

191 isotopic pattern, r.t. ) coul d be validated by the fragmentation patterns seen upon execution of several stages of MS (MS 2 , MS 3 , etc. ). Lastly, and arguably the most important capability is compound identification by fragmentation analysis. Similar to validating an identified compound , MS n could provide instrumental identification where HRMS comes up short. In many cases, feature extraction based on accurate mass will produce a list of several compounds, all of which could debatably pertain to the chosen m/z , with no foreseeable way to narrow it down any further. MS n analysis could provide definitive justification for choosing one compound over the rest based on the predictable fragmentation pattern of certain species. For example, molecular feature extraction helped to determine that m /z 260.1606 is produced by a tripeptide. Yet, this single tripeptide could be composed of any combination of a handful of amino acids. If MS n were to be performed on this m/z , elucidation of which specific amino acids are present could be achievable, and f urther probing of the data may well lead to the exact structure. Direct Analysis in Real Time, High Resolution Mass Spectrometry First cell culture iteration The aforementioned set of cell cultures that were grown outside the research group of the author, and maintained on separate media were also analyzed by DART/HRMS. Analysis comprised pooled samples of the isolated metabolite solution from FM, MM, XM (grown on EMEM), AN, and NN (grown on DCBM), in addition to pooled samples of the precipitated protein l ysate from XM. All samples were ionized with no pre treatment or pre concen tration of any kind. Figure 4 16 displays PCA and PLS DA scores plots of this cell culture data. Once again, first component separation of normal and melanoma samples is observed.

PAGE 192

192 A s before, the data presented in this section should be seen as proof of concept, as all statistical analyses performed on it are thrown into question since the greatest variance is most likely the result of different media components. Nevertheless, analysi s of this first culture iteration proved valuable in helping direct future analyses by allowing instrumental optimization, and providing evidence that direct analysis of both the isolated metabolite solution and the precipitated lysed cellular protein was achievable. Biopsied human tissue Six biopsied human skin tissue samples were analyzed by DART/HRMS. All sampled were ionized with no pre treatment or pre concentration of any kind. Three samples were normal tissue and three samples were melanoma lesions. Of these six, two came from the same individual, one melanoma lesion and one sample of adjacent skin tissue. The other four were all biopsied from separate people. PLS DA scores and loadings pl ots are displayed in Figure 4 17 , and reveal first component se paration of normal and melanoma samples. Only one female sample was analyzed, the rest were obtained from males, thus accurate separation of specimens by gender is not possible at this time, yet some distinction was achieved. The three purple circles near the first component line lying to the far left along the second component (C1: 0, C2: 40) correspond to the female tissue sample, and are nearly separated from all of the male tissue samples. Future analysis of a greater number female tissue samples would allow definitive verification of gender separation. PCA did not provide component separation of tissue samples as it did for cell cultures. However, this does not detract from the results observed by PLS DA. Since this research is still in the classificat ion phase, PLS DA delivers a more accurate representation of the distinct groupings, whereas PCA is

PAGE 193

193 best for validation of class separation. Achieving separation by PCA is an eventual goal, but of little concern at this s tage. It should be noted that PCA ( and PLS DA) separation was achieved if analysis was limited to the adjacent melanoma and normal tissue samples collected from the same individual. While the sample size is only one, it represents an important discovery as there is clear separation between cancer and control while presumably avoiding differences in gender, age, and exogenous sources (e.g., lotion, fragrance, diet) since they both come from a single person where all of those factors would be the same for both samples types. I ncreasing the tot al number of tissue specimens to much greater than six , and adjacent melanoma/normal specimens to greater than one individual, would provide a more definitive probe into biomarkers identification. It may even reveal PCA component separation with an n > 1, since the variables associated with melanoma vs normal samples would become more apparent as individual differences are weeded out in larger groups. Furthermore, it would provide a s tatistical model with more representat ive information of the population at large. Besides PCA and PLS DA, other statistical techniques, including ANOVA/t test, HCA, and RdF, were performed on the data as well. These plots were further mined for data features to identify compounds of interest that differ between sample groupings. This will be discussed in more detail later. This particular section is of tremendous importance in this work, and best represents the ultimate goal of this research, direct sampling. Direct ionization off a ounds detected toward those of higher

PAGE 194

194 volatility, as they are most easily ionized. A compound that can be more easily ionized requires less energy to eject/desorb from a surface, thus, higher volatility means a greater probability for application in noninv asive screenings. The results here demonstrate the utility of direct sampling methods and their ability to observe differences in diseased vs control samples with no sample pre treatment. Compound Identification PLS DA, PCA, and several other statistical m ethods were conducted on normal melanocyte cells and melanoma cells. Multivariate analysis is used primarily for feature reduction to allow potentially pertinent data to be mined from the thousands of metabolic features present in a single sample. 51 , 52 In brief, to conduct PLS DA, features ( m/z , r.t., peak area, etc.) from every individual sample are all compared to one another. Those features that rev eal the greatest variance from one sample to the next are assigned to a principal component (PC). Only a handful of principal components are typically established, meaning the data is reduced from thousands of features to a manageable number of components. Then, two PCs are plotted as a function of one another producing a 2 dimensional representation of the greatest variance between the samples, called a scores plot. If class separation is achieve d (i.e., melanoma samples are distinct ly separated from norma l samples ), there is a very probable significant difference in the data sets. In order to determine what specific features are responsible for this variation, several techniques are employed in tandem: ANOVA/t testing, HCA, and RdF. In addition, a loading s plot is created directly from the scores plot information, if the data permits. This is done by calculating the standard deviation of all of the samples at each m/z , and multiplying this value by the PC score determined for that particular m/z value.

PAGE 195

195 The se new values will be either positive or negative, and vary in their intensity. The more intense the loading value, the greater the variation between samples given by that specific m/ z. Based on the scores plot, it can be seen which polarity (positive or n egative) accounts for each sample type (normal or melanoma). For example, if m/ z 166 had a much greater signal intensity produced in the MS of the melanoma cells vs the normal cells, the loadings plot would reflect this by revealing a high loadings intensi ty at m/z 166 in the direction of melanoma. Therefore, if melanoma is negatively loaded, the large peak at 166 will be greatly down loaded. See again Figure 4 6 for examples of these scores and loadings plots. However, a loadings plot is typically biased t oward features of large variation between samples, not necessarily those of greatest biological significance. This is where use of more than one technique comes in handy. Every method has an intrinsic bias, but each is prejudiced toward certain features fo r different reasons based off how the analysis is calculated. Therefore, using numerous statistical methods on the same data will provide the best chance of identifying biomarkers that will have a large biomedical impact. PCA is performed in a similar fas hion to PLS DA, the only difference is the way the algorithm governing the method is run. PLS DA is a supervised method while PCA is unsupervised. 51 , 52 Basically, when running PLS DA, sample classes are given a bias based on how they are input into the program. For example, if all of the melanoma cells normal melanocyte cells are entered under between the classes is prioritized (e.g., melanoma vs. normal) . However, when

PAGE 196

196 conducting PCA, each sample is seen as an individual entity, so the greatest variance between all samples is p rioritized (e.g., female melanoma sample x vs male melanoma y vs. adult melanocyte sample z vs. etc.) . Neither method can objectively be considered better than the other as they each have merit depending on the current situation. In this research PLS DA pr oved most useful in identifying new putative biomarkers as it inherently pull s out variation between melanoma and normal samples. However, PCA could still be utilized to great benefit. If class separation by PCA was achieved, that proves there was a more o bjective difference between samples from those classes. Hence, both methods were used. High resolution mass spectral data provided the capability to perform compound elucidation based on accurate mass, as well as the retention time and compound polarity wh en using LC. See Figure 4 1 8 for a visual workflow of the steps necessary to perform identification of putative compounds. First, the multivariate analysis was performed on the accurate mass data, producing g raphical diff erences in sample variation . This a llowed selection of specific accurate mass values from the mass spectrum of each sample. An online database (METLIN) 135 was used to input the m/z values allowing generation of a list of potential compounds that might correlate to the searched for accurate mass. Based on ppm error, biofunction of the compound, and r.t. and polarity if applicable, 132 134 compound identities were selected to best represent potential biomarkers for melanoma (Table 4 5) . Comprehensive metabolomic studies (or studies into any biological class, i.e., lipidomics, proteomics, etc.) typically involves a collaborative efforts between several groups of scientists mining data for months, if not years. This data has only been probed by a single person over the course of a couple

