Publication - AC Birox

Please cite this article in press as: A.A. Inamdar, et al., Prediction of clinical outcomes using the pyrolysis, gas chromatography, and
differential mobility spectrometry (Py-GC-DMS) system, J. Anal. Appl. Pyrol. (2016), http://dx.doi.org/10.1016/j.jaap.2016.02.019
ARTICLE IN PRESSG Model
JAAP-3680; No.of Pages10
Journal of Analytical and Applied Pyrolysis xxx (2016) xxx–xxx
Contents lists available at ScienceDirect
Journal of Analytical and Applied Pyrolysis
journal homepage: www.elsevier.com/locate/jaap
Prediction of clinical outcomes using the pyrolysis, gas
chromatography, and differential mobility spectrometry
(Py-GC-DMS) system
Arati A. Inamdara
, Parag Borgaonkarb
, Yvonne K. Remachea
, Shalini Nairb
,
Waleed Maswadehc
, Amit Limayeb
, Arnold P. Snyderc
, Andrew Pecorad
, Andre Goyd
,
K. Stephen Suha,∗
a
The Genomics and Biomarkers Program, The John Theurer Cancer Center, Hackensack University Medical Center, Hackensack, NJ 07601, United States
b
AC Birox, LLC, Newark, NJ 07102, United States
c
U.S. Army Edgewood Chemical Biological Center (ECBC), Aberdeen Proving Ground, MD 21010, United States
d
Clinical Divisions, John Theurer Cancer Center, Hackensack University Medical Center, Hackensack, NJ 07601, United States
a r t i c l e i n f o
Article history:
Received 2 October 2015
Received in revised form 18 February 2016
Accepted 24 February 2016
Available online xxx
Keywords:
Biomarker
Clinical outcome
Differential mobility spectrometry
Ionization signature
Gas chromatography
Mantle cell lymphoma
Pyrolysis
a b s t r a c t
Biological and molecular heterogeneity of human diseases especially cancers contributes to variations in
treatment response, clinical outcome, and survival. The addition of new disease- and condition-specific
biomarkers to existing clinical markers to track cancer heterogeneity provides possibilities for further
assisting clinicians in predicting clinical outcomes and making choices of treatment options. Ionization
patterns derived from biological specimens can be adapted for use with existing clinical markers for early
detection, patient risk stratification, treatment decision making, and monitoring disease progression. In
order to demonstrate the application of pyrolysis, gas chromatography, and differential mobility spec-
trometry (Py-GC-DMS) for human diseases to predict the outcome of diseases, we analyzed the ionized
spectral signals generated by instrument ACB2000 (ACBirox universal detector 2000, ACBirox LLC, NJ,
USA) from the serum samples of Mantle Cell Lymphoma (MCL) patients. Here, we have used mantle
cell lymphoma as a disease model for a conceptual study only and based on the ionization patterns of
the analyzed serum samples, we developed a multivariate algorithm comprised of variable selection and
reduction steps followed by receiver operating characteristic curve (ROC) analysis to predict the probabil-
ity of a good or poor clinical outcome as a means of estimating the likely success of a particular treatment
option. Our preliminary study performed with small cohort provides a proof of concept demonstrating
the ability of this system to predict the clinical outcome for human diseases with high accuracy suggest-
ing the promising application of pyrolysis, gas chromatography, and differential mobility spectrometry
(Py-GC-DMS) in the field of medicine.
© 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND
license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
Abbreviations: bcl-1, b-cell lymphoma-isoform 1; CDK4, Cyclin dependent
kinase 4; CDKN2A, cyclin dependent kinase inhibitor 2A; MIB-1, mindbomb 1; SOX-
11, sry-related HMG box; IL-2R␣, interleukin-2 receptor alpha; IL-8, interleukin-8;
MIP-1␤, macrophage inflammatory protein-1 beta.