PAGE 197

197 months. Therefore, this table is far from complete as much of the data has yet to be explored fully. However, it does represent a major achievement, and the primary goal of this work, which was to begin production of a table of putative markers. Selection of Putative Biomarkers As mentioned, there are several reasons a c ompound may be selected over others of equal potential. Select compounds from the table will be discussed in more detail here. To start, the most recognizable finding is of m/z 166.0863, pertaining to phenylalanin e. It is well established that there is an observed increase in phenylalanine concentration with the presence of melanoma tumors. 134 This is interesting since p henylalanine is an essential amino acid, i.e., it cannot be synthesized by the typical metabolic processes of the human body , hence, an increase in its abundance must be linked to a change in diet or alterations in the metabolic pathway . Moreover, phenylal anine is a precursor to melanin, the skin darkening pigment, th us providing an even stronger correlation to melanoma pathology. Lastly, the mass error (ppm) is essentially zero, making phenylalanine an ideal choi ce as a biomarker for melanoma. Another note worthy discovery was the potential relation of several chosen markers through related metabolic pathways. Choline ( m/z 104.1070) is an essential nutrient, which can be oxidized to form betaine ( m/z 118.0864), itself a methyl donating amino acid vital to th e production of methionine ( m/z 150.0586). 134 , 140 All three of these compounds demonstrate a significant i ncrease in abundance in melanoma samples compared to normal samples. C onversely, there is a decrease in the observed level of creatine ( m/z 132.0770) present in melanoma samples compared to normal . Creatine is synthesized, in part, by methionine. If methio nine is prevented from reacting, as assumed by its relatively higher abundance, this could cause a reduction in creatine

PAGE 198

198 levels. It stands to reason that such an integral metabolic relationship would advocate a parallel increase or decrease in the concentr ation s of each species in the presence of melanoma. If the amount of one metabolite is altered, a subsequent shift in the concentrations of the ensuing products would be foreseeable. Similarly, two more markers, tryptophan ( m/z 205.0969) and aminomuconic a cid ( m/z 175.0713), were established through a related degradation pathway . Amino muconic acid is a major breakdown product of tryptophan 134 and demonstrates a s ignificantly lower abundance in melanoma vs normal samples, while the opposite is observed for tryptophan . This may sug gest that melanoma disrupts the degradation pathway causing tryptophan (or another metabolite involved in the degradation) to be stored i n higher abundance, consequently decreasing the amount of aminomuconic acid produced. Finally, there were dozens (if not more) dipeptides and tripeptides that were, thus far, only narrowed down to one of a few possibilities. In these cases, the identities of the two or three amino acids that make up the peptide chain is limited to three or four acids, bringing complete identification very close, but not quite achievable yet. In these instances, MS n would be instrumental in determining the definitive makeup of the peptides, and might even shed light on the sequence. Alternatively, a standard of each of the possible peptides could be run by the same analytical method and matched to the chromatographic peak seen for this compound. In either case, a validative a nalysis is required. For example, m/z 231.1702 is presumably a dipeptide made up of valine and either leucine or isoleucine. However, the order of the two acids, and which isomer of leucine is present , is unknown since the masses are identical. As amino ac ids are the

PAGE 199

199 major building blocks of all biochemical molecules and reactions, changes in the abundance of di , tri , and higher order polypeptides will likely play a key role in filling out a large portion of any comprehensive table of biomarkers that is i nevitably established. Therefore, identification and validation of these seemingly generic small peptide chains will prove vital in the long run. Conclusion This research is the first presentation utilizing a variety of ambient ionization techniques to pro duce a list of putative biomarkers for melanoma by mass spectrometry. It is also the first instance of using DART/MS to analyze human tissue samples. Three distinct instrumental methodologies were applied for analysis of mammalian cell cultures as well as human tissue biopsies, comprising both normal and melanoma samples. The use of multivariate data analysis revealed a significant difference in abundance between normal and melanoma afflicted samples. LC/MS was utilized to validate the other, more novel met hods. More importantly, it was incredibly useful for global metabolomics, providing broad insight into the differences between melanoma and normal samples. Accurate mass data affords compound identification, while coupling to chromatography improves identi fication accuracy due to the added features of r.t. and polarity. This distinction between sample groupings was further revealed by multivariate data analysis. PLS DA more often showcased better separation between melanoma and normal samples, helping to de termine more accurate biomarkers. However, PCA frequently resulted in first component separation of melanoma vs normal samples. This is important to note since PCA is an unsupervised method, thus it can be confidently stated that there is definitive eviden ce for a distinct difference in the volatile metabolome of melanoma samples vs

PAGE 200

200 normal. Moreover, clustering analysis revealed branching amongst melanoma and normal samples, and ANOVA, t test, and RdF modeling all produced significant fold change difference s between the diseased and control groups. Features that were difficult to distinguish with LC/MS were targeted with APCI/MS n for analysis of their specific fragmentation patterns. By comparison to databases of known fragmentation spectra, more potential m arkers were identified. Moreover, previously recognized compounds could also be verified by their characteristic fragmentation. Generally, tandem MS is ideal for targeted metabolomics. Finally, DART/HRMS afforded direct analysis of samples. All samples, no matter the type (i.e., isolate solution, lysate precipitate, tissue), were analyzed without pretreatment or preconcentration of any kind, which is impossible with the other two methods. Direct sampling off the surface of a tissue inherently prejudices the compounds identified toward those of higher volatility, as they are most easily ionized. The greater the volatility of the compound, the greater the potential for application in noninvasive screenings. This is because a compound that can be more easily io nized requires less energy to eject/desorb from a surface (i.e., live human tissue), therefore allowing creation of a novel method for analysis directly off of skin. Overall, a table of prospective biomarkers for melanoma was created based on the data coll ected and analyzed in this work (Table 4 5). The use of several multivariate data analysis techniques, justification according to numerous features ( m/z , r.t., etc.), and verification by several different instrumental methodologies should provide adequate reason to believe these compounds are of significance. The data collected is still a gold mine of potential data yet to be mined, as Table 4 5 represents a small

PAGE 201

201 fraction of the markers identified from ions detected in positive ion mode. Furthermore, this does not take into account the negative polarity data collected, which itself could hold even more prospective markers. Future studies exploring potential biomarkers for melanoma should increase the number of human samples investigated, both cell cultures and tissue. Pooling more samples will allow further verification (or demotion) of the prospective markers found here, as well as possible identification of even more. Lastly, LC analysis with a hydrophilic column could produce more efficient retention and baseline separation of the polar compounds that seem to dominate the multivariate statics.

PAGE 202

202 Figure 4 1. Normal vs cancer cell division. Normal cells divide at a controlled rate and when one cell becomes damaged the cell dies (top). Cancer cells involv e the uncontrolled proliferation of mutated or damaged cells (bottom) Adapted from the National Cancer Institute . 109

PAGE 203

203 Figure 4 2. Stage progression of melanoma as it originates in melanocytes, spreads into deeper layers of the skin, then metastasizes.

PAGE 204

204 Figure 4 3. Survival rates after melanoma d iagnosis. Percentage of patients who live at least 5 (orange) and 10 (grey) years after diagnosis by stage at which cancer was detected. There is a significant decrease in chance of survival the later melanoma is detected. Stage IIIA has a higher survival rate than some stage II cancers, this may result from less advanced primary tumors often seen with IIIA cancers, how ever this remains uncertain. 108

PAGE 205

205 Figure 4 4. Visual distinction and interpretation between benig n moles and malignant melanoma adapted and reprinted with permission from Americ an Academy of Dermatology . 111

PAGE 206

206 Figure 4 5. llustrates that metabolomics involves the study of low molecular weight species (metabolites), the downstream products of all of the other omics areas, and as such are the closet to the phenotype, and therefore the function of biological molecules.

PAGE 207

207 Table 4 1. MS and chromatographic parameters for QE experimentation. MS Parameters Chromatographic Parameters Chromatographic Gradient Scan Range ( m/z ) 70 100 0 Column ACE Excel 2 C18 PFP 100 x 2.1 mm Time (min.) Solvent A (%) 0.1 % FA/H 2 O Solvent B (%) ACN Injection Volume (µL) 5 Pump Thermo Dionex UltiMate 3000 0 100 0 Flow Rate (µL/min) 350 Column Temperature 35 °C 1 100 0 Sheath Gas (a.u.) 50 Injec tion Volume 5 µL 11 35 65 Auxiliary Gas (a.u.) 10 Flow Rate 350 µL/min 13 35 65 Spray Voltage (kV) 3.5 Solvent A 0.1% formic acid/water 18 5 95 Capillary Temperature (°C) 300 Solvent B acetonitrile 20 5 95 ESI Heater Temperature (°C) 350 21 100 0 Split Lens (V) 40 22 100 0 Resolution 35,000 Polarity pos/neg switching

PAGE 208

208 Table 4 2. MS parameters for LTQ experimentation on cell cultures. Parameter Set Value Scan Range ( m/z ) 50 1000 Injection Volume (µL) continuous Flow Rate (µL/min) 10 Sheath Gas (aux units) 20 Auxiliary Gas (aux units) n/a Spray Current (µA) 4.5 Capillary Temperature (°C) 225 Vaporizer Temperature (°C) 245 Tube Lens 70 Resolution unit Polarity positive

PAGE 209

209 Table 4 3. MS parameter s for TOF experimentation. Parameter Set Value Scan Range ( m/z ) 70 1000 Injection Volume (µL) variable Gas Flow Rate (L/min) 3.5 DART Gas Stream (°C) 200 500 Capillary Temperature (°C) 300 Skimmer Voltage (V) 70 Resolution 12,000 Polarity posit ive

PAGE 210

210 Table 4 4. Comparison of instrumental methodologies utilized for the metabolomics studies. Ion Trap Mass Spectrometry Liquid Chromatography High Resolution Mass Spectrometry Direct Analysis in Real Time High Resolution Mass Spectrometry Therm o Scientific LTQ XL Thermo Scientific Q Exactive Agilent 6220 TOF Targeted metabolomics Global metabolomics Global Metabolomics Tandem MS (MS n ) Accurate Mass Accurate Mass Quick time of analysis Pre MS separation Direct sampling

PAGE 211

211 Figur e 4 6. PLS DA scores (A) and loading (B) plots from UHPLC/HRMS of cell cultures displaying first component separation of normal melanocyte cells vs. melanoma cells, and second component separation of female vs. male cells. Masses of greater significance i n normal cells are positively loaded (green), while masses of greater significance in melanoma cells are negatively loaded (purple).

PAGE 212

212 Figure 4 7. PCA scores (A) and loading (B) plots from UHPLC/HRMS of cell cultures displaying first component sep aration of normal melanocyte cells vs. melanoma cells, and second component separation of female vs. male cells. Masses of greater significance in normal cells are positively loaded (green), while masses of greater significance in melanoma cells are negati vely loaded (purple).

PAGE 213

213 Figure 4 8. PCA and PLS DA scores plots from UHPLC/HRMS of cell cultures displaying first component separation of normal melanocyte cells vs. melanoma cells, and second component separation of female vs. male cells.

PAGE 214

214 Fi gure 4 9. HCA from UHPLC/HRMS analysis of cell cultures using ANOVA/t test statistics reveals branching amongst two groups, normal and melanoma. The top 50 features were ranked to retain the most contrasting pattern. Further branching is subsequently defi ned between male, female, and hybrid melanoma cells.

PAGE 215

215 Figure 4 10. HCA from UHPLC/HRMS analysis of cell cultures using Random Forest modeling reveals branching amongst two groups, normal and melanoma. The top 50 features were ranked to reta in the most contrasting pattern. Further branching is subsequently defined between male, female, and hybrid melanoma cells.

PAGE 216

216 Figure 4 11. Mean decreased accuracy (MDA) plot for top 15 m/z features that contribute most to the RdF training se t, from UHPLC/HRMS analysis of cell cultures. The mean of each samples contribution to the data model is calculated and then one sample is removed and the mean re calculated to obtain a MDA value. The greater the contribution of that value to the mean, the greater the MDA value. Using Random Forest modeling reveals branching amongst two groups, normal and melanoma. The columns to the right of each plot indicate whether a given feature is up or down regulated in melanoma (M) vs normal (N) samples .

PAGE 217

217 Figure 4 12. Plot of top 25 m/z features with the greatest variable information processing (VIP) scores by PLS DA, from UHPLC/HRMS analysis of cell cultures. The greater the VIP score, the greater the contribution of that feature to the variance associated with that particular PLS DA component. The columns to the right of each plot indicate whether a given feature is up or down regulated in melanoma (M) vs normal (N) samples.

PAGE 218

218 Figure 4 13 . TIC from UHPLC/HRMS analysis of AN, and comparison of select r.t. (1 5.5 and 9.5 11 min) of same TIC and TIC from UHPLC/HRMS analysis of XM (inset). Notable differences between normal and melanoma samples is observed around 1.9 2.3, 3.9, 4.6 4.7, and 10.6 10.8.

PAGE 219

219 Figure 4 14 . PC A and PLS DA scores plots from APCI/MS of cell cultures displaying first component separation of normal melanocyte cells vs. melanoma cells.

PAGE 220

220 Figure 4 15 . PLS DA loadings plot along C1 from APCI/MS of cell cultures, masses of greater significanc e in normal cells are positively loaded (green) and masses of greater significance in melanoma cells are negatively loaded (purple). Insets display MS 2 spectra of m/z 520 and 369.

PAGE 221

221 Figure 4 16 . PCA and PLS DA scores plots from DART/HRMS of cell cultur es displaying first component separation of normal melanocyte cells vs. melanoma cells, and moderate separation of female vs. male cells along the second component.

PAGE 222

222 Figure 4 17 . PLS DA scores (A) and loading (B) plots from DART/HRMS of biopsied tissue specimens displaying first component separation of normal vs. melanoma samples. Masses of greater significance in normal tissue are positively loaded (green), while masses of greater significance in melanoma tissue are negatively loaded (purple).

PAGE 223

223 Figure 4 18 . Workflow for compound identification. First, single m/z deemed significantly different between sample class types by statistical analysis. Next, search m/z in metabolite database. Last, choose potential compound ID from database base d on multiple criteria, e.g., ppm, r.t., pola rity, biofunction .