∗ Corresponding author at: The Genomics and Biomarkers Program, John Theurer
Cancer Center at Hackensack University Medical Center, 40 Prospect Avenue, David
Jurist Research Building, Hackensack, NJ 07601, United States.
E-mail addresses: AInamdar@HackensackUMC.org (A.A. Inamdar),
borgaonkar.parag@gmail.com (P. Borgaonkar), YRemache@HackensackUMC.org
(Y.K. Remache), shalini.nair@acbirox.com (S. Nair), wmmaswad2001@yahoo.com
(W. Maswadeh), al.limaye@logistic-solutions.com (A. Limaye),
Arnold.P.Snyder.civ@mail.mil (A.P. Snyder), APecora@HackensackUMC.org
(A. Pecora), AGoy@HackensackUMC.org (A. Goy), KSuh@HackensackUMC.org
(K.S. Suh).
1. Introduction
Despite advances in the treatment of hematological malignan-
cies, a subset of patients continues to develop progressive disease,
to be refractory to treatment, or to relapse shortly after treat-
ment. Hematological malignancies often pose a challenge in terms
of early detection, diagnosis, and prediction of clinical outcomes
[1]. With robust biomarker-based detection methods, extremely
sick patients can be stratified prior to treatment and recognized
as a high-risk group with poor prognosis. However, conventional
methods to detect clinical biomarkers usually involve invasive pro-
cedures for procuring blood or tissue biopsy, and require lengthy
laboratory diagnostic confirmation [2]. Recently, corroborative
research efforts in the field of oncology have led to use of molec-
ular biomarkers for early diagnosis and accurate prediction of
http://dx.doi.org/10.1016/j.jaap.2016.02.019
0165-2370/© 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.
0/).

2 A.A. Inamdar et al. / Journal of Analytical and Applied Pyrolysis xxx (2016) xxx–xxx
prognosis for various cancers [3]. However, no simple, rapid, and
non-invasive test exists that is capable of performing initial patient
selection, accurately predicting treatment response, and assigning
the probability of good or poor outcome. We used a technol-
ogy that combines pyrolysis, gas chromatography, and differential
mobility spectrometry (Py-GC-DMS) for detecting ionized particles
from human bio-specimens derived from Mantle cell lymphoma
(MCL) patients as an example to assess the diagnostic ability of
this technology to directly predict or supplement other biological
biomarkers to predict clinical outcomes.
MCL is cytogenetically characterized by a t(11;14) transloca-
tion and a bcl-1 rearrangement resulting in the overexpression
of cyclin D1 [4–6]. Presence of complex karyotypes, deregula-
tion of key cell cycle regulators associated with p53 mutations,
CDK4 activation, p16/CDKN2A inactivation, and inactivation of
DNA damage response pathways are associated with aggressive
clinical behavior and poor prognosis [7–9]. A high proliferation
index as assessed with Ki67/MIB-1 immunostaining, SOX11 over-
expression, p53 alterations, and blastoid morphology have been
reported to predict a poor outcome in MCL patients [10–12]. Other
serum biomarkers such as higher levels of ␤-2 microglobulin, free
immunoglobulin light chain in serum (either monoclonal or poly-
clonal), IL-2R␣, IL-8, and MIP-1␤ correlate with poor prognosis and
inferior outcomes [13–15]. Although morphological features along
with genetic and serum biomarkers appear to be useful for predict-
ing outcomes, their prognostic significance has not been confirmed
in all studies [16,17]. Furthermore, assessment of overall prognosis
using the MCL International Prognostic Index (MIPI) lacks accuracy
and is dependent on the treatment regimen [18]. In view of these
deficits, it is evident that there is a lack of diagnostic methodology
that can predict the clinical outcome before initiation of treatment
for MCL.