PAGE 224

224 Table 4 5. Potential biomarkers for m elanoma. m/z (± ppm) Fold Change (M elanoma /N ormal ) Ion Proposed Formula Prospective Marker 86.0964 (± 0.00) 2.40 [M+NH 4 ] + C 5 H 8 Isoprene 104.1 070 (± 0.00) 2.31 [M+H] + C 5 H 13 NO Choline 118.0862 ( 0.85) 2.48 [M+H] + C 5 H 11 NO 2 Betaine 150.0586 (+ 1.99) 4.68 [M+H] + C 5 H 11 NO 2 S Methionine 166.0863 (± 0.00) 2.93 [M+H] + C 9 H 11 NO 2 Phenylalanine 205.0969 ( 1.46) 2.08 [M+H] + C 11 H 12 N 2 O 2 Tryptophan 23 1.1702 ( 0.43) 2.16 [M+H] + C 11 H 22 N 2 O 3 Dipeptide (V + I or L) 175.0713 (± 0.00) 0.37 [M+NH 4 ] + C 6 H 7 NO 4 Aminomuconic acid 194.0422 ( 1.03) 0.32 [M+Na] + C 7 H 9 NO 4 Tetrahydrodipicolinate 251.0699 (+ 1.19) 0.50 [M+NH 4 ] + C 8 H 11 NO 5 S Dopamine 3 (or 4 ) sulfate 348.0706 (+ 0.59) 0.49 [M+H] + C 10 H 14 N 5 O 7 P Adenosine monophosphate

PAGE 225

225 CHAPTER 5 CONCLUSIONS AND FUTURE WORK Summary This dissertation has presented three distinct projects utilizing mass spectrometry (MS), field asymmetric ion mobility spectrometry (FAIMS ), and ultraviolet spectrometry (UV), representing a wide range of biomedical and clinical applications. This includes modification of solvent systems for optimal solubility and stability of capsaicin, development of novel methodologies for analysis of hum an breath by FAIMS/MS and standalone FAIMS, as well as metabolomics analysis of melanoma for discovery and characterization of potential biomarkers. The first chapter gave a brief look into the principles and underlying mechanisms behind the techniques pre sented in the body of this work. Chapter 2 detailed the first application of one of these methods. UV was utilized to quantify capsaicin concentrations prepared in varying solvent systems. This chapter provided an in depth study into the solubility and sta bility of capsaicin containing solutions. Furthermore, this work represents a step toward creation of a standardized method for the preparation of capsaicin solutions to be used in tussigenic challenges. As of yet, no such method has been adopted by clinic ians. Therefore, developing a step by step procedure for creating capsaicin solutions with optimal solubility, while remaining safe for human consumption, stable for an extended shelf life, and lacking a foul taste was crucial to establishing a universal m ethod. In the next chapter, human breath and simulated breath were analyzed by MS and FAIMS/MS for characterization of volatile exogenous and endogenous species. Additionally, the utility of standalone FAIMS as a viable technique for real time breath

PAGE 226

226 detec tion was explored. In order to perform analysis of human breath, introduction and collection methods were developed for analysis by APCI/MS, EESI/MS, FAIMS/MS, and standalone FAIMS. Moreover, certain samples could not be analyzed directly in human subjects (i.e., THC), thus a breath simulation system was created and implemented for analysis of these compounds. All compounds were first characterized by MS to validate the breath sampling methods. Once established, the flavorant compounds were detected on the standalone FAIMS device, while THC was detected by FAIMS/MS. A one way mouthpiece was used to introduce the breath samples to the standalone device. Conversely, a breath simulation system utilizing nitrogen bubbled through solvent spiked with analyte was u sed to introduce THC to a home built FAIMS cell. An extractive electrospray ionization (EESI) setup was developed for FAIMS/MS analysis. THC was detected at low concentrations by direct injection (ESI/FAIMS/MS) as well as in simulated breath by EESI/FAIMS/ MS. Real time, direct breath detection of exogenous volatile flavorants was demonstrated by standalone FAIMS. Similarly, THC in solution and simulated breath was detected on line and in real time. Detection of breath by standalone FAIMS and ionization by EESI coupled to FAIMS/MS both established novel methodologies never demonstrated prior to those studies being conducted. Finally, Chapter 4 involved a comprehensive metabolomics study with the aim of identifying and validating biomarkers of melanoma. Prese nted here was the first work to utilize a variety of ambient ionization techniques to produce a list of putative biomarkers for melanoma by mass spectrometry. Moreover, it is the first example of using

PAGE 227

227 DART/MS to analyze human tissue samples. Overall, t hre e distinct instrumental methodologies were employed for analysis of mammalian cell cultures a s well as human tissue biopsies. M ultivariate data analysis revealed a significant difference in the abundance of many species between normal and melanoma afflicte d samples. Initially , ultra high performance liquid chromatography (UHPLC) high resolution mass spectrometry (HRMS) provided validation of methodologies as it is the current gold standard for metabolomics analysis. Moreover, LC/HRMS allowed untargeted glob al analysis of cell cultures. Complimentary to LC/HRMS, the cell cultures were also analyzed by APCI/MS n and DART/HRMS. Using several disparate techniques permitted confirmation of certain compounds that produced spectral peaks by more than one method, as well as identification of supplemental markers unique to a particular method that would have otherwise gone unnoticed if a singular method was chosen. This research also moved beyond cell cultures to analyze biopsied human tissue samples by DART/HRMS. Anal yses were performed with no pretreatment or preconcentration of the tissue. Multivariate data analysis was performed on all of the data collected (i.e., PCA, PLS DA, ANOVA, Hierarchical Clustering Analysis, Random Forest, t test). The accurate mass spectra allowed molecular feature extraction, and subsequent compound identification when run through several popular online metabolomics databases (i.e., HMDb, METLIN). Tandem MS data helped to further verify uncertain identities. From this, a table of prospecti ve biomarkers for melanoma was established. Future Work At its core, science has always been about striving for new and better answers, and looking forward to what can be accomplished in the future while reminding oneself

PAGE 228

228 of the hurdles and accomplishments of the past. Therefore, while each project in this manuscript has been brought to a relative conclusion, there is plenty more to be discovered. In this vain, the following discussion provides insight into the potential directions the projects detailed in this work might go. However, this should be seen more as a gentle guide than a permanent roadmap. Tussigenic challenges have been used to assess cough threshold for decades, yet standard methods for preparation of the capsaicin containing solutions vital t o the test are still missing from the literature. Several attempts have been made, 59 , 60 but the use of certain solvents, namely surfactants like Tween, has brought about new issues of discomfort and taste. Chapter 2 of this document outlined a procedure to create solutions of capsaicin in a reproducible manner that have an adequate shelf life, but this was only laying the groundwork. The solvent system should be refined further to create a solution that continues to impr ove in solubility and stability, as well as taste. A resurgence in analysis of breath has taken place in the last decade. It has been known for a long time that exhaled breath held diagnostic implications, however technology was never sufficient enough to allow adequate sensitivity and selectivity of breath samples for such a technique to become viable, until recently that is. One particular method vying for attention in the field of breath research is FAIMS; many of the advantages to this technique are det ailed in Chapter 3. There is discussed work using standalone FAIMS and FAIMS/MS to analyze breath samples. However, the FAIMS devices were vastly different, thus the data observed on the standalone device is not comparable to that demonstrated by FAIMS/MS. Therefore, the foremost goal should be the development of a FAIMS device that can couple to MS for peak

PAGE 229

229 identification, while subsequently allowing detachment and analysis as a standalone system. This would afford clinical and field use of FAIMS, while pr oviding MS validation in the lab whenever more extensive analysis is required. Other issues only touched on above concern the types of compounds in breath that may be analyzed, the geometry of the FAIMS cell, and the application of the FAIMS waveform itsel f. It would take an entire manuscript alone to debate the implications and necessities of every individual analyte that might be of interest for different applications of breath analysis, hence a broad overview will be given here. In general, analysis of b reath could easily revolutionize two fields: medical and law enforcement. Medically speaking, diseases that are already associated with human exhaled breath and the respiratory system would be the easiest look into. Cancers of the lung and breast, as well as ailments like asthma and chronic obstructive pulmonary disease would be obvious choices; diabetes would be an interesting alternative since its recognition in the breath has been around for millennia, yet no definitive breath test has ever been develop ed. As for law enforcement, legalization of drugs that are smoked or inhaled, such as marijuana, will quickly become an issue for police as they will soon require a roadside test for assessment of intoxication. Due to the blood breath interface in the lung s, any bloodborne marker for any illicit or prescription drugs should be detectable in the breath, therefore a device will need to be developed to monitor these compounds in a portable manner; hence the use of FAIMS analysis. Along those lines, the geometr y of the FAIMS cell will dictate the ability the FAIMS device would have to analyze a certain type of sample. Indeed, planar FAIMS affords the greatest selectivity, but sensitivity and transmission will be the tallest hurdle to overcome in breath analysis due to the trace

PAGE 230

230 quantities of analytes and variable nature of breath samples. Hence, hemispherical FAIMS will probably be the ideal geometry, as its curvature allows the most efficient air flow over its surface (relative to flat plates in planar and moder ate curvature in cylindrical), and its focusing of ions in all directions. Moreover, if selectively is an issue for hemispherical geometries, modification of the waveform could alleviate this setback. One of the easiest and most efficient ways to create a portable yet selective FAIMS device would be to use a square waveform. Preliminary experiments into square wave FAIMS on a miniature device indicate substantial increases in resolution as a result if the inherent shape of the waveform as well as the abilit y to alter the duty cycle. Thus, combining square wave FAIMS, with a portable device, monitoring select analytes of interest could allow development of a field ready and clinically amenable device capable of detecting biomedically and road safety relevant compounds. Finally, Chapter 4 entails metabolomics of melanoma. Described there are experiments revealing significant distinction between melanoma and normal samples comprised of cell cultures and human tissue. While analysis directly off tissue was achiev ed, the methodology was far from optimized. Furthermore, future studies may want to focus efforts on analysis of tissue extractions, whether by the UHPLC/MS method detailed here, or more prudently by GC/MS to compare results more directly with DART, itself a gas analysis method. Therefore, an ideal technique for ionization and detection of trace biomarkers for melanoma should be sought after, and two main routes of analysis could be explored. First, to continue down the path of direct tissue analysis. DART ionization was employed here, yet desorption electrospray ionization (DESI) and laser ablation electrospray ionization (LAESI) could provide useful insight

PAGE 231

231 into localization of select biomarkers. While DART is an excellent tool for rapid analysis of any su rface, is does not allow localization of markers. DESI and LAESI, while differing in their method of analyte extraction (Figures 5 1 and 5 2), both allow projection of spectral data onto a 2D coordinate plane. Ionization directly off of human skin have bee n detailed before, 141 144 and further adaptation could provide clinical applicability. Another solution would be to develop a technique for headspace analysis, as this would provide the least invasive procedure possible. Use of a device to sample headspace produced by skin emanations has also been studied previously. 145 Improving on current methods may prove most fruitful, especially in combination with the more volatile biomarkers. If developing a useful device that attaches directly to the skin is impractical, another venture may involve analysis of swabs of the skin site. Swabbing a skin lesions with alcohol or saline to sanitize the biopsy site is common practice in the dermatology office, however, those swabs of immediately trashed as they are of no use. On the other hand, if there were a way to analyze those swabs directly, i.e., DART/MS, a noninvasive screening could be as simple as swabbing a potential lesions and performing subsequent analysis on the swab, rather than the human tissue or emanation above the skin. To conc lude, there are vast avenues to be explored to further the work presented here. Each has its merits and its issues, yet the one thing all of these techniques have in common is potential. Potential is the key to scientific progress, whether it leads to succ ess or failure, if something has the potential to try it, knowledge will be gained and progress will be made. Breath analysis by FAIMS and noninvasive metabolic screening for melanoma are two concepts that I truly believe have that promise to revolutionize the

PAGE 232

232 biomedical community, and pave the way for drug monitoring and disease diagnosis in the future.

PAGE 233

233 Figure 5 1. DESI process. T he ESI spray is directed toward the sample stage where charged solvent droplets desorb analyte particles. These charged ana lyte droplets are directed toward the MS inlet by an electric potential between the stage and the inlet.

PAGE 234

234 Figure 5 2. LAESI process. N ear IR laser pulse is shot at the sample, ablating away some of the analyte. The ablated particles interact wi th the charged ESI solvent droplets to produce ionized analyte droplets, which then enter the MS.

PAGE 235

235 LIST OF REFERENCES (1) Skoog, D. A.; Holler, F. J.;Crouch, S. R. Principles of Instrumental Analysis . 6th ed.; Thomso n Brooks/Cole: Belmont, CA, 2007. (2) Willis, B. G. Hewlett Packard J. 1980 , 31 , 3 5. (3) Howell, J. A., in Handbook of intrumental techniques for analytical chemistry ; F. Settle, Ed.; Prentice Hall: Upper Saddle River, NJ, 1997; Chap. 25, pp 481 506. (4) Domin, M.;Cody, R., Eds. Ambient Ionization Mass Spectrometry . Royal Society of Chemistry: London, UK, 2014; p 508. (5) Huang, M. Z.; Yuan, C. H.; Cheng, S. C.; Cho, Y. T.;Shiea, J. Annu. Rev. Anal. Chem. 2010 , 3 , 43 65. (6) Cooks, R. G.; Ouyang, Z. ; Takats, Z.;Wiseman, J. M. Science 2006 , 311 , 1566 1570. (7) Monge, M. E.; Harris, G. A.; Dwivedi, P.;Fernandez, F. M. Chem. Rev. 2013 , 113 , 2269 2308. (8) Cody, R. B.; Laramee, J. A.;Durst, H. D. Anal. Chem. 2005 , 77 , 2297 302. (9) Carroll, D. I.; Dzidic, I.; Stillwell, R. N.; Horning, M. G.;Horning, E. C. Anal. Chem. 1974 , 46 , 706 10. (10) Horning, E. C.; Carroll, D. I.; Dzidic, I.;Stillwell, R. N. Pure Appl. Chem. 1978 , 50 , 113 27. (11) Watson, J. T.;Sparkman, O. D. Introduction to Mass Spectr ometry: Instrumentation, Applications, and Strategies for Data Interpretation . 4th ed.; Wiley Inc.: West Sussex, UK, 2007. (12) Finnigan LTQ Hardware Manual, Revision A. Thermo Electron Corporation: San Jose, CA, 2003, p 202. (13) Martin, C. J.; Das, S.; Young, A. C. J Appl Physiol Respir Environ Exerc Physiol 1979 , 47 , 319 24. (14) Iribarne, J. V.;Thomson, B. A. J. Chem. Phys. 1976 , 64 , 2287 94. (15) Dole, M.; Mack, L. L.; Hines, R. L.; Mobley, R. C.; Ferguson, L. D.;Alice, M. B. J. Chem. Phys. 1968 , 49 , 2240 9. (16) Fernandez de la Mora, J. Anal. Chim. Acta. 2000 , 406 , 93 104.