Gas chromatography (GC)-related technologies have shown
cost-effective utility in clinical diagnostic settings to resolve issues
posed by the limitations of current clinical diagnostic procedures
and laboratory associated costs. Several GC and other related “sniff
and tell” technologies were previously developed for detection and
biosensor related purposes [19–21]; these have included detection
of infections and diagnosis of metabolic disorders [22,23]. A com-
bination of gas chromatography and mass spectrometry (GC–MS)
has been used for cancer detection via a “breath analyzer” that
detects volatile organic compounds (VOC), which are elaborated
from different types of cancer, including lung and breast cancers
[24–26]. More recently, GC-related technologies have been more
widely applied in clinical settings to identify serum metabolites and
biomarkers for different cancer types, including gastric, colorectal,
esophageal, and pancreatic cancers [27–30]. Similarly, GC–MS tech-
nologies have also been developed for diagnosis of hematological
malignancies [31]. Ionization pattern recognition has been used in
detection technologies, and has recently been under investigation
for potential use in diagnosis of cancer and infectious diseases and
for rapid analysis of potential bioterror agents [32–36]. For detec-
tion of pathogens or markers of interest, approaches to ionization
pattern recognition have used a multivariate algorithm that can
immediately differentiate the pathogens or markers only present
in the experimental or case sets as opposed to control sets [37].
In this study, we analyzed twenty one serum samples of MCL
patients using “ionized” biochemical signatures produced by the
ACB2000 instrument. Signatures were first quantified, and the
signal intensities were then differentiated using new multivari-
ate data analysis algorithm approach and were finally associated
with the clinical phenotype of interest to generate an ionization
signature map. In combination with analysis of DNA, protein, or
metabolite expression, the ionization signature map may provide
a biomarker panel that could permit patient-stratification and pre-
diction of clinical outcome with high sensitivity and specificity. The
ionization signature map generated by this technology provides a
“second opinion” that is independent of existing clinical biomarkers
and established scoring systems, thus, enhancing clinical decision-
making for human diseases. We thus illustrate the application of
pyrolysis, gas chromatography, and differential mobility spectrom-
etry system in the field of cancer medicine.
2. Materials and methods
2.1. Study population
This study was approved by the Hackensack University Med-
ical Center’s (HUMC) Institutional Review Board (IRB) (reference
# CR00002851) and was performed under MCL study proto-
col number, PRO00001689 titled “Biomarker studies on Mantle
Cell Lymphoma”. Informed consents were obtained from all
patients prior to enrollment. Aliquots of serum samples obtained
under standard of care procedures were registered in the
HUMC Tissue Repository and stored at −80 ◦C. Serum samples
from untreated, newly diagnosed, chemo-naive (CN) (n = 15) and
relapsed, chemo-exposed (CE) (n = 6) MCL patients (total n = 21)
were studied. Clinical characteristics of the patients used in this
study include leukemic phase status (increased WBC/lymphocyte
count); blastoid (presence of blastoid cells) status; MIPI score (low,
intermediate, and high); pre-treatment history (CN versus CE), type
of treatment regimen administered, clinical outcome [good out-
come (GO) versus poor outcome (PO)], progression-free survival
(PFS, from date of first line treatment), and current status (alive or
deceased) (Supplementary file 1, Table 1).
To evaluate the system, MCL patients were divided into
GO (n = 12/21) and PO (n = 9/21) based on a median overall
survival cut-off of 35 months irrespective of their prior treat-
ment status. Among 15 Chemo-naïve MCL patients, 11 patients
who subsequently received chemotherapeutic regimen, HCVAD
(Hyper-fractionated–Cyclophosphamide, Vincristine, Adriamycin
and Dexamethasone) with or without rituximab as first line
treatment (n = 11) were further divided into good outcome (GO,
n = 6/11) or poor outcome (PO = 5/11) according to the follow-
ing clinical outcome criteria: GO = relapse-free/progression free
survival > 35 months after initial treatment, or PO = more than 2
relapses or death < 35 months after initial treatment (Supplemen-
tary file 1, Table 1, gray portion). Since presence of leukemic phase,
blastoid condition, and high MIPI score in MCL are associated with
poor prognosis, we performed contingency analysis to determine
the association of these clinical features with the prognosis for the
MCL cohort used in our study. Our analysis suggested that there
was no significant association between these clinical features and
clinical outcome (as defined above) in this study (Fig. 1).