PAGE 236

236 (17) Nguyen, S.;Fenn, J. B. P. Natl. Acad. Sci. USA 2007 , 104 , 1111 1117. (18) Chen, H.; Venter, A.;Cooks, R. G. Chem. Commun. 2006 , 2042 2044. (19) Chen, H.; Gamez, G.; Zenobi, R. J. Am. Soc. Mass. Spectrom. 2009 , 20 , 1947 63. (20) Berchtold, C.; Meier, L.;Zenobi, R. Int. J. Ion Mobil. Spectrom. 2011 , 299 , 145 150. (21) Chen, H.; Wortmann, A.; Zhang, W.;Zenobi, R. Angew. Chem. Int. Ed. 2007 , 46 , 580 583. (22) Law, W . S.; Wang, R.; Hu, B.; Berchtold, C.; Meier, L.; Chen, H.;Zenobi, R. Anal. Chem. 2010 , 82 , 4494 4500. (23) Abbott, C. E. Rev. Geophys. 1977 , 15 , 363 374. (24) Orme, M. Prog. Energ. Combust. 1997 , 23 , 65 79. (25) Agilent 6200 Series TOF and 6500 Seri es Q TOF LC/MS System Concepts Guide. Rev A. Agilent Technologies, Inc: Santa Clara, CA, 2012. (26) Exactive Operating Manual, Revision C. Thermo Fisher Scientific Inc.: Waltham, MA, 2010, p 170. (27) March, R. E. J. Mass Spectrom. 1997 , 32 , 351 369. ( 28) mzCloud: Advanced Mass Spectral Database (version , 2015. (29) National Institute of Advances Inducstrial Science and Technology SDBSWeb: , 2015. (30) National Institute of Standards and Technology Chemistry WebBook: , 2015. (31) Kanu, A. B.; Dwivedi, P.; Tam, M.; Matz, L.;Hill, H. H., Jr. J. Mass Spectrom. 2008 , 43 , 1 22. (32) McDaniel, E. W.; Martin, D. W.;Barnes, W. S. Rev. Sci. Instrum. 1962 , 33 , 2 7. (33) Albritton, D. L.; Miller, T. M.; Martin, D. W.;McDaniel, E. W. Phys. Rev. 1968 , 171 , 94 102. (34) Moseley, J. T.; Snuggs, R. M.; Marti n, D. W.;McDaniel, E. W. Phys. Rev. 1969 , 178 , 240 8.

PAGE 237

237 (35) Martin, D. W.;McDaniel, E. W. Mass spectrometric studies of mobilities, diffusion, and reactions of ions in gases ; Georgia Inst. Technol.: 1970. (36) Buryakov, I. A.; Krylov, E. V.; Nazarov, E. G .;Rasulev, U. K. Int. J. Ion Mobil. Spectrom. 1993 , 128 , 143 8. (37) Guevremont, R. J.Chromatogr. A 2004 , 1058 , 3 19. (38) Purves, R. W.; Guevremont, R.; Day, S.; Pipich, C. W.;Matyjaszczyk, M. S. Rev. Sci. Instrum. 1998 , 69 , 4094 4105. (39) Rorrer, L. C.;Yost, R. A. Int. J. Ion Mobil. Spectrom. 2011 , 300 , 173 181. (40) Krylov, E. V. Int. J. Ion Mobil. Spectrom. 2003 , 225 , 39 51. (41) Bryant, J. G.; Prieto, M.; Prox, T. A.;Yost, R. A. Int. J. Ion Mobil. Spectrom. 2010 , 298 , 41 44. (42) Prieto, M . High field asymmetric waveform ion mobility spectrometry: Evaluation of novel cell geometries and techniques. Dissertation, University of Florida, Gainesville, FL, 2010. (43) Shvartsburg, A. A.; Smith, R. D.; Wilks, A.; Koehl, A.; Ruiz Alonso, D.;Boyle, B. Anal. Chem. 2009 , 81 , 6489 6495. (44) Prieto, M.; Tsai, C. W.; Boumsellek, S.; Ferran, R.; Kaminsky, I.; Harris, S.;Yost, R. A. Anal. Chem. 2011 , 83 , 9237 9243. (45) Yost, R. A.; Tsai, C.; Beekman, C.; Boock, J. J.;Costanzo, M., presented at 244th ACS National Meeting & Exposition , Philidelphia, PA, 2012 . (46) Costanzo, M. T.; Kazaleh, M. S.; Kemperman, R. H. J.;Yost, R. A., presented at Proceedings of the 63rd ASMS Conference on Mass Spectrometry and Allied Topics , St. Louis, MO, 2015 . (47) Costa nzo, M. T.; Tsai, C.;Yost, R. A., presented at Proceedings of the 60th ASMS Conference on Mass Spectrometry and Allied Topics , Vancouver, BC, CA, 2012 . (48) Beekman, C.; Tasi, C.; Ugarov, M.; Stafford, G.;Yost, R., presented at Proceedings of the 60th ASM S Conference on Mass Spectrometry and Allied Topics , Vancouver, Canada, 2012 . (49) Giddings, J. C. Unified Separation Science . Wiley: 1991; p 320 pp. (50) Martin, A. J. P.;Synge, R. L. M. Biochem. J. 1941 , 35 , 1358 68.

PAGE 238

238 (51) Jolliffe, I. T. Principal Com penent Analysis . 2nd ed.; Springer Verlag: New York, NY, 2002. (52) Xia, J.; Psychogios, N.; Young, N.;Wishart, D. S. Nucleic Acids Res. 2009 , 37 , W652 W660. (53) Menger, R. F. Multivariate data analysis strategies for biological applications of matrix assisted laser desorption/ionization mass spectrometric imaging. Dissertation, University of Florida, Gainesville, FL, 2013. (54) O'Connell, F.; Thomas, V. E.; Pride, N. B.;Fuller, R. W. Am. J. Respir. Crit. Care. Med. 1994 , 150 , 374 80. (55) Fuller, R. W. Resp. Med. 1991 , 85 Suppl A , 31 4. (56) Irwin, R. S.; Corrao, W. M.;Pratter, M. R. Am. Rev. Respir. Dis. 1981 , 123 , 413 7. (57) Turgut, C.; Newby, B.;Cutright, T. J. Environ. Sci. Pollut. R. 2004 , 11 , 7 10. (58) O'Neil, M. The Merck Index: An Enc yclopedia of Chemicals, Drugs, and Biologicals . 14th ed.; Whitehouse Station, NJ, 2006. (59) Kopec, S. E.; DeBellis, R. J.;Irwin, R. S. Pulm. Pharmacol. Ther. 2002 , 15 , 529 534. (60) Kopec, S. E.; Irwin, R. S.; Debellis, R. J.; Bohlke, M. B.;Maher, T. J . Cough 2008 , 4 , 3. (61) Fujimura, M.; Kasahara, K.; Kamio, Y.; Naruse, M.; Hashimoto, T.;Matsuda, T. Eur. Respir. J. 1996 , 9 , 1624 1626. (62) Fujimura, M.; Kasahara, K.; Yasui, M.; Myou, S.; Ishiura, Y.; Kamio, Y.; Hashimoto, T.;Matsuda, T. Eur. Respi r. J. 1998 , 11 , 1060 3. (63) Davenport, P. W.; Sapienza, C. M.;Bolser, D. C. Eur. Respir. Rev. 2002 , 12 , 249 253. (64) Dicpinigaitis, P. V.; Allusson, V. R.; Baldanti, A.;Nalamati, J. R. Respiration 2001 , 68 , 480 2. (65) Dicpinigaitis, P. V.;Alva, R. V. Chest 2005 , 128 , 196 202. (66) Vovk, A.; Bolser, D. C.; Hey, J. A.; Danzig, M.; Vickroy, T.; Berry, R.; Martin, A. D.;Davenport, P. W. Pulm. Pharmacol. Ther. 2007 , 20 , 423 432. (67) Baeyens, F.; Hendrickx, H.; Crombez, G.;Hermans, D. Appetite 1998 , 31 , 185 204.

PAGE 239

239 (68) Coppens, E.; Vansteenwegen, D.; Baeyens, F.; Vandenbulcke, M.; Van, P. W.;Eelen, P. Neuropsychologia 2006 , 44 , 840 3. (69) Dickinson, A.;Brown, K. J. Learn Behav. 2007 , 35 , 36 42. (70) Midgren, B.; Hansson, L.; Karlsson, J. A.; Simo nsson, B. G.;Persson, C. G. A. Am. Rev. Respir. Dis. 1992 , 146 , 347 51. (71) Hegland, K. W.; Bolser, D. C.;Davenport, P. W. J. Appl. Physiol. 2012 , 113 , 39 46. (72) Hegland, K. W.; Troche, M. S.;Davenport, P. W. Front. Physiol. 2013 , 4 , 167. (73) Cao , W.;Duan, Y. Cr. Rev. Anal. Chem. 2007 , 37 , 3 13. (74) Modak, A. S. J. Breath R. 2010 , 4 , 017002/1 017002/6. (75) Popov, T. A. Ann. Allerg. Asthma Im. 2011 , 106 , 451 456. (76) Phillips, M. Sci. Am. 1992 , 267 , 74 9. (77) Murugan, A.; Prys Picard, C .;Calhoun, W. J. Curr. Opin. Pulm. Med. 2009 , 15 , 12 8. (78) Corradi, M.; Gergelova, P.;Mutti, A. Eur. Respir. Monogr. 2010 , 140 151. (79) Wehinger, A.; Schmid, A.; Mechtcheriakov, S.; Ledochowski, M.; Grabmer, C.; Gastl, G. A.;Amann, A. Int. J. Ion Mo bil. Spectrom. 2007 , 265 , 49 59. (80) Pauling, L.; Robinson, A. B.; Teranishi, R.;Cary, P. P. Natl. Acad. Sci. USA 1971 , 68 , 2374 6. (81) Diamant, Z.; Boot, J. D.; Mantzouranis, E.; Flohr, R.; Sterk, P. J.;van, W. R. G. Pulm. Pharmacol. Ther. 2010 , 23 , 468 481. (82) Mazzone, P. J. Eur. Respir. Monogr. 2010 , 130 139. (83) Hansen, J. E.; Ampaya, E. P.; Bryant, G. H.;Navin, J. J. J. Appl. Physiol. 1975 , 38 , 983 9. (84) Wozniak, J. A.; Davenport, P. W.;Kosch, P. C. J. Appl. Physiol. 1988 , 65 , 633 9. (85) Snell, N.;Newbold, P. Curr. Opin. Pharmacol. 2008 , 8 , 222 235. (86) Riess, U.; Tegtbur, U.; Fauck, C.; Fuhrmann, F.; Markewitz, D.;Salthammer, T. Anal. Chim. Acta. 2010 , 669 , 53 62.

PAGE 240

240 (87) Greenwald, R.; Fitzpatrick, A. M.; Gaston, B.; Marozkina, N. V.; Erzurum, S.;Teague, W. G. PloS one 2010 , 5 , e11919. (88) Kazani, S.;Israel, E. J. Breath R. 2010 , 4 , 047001/1 047001/13. (89) Grob, N. M.; Aytekin, M.;Dweik, R. A. J. Breath R. 2008 , 2 , 037004/1 037004/18. (90) Beauchamp, J.; Herbig, J.; Gutmann , R.;Hansel, A. J. Breath R. 2008 , 2 , 046001/1 046001/19. (91) Amann, A.; Poupart, G.; Telser, S.; Ledochowski, M.; Schmid, A.;Mechtcheriakov, S. Int. J. Ion Mobil. Spectrom. 2004 , 239 , 227 233. (92) Williams, J.; Poschl, U.; Crutzen, P. J.; Hansel, A. ; Holzinger, R.; Warneke, C.; Lindinger, W.;Lelieveld, J. J. Atmos. Chem. 2001 , 38 , 133 166. (93) Pollard, M.; Hilton, C.; Li, H.; Kaplan, K.; Yost, R.;Hill, H. Int. J. Ion Mobil. Spectrom. 2011 , 14 , 15 22. (94) Bessa, V.; Darwiche, K.; Teschler, H.; Sommerwerck, U.; Rabis, T.; Baumbach, J.;Freitag, L. Int. J. Ion Mobil. Spectrom. 2011 , 14 , 7 13. (95) Westhoff, M.; Litterst, P.; Freitag, L.; Urfer, W.; Bader, S.;Baumbach, J. I. Thorax 2009 , 64 , 744 8. (96 ) Maddula, S.; Rabis, T.; Sommerwerck, U.; Anhenn, O.; Darwiche, K.; Freitag, L.; Teschler, H.;Baumbach, J. Int. J. Ion Mobil. Spectrom. 2011 , 14 , 197 206. (97) Fink, T.; Baumbach, J. I.;Kreuer, S. J. Breath Res. 2014 , 8 , 027104/1 027104/11, 11 pp. (98 ) Owlstone Nanotech White Paper. Owlstone Nanotech Inc.: Cambridge, UK, 2006, p 13. (99) Sinues, P. M.; Kohler, M.;Zenobi, R. Anal. Chem. 2013 , 85 , 369 73. (100) Florida Statute 893.13 . The Florida Legislature: , 2014. accessed 2015 05 25. (101) King, J.; Kupferthaler, A.; Unterkofler, K.; Koc, H.; Teschl, S.; Teschl, G.; Miekisch, W.; Schubert, J.; Hinterhuber, H.;Amann, A. J. Breath R. 2009 , 3 , 027006/1 027006/16. (102) Teshima, N.; Li, J.; Toda, K.;Dasgupta, P. K. Anal. Chim. Acta. 2005 , 535 , 189 199.

PAGE 241

241 (103) Harvey, B. A.;Barra, J. Eur. J. Pharm. Biopharm. 2003 , 55 , 261 269. (104) Am. J. Respir. Crit. Care. Med. 2005 , 171 , 912 30. (105) Sinues, P. M. L.; Kohl er, M.;Zenobi, R. PloS one 2013 , 8 , e59909. (106) Seed, B. Silanizing Glassware. John Wiley & Sons, Inc: 1998, pp 13:3E: A.3E.1 A.3E.2. (107) Pulmonary alveolus . Encyclopædia Britannica Inc.: alveolus , 2013. accessed 2015 05 20. (108) American Cancer Society: , 2015. accessed 2015 06 28. (109) National Cancer Institute: , 2015. accessed 2015 06 28. (110) Uong, A.;Zon, L. I. J. Cell Physiol. 2010 , 222 , 38 41. (111) American Academy of Dermatology: , Online, 2015. acc essed 2015 06 28. (112) Kwak, J.; Gallagher, M.; Ozdener, M. H.; Wysocki, C. J.; Goldsmith, B. R.; Isamah, A.; Faranda, A.; Fakharzadeh, S. S.; Herlyn, M.; Johnson, A. T.;Preti, G. J. Chromatogr. B 2013 , 931 , 90 6. (113) Williams, H.;Pembroke, A. Lancet 1989 , 1 , 734. (114) Buszewski, B.; Rudnicka, J.; Ligor, T.; Walczak, M.; Jezierski, T.;Amann, A. TrAC, Trends Anal. Chem. 2012 , 38 , 1 12. (115) Nguyen, T. Can Bees Be Trained to Sniff Out Cancer? Smithsonian Magazine: Online, 2013. (116) Amundsen, T. ; Sundstrom, S.; Buvik, T.; Gederaas, O. A.;Haaverstad, R. Acta. Oncol. 2014 , 53 , 307 15. (117) Rudnicka, J.; Walczak, M.; Kowalkowski, T.; Jezierski, T.;Buszewski, B. Sens. Actuators, B 2014 , 202 , 615 621. (118) Taverna, G.; Tidu, L.; Grizzi, F.; Torr i, V.; Mandressi, A.; Sardella, P.; La Torre, G.; Cocciolone, G.; Seveso, M.; Giusti, G.; Hurle, R.; Santoro, A.;Graziotti, P. J. Urol. (N. Y., NY, U. S.) 2015 , 193 , 1382 1387. (119) Meij, T. G.; Ben Larbi, I.; Schee, M. P.; Lentferink, Y. E.; Paff, T.; Terhaar sive Droste, J. S.; Mulder, C. J.; Bodegraven, A. A.;Boer, N. K. Int. J. Cancer 2014 , 134 , 1132 1138.