2.2. ACB2000 Py-GC-DMS detection system
A serum sample (2.5 ␮l) was applied to the sample probe, which
was inserted into the pyro-tube (1) and heated to 300 ◦C to gen-
erate pyrolytic fragments in the vapor phase. Vaporized pyrolytic
analytes were transported by a flow module (5) which injected into
the GC module (3) where the analytes were separated. The sepa-
rated analytes entered an ionization chamber inside the Differential
Mobility Spectrum (DMS) module (4), and the ionized analytes
were detected (Fig. 2). Typical DMS settings used for separation of
analytes were based on ion differential mobility in alternating elec-
tric fields, in the compensation voltage (V) linearly increased from
−40 V to +15 V in approximately one second. Positive and negative
ion signal intensities were simultaneously recorded by DMS at each
compensation voltage. The output spectrum generated by the DMS
consisted of 60 signal intensity values each from the negative and

A.A. Inamdar et al. / Journal of Analytical and Applied Pyrolysis xxx (2016) xxx–xxx 3
Table 1
Summary of the clinical information of the MCL sample cohorts used in the study.
Patient No.GenderAGE @ DXStage @ DxTreatment status
at blood
procurement
Leukemic @ DxMIPI SCORE
1 = Low (0–3);
2 = Intermediate
(4–5); 3 = High
(6–11)
␤2-MG @ DxBlastoid @ Dx1st line RxPFS (mth)Current StatusClinical Outcome
1 M 51 4 CN Y 2 6.17 N HCVAD 82.1 A GO
2 F 59 4 CN Y 3 10.3 N HCVAD 87.3 A GO
3 M 56 4 CN N 1 2.3 N HCVAD 47.3 A GO
4 M 53 4 CN N 1 1.59 Y HCVAD 69.7 A GO
5 M 43 4 CN Y 1 2.91 N HCVAD 80.3 A GO
6 M 57 4 CN Y 3 NA Y HCVAD 73.4 A GO
7 M 65 4 CN Y 1 3.54 Y P05165 66.8 A GO
8 M 60 4 CE N 2 4.81 N HCVAD 30.4 A GO
9 M 64 4 CN N 1 2.14 N HCVAD 51.5 A GO
10 M 82 4 CN Y 3 7.03 N None 36.0 D GO
11 F 28 4 CN N 1 NA N None 84 A GO
12 M 64 4 CE N 1 3.7 N HCVAD 43.8 A PO
13 M 48 4 CN N 1 1.88 N HCVAD 29.5 D PO
14 F 45 4 CN N 1 4.56 N HCVAD 12.9 D PO
15 M 55 4 CN Y 2 2.42 N HCVAD 8.3 D PO
16 F 73 4 CN Y 3 11.1 Y HCVAD 1.8 D PO
17 M 68 4 CN Y 3 7.6 N HCVAD 1.2 D PO
18 M 62 4 CE N 1 NA N FC/F 14.7 D PO
19 F 67 4 CE N 1 NA N HCVAD 18.9 D PO
20 F 65 4 CE N 1 NA N CHOP 8.7 A PO
21 F 79 4 CE Y 3 6.98 N FC/F 26.2 D PO
DX = diagnosis; M = male; F = female; CN = Chemo-naïve; CE = chemo-exposed; N = no; Y = yes; ␤2-MG = ␤2-microglobulin; HCVAD = Hyper-fractionated−Cyclophosphamide,
Vincristine, Adriamycin and Dexamethasone; CHOP = Cyclophosphamide, Doxorubicin [hydroxydaunomycin], Predinosolone; F = Fludarabine; P–05165 = protocol # 05165;
NA = not-available; A = alive, D = deceased
Fig. 1. Lack of association between clinical outcome and clinical features. Contingency analysis was performed to determine the association between clinical outcome (good
or poor) in MCL patients and clinical markers including leukemic phase (A) (p = 0.575), blastoid condition (B) (p = 0.4135), MIPI score (C) (p = 0.7141), and ␤2-microglobulin
(D) (p = 0.8454). No-significant association between clinical outcome and defined clinical features was found. The contingency analysis was performed and analyzed using
Fisher’s exact test or chi-square test.