PAGE 242

242 (120) Queralto, N.; Berliner, A. N.; Goldsmith, B.; Martino, R.; Rhodes, P.;Lim, S. H. J. Breath Res. 2014 , 8 , 027112/1 027112/13, 13 pp. (121) Wu, Y.; Huo, D.; Hou, C.; Fa, H.; Yang, M.;Luo, X. Chem. Res. Chin. Univ. 2014 , 30 , 572 577. (122) Lei, J. c.; Hou, C. j.; Huo, D. q.; Luo, X. g.; Bao, M. z.; Li, X.; Yang, M.;Fa, H. b. Rev. Sci. Instrum. 2015 , 86 , 025106/1 025106/8. (123) Abaffy, T.; Duncan, R.; Riemer, D. D.; Tietje, O.; Elgart, G.; Milikowski, C.;DeFazio, R. A. PloS one 2010 , 5 , e13813. (124) Abaffy, T.; Moller, M. G.; Riemer, D. D.; Milikowski, C.;DeFazio, R. A. Metabolomics : Official journal of the Metabolomic Society 2013 , 9 , 998 1008. (125) Buszewski, B.; Rudnicka, J.; Walczak, M.;Jezierski, T. Nova Biotechnol. Chim. 2014 , 13 , 21 27, 7 pp. (126) ATCC: Global Resource Center: , 2015. (127) Smith, C. A.; Want, E. J.; O'Maille, G.; Abagyan, R.;Siuzdak, G. Anal. Chem. 2006 , 78 , 779 787. (128) Tautenhahn, R.; Bottcher, C.;Neumann, S. BMC Bioinformatics 2008 , 9 , 504. (129) Benton, H. P.; Want, E. J.;Ebbels, T. M. D. BMC Bioinformatics 2010 , 26 , 2488 2489. (130) Xia, J.; Mandal, R.; Sinelnikov, I. V.; Broadhurst, D.;Wishart, D. S. Nucleic Acids Res. 2012 , 40 , W127 W133. (131) Xia, J.; Sinelnikov, I. V.; Han, B.;Wishart, D. S. Nucleic Acids Res. 2015 . (132) Wishart, D. S.; Tzur, D.; Knox, C.; Eisner, R.; Guo, A. C.; Young, N.; Cheng, D.; Jewell, K.; Arndt, D.; Sawhney, S.; Fung, C.; Nikolai, L.; Lewis, M.; Coutouly, M. A.; Forsythe, I.; Tang, P.; Shrivastava, S.; Jeroncic, K.; Stothard, P.; Amegbey, G.; Block, D.; Hau, D. D.; Wagner, J.; Miniaci, J.; Clements, M.; Ge bremedhin, M.; Guo, N.; Zhang, Y.; Duggan, G. E.; Macinnis, G. D.; Weljie, A. M.; Dowlatabadi, R.; Bamforth, F.; Clive, D.; Greiner, R.; Li, L.; Marrie, T.; Sykes, B. D.; Vogel, H. J.;Querengesser, L. Nucleic Acids Res. 2007 , 35 , D521 6.

PAGE 243

243 (133) Wishar t, D. S.; Knox, C.; Guo, A. C.; Eisner, R.; Young, N.; Gautam, B.; Hau, D. D.; Psychogios, N.; Dong, E.; Bouatra, S.; Mandal, R.; Sinelnikov, I.; Xia, J.; Jia, L.; Cruz, J. A.; Lim, E.; Sobsey, C. A.; Shrivastava, S.; Huang, P.; Liu, P.; Fang, L.; Peng, J. ; Fradette, R.; Cheng, D.; Tzur, D.; Clements, M.; Lewis, A.; De Souza, A.; Zuniga, A.; Dawe, M.; Xiong, Y.; Clive, D.; Greiner, R.; Nazyrova, A.; Shaykhutdinov, R.; Li, L.; Vogel, H. J.;Forsythe, I. Nucleic Acids Res. 2009 , 37 , D603 10. (134) Wishart, D . S.; Jewison, T.; Guo, A. C.; Wilson, M.; Knox, C.; Liu, Y.; Djoumbou, Y.; Mandal, R.; Aziat, F.; Dong, E.; Bouatra, S.; Sinelnikov, I.; Arndt, D.; Xia, J.; Liu, P.; Yallou, F.; Bjorndahl, T.; Perez Pineiro, R.; Eisner, R.; Allen, F.; Neveu, V.; Greiner, R.;Scalbert, A. Nucleic Acids Res. 2013 , 41 , D801 7. (135) METLIN Metabolite Database: , 2015. (136) Miller, J. C.;Miller, J. N. Statistics and Chemometrics for Analytical Chemistry, 4th Edition . Prentice Hall: 2000; p 271 pp. (137) Rokach, L.;Maimon, O., in Data Mining and Knowledge Discovery Handbook ; O. Maimon, L. Rokach, Ed.; Springer US: 2005; Chap. 15, pp 321 352. (138) Breiman, L. Machine Learning 2001 , 45 , 5 32. (139) Ludolph, B. S.; Jeng, C. Y.; Chu, A. H. T.;Langer, S. H. J. Chromatogr. A 1994 , 660 , 3 15. (140) Zeisel, S. H.;da Costa, K. A. Nutrition reviews 2009 , 67 , 615 623. (141) Chen, H.; Wortmann, A.;Zenobi, R. J. Mass Spectrom. 2007 , 42 , 1123 1135. (142) Zhao, M.; Zhang, S.; Yang, C.; Xu, Y.; Wen, Y.; Sun, L.;Zhang, X. J. Forensic Sci. 2008 , 53 , 807 811. (143) Katona, M.; Denes, J.; Skoumal, R.; Toth, M.;Takats, Z. Analyst 2011 , 136 , 835 840. (144) Martin, H. J.; Reynolds, J. C.; Riazanskaia, S.;Thomas, C. L. P . Analyst 2014 , 139 , 4279 4286. (145) Chen, H.; Hu, B.; Hu, Y.; Huan, Y.; Zhou, Z.;Qiao, X. J. Am. Soc. Mass. Spectrom. 2009 , 20 , 719 22.

PAGE 244

2 44 BIOGRAPHICAL SKETCH Michael Thomas Costanzo was born in 1989, and grew up in the City of Tonawanda, a small sub urb of Buffalo, NY. In addition to snowman construction skills and learning to speak at an incredible speed, growing up in Buffalo gave Michael the mindset to never accept failure, no matter the difficulties (as demonstrated by every professional sports te am that calls Buffalo home). While he has always had a passion a chemist began to take shape. After graduating in 2006, he attended the State University of New Yo rk at Buffalo to obtain his Bachelor of Science in chemistry. While there, he examined products of enzymatic protein digestion by mass spectrometry under Dr. Troy Wood, in the interest of studying metabolic markers of autism in young children. By 2010, he was moving out of NY for the first time in his life to pursue his analytical chemistry aspirations at the University of Florida. Conducting research under Dr. Richard A. Yost, Michael has had the unique opportunity to engage in a variety of experimental me thodologies and instrumentation. From UV spectroscopy, to human breath analysis utilizing novel mobility separation devices, to metabolomics studies of melanoma by mass spectrometry, he has acquired a vast knowledge of analytical techniques and an even gre ater desire for scientific progress than he thought possible. It is in this vain that Michael will continue to expand his chemistry horizons as he strives for a career that allows him to continue laboratory research while also allowing him to test drive hi s marketing skills.