positive ions (120 points total) acquired at an average rate of 1.5
spectra per second. The system generated a single relative intensity
(RI) for each sample by adding the intensity of signals correspond-
ing to the detected ions. Each sample was run in triplicate, and
average relative intensity from three runs was calculated for each
sample.
2.3. A new multivariate data analysis approach: variable
selection and reduction using receiver operating characteristics
(VSR-ROC) curve method algorithm
A new multivariate algorithm was developed using ROC statis-
tics described by Maswadeh and Snyder [37] for the classification
and identification of any two groups; in the current study, the
groups were good outcome (GO) and poor outcome (PO). The

Fig. 2. ACB2000 Py-GC-DMS detection system. The system and close-up view of the
components involved with sample processing: (1) Pyro-tube, (2) GC injector, (3)
GC, (4) DMS, (5) flow control module, (6) electronic control boards, (7) internal PC
adapter, (8) DC input.
data analysis algorithm comprised two components. The first com-
ponent of the algorithm was variable ranking and selection; the
second component was multivariate analysis, performed by com-
bining the highest ranked variables into a single variable, which was
used to generate a ROC curve. The ACD (area between the curve
and the diagonal) was calculated to separate GO and PO groups
at the maximum degree. The major variables, True Positive (TP),
False Positive (FP), True Negative (TN), and False Negative (FN) of
ROC curve statistics [37] as well as new variables unique to this
analysis such as Percentage Separation, Positive Population Center
likelihood (PPCL), Negative Population Center likelihood (NPCL),
Confidence level (CL) were further calculated for the two groups
(Supplementary file 2, part A and B).
2.3.1. Algorithm component 1: variable selection and ranking
Using the ACB2000 system, experimental data files were gener-
ated for serum samples of 21 MCL patients with known clinical
outcomes (12 with good outcomes, 9 with poor outcomes). As
described in Supplementary file 1 (part A) each experimental data
file generated a relative intensity for each sample. Upon subtract-
ing the variables (pixels) corresponding to ionic patterns found to
occur in common to both GO and PO groups, unique ionization sig-
natures consisting of 572 variables (pixels) were recognized. These
variables were further used to calculate the ACD [37] between the
GO and PO groups at each of 572 variables. For example, Table A1
shows the average intensity of signal for pixel (variable) # 195 for
MCL patients belonging to GO and PO groups. Fig. A1 shows the
frequency plot generated using pixel signals listed in Table A1 (see
Supplementary file 2, part A). Here, the frequency plot indicates a
moderate separation of good (green diamond, ) from poor (red
box, ) clinical outcome (Supplementary file 2, Fig. A1). To improve
the distinction between GO and PO patients, the ACD were calcu-
lated for these 572 variables, and the variables that gave the best
possible degree of separation between data-points of the GO and PO
groups were used for further analysis. The low ACD variables were
considered to be noise or to represent variables that were not suffi-
ciently sensitive to discriminate between GO and PO groups [37]. In
total, 59 variables, representing unique biomarkers, were identified
that were capable of distinguishing between the two groups with
a high degree of separation. These 59 variables, arranged in order
of highest to lowest ACD, are listed in Table A2 (Supplementary file
2, part A).
2.3.2. Algorithm component 2: top-variables combination and
ROC statistics
The second component of our new multivariate data analysis
approach utilizes mathematical steps (Steps 1 through 4, which are
explained in Supplementary file 2, part B) described by Maswadeh
and Snyder [37]. In brief, a new ACD was calculated from the ACDs
of first two variables (for example, #195 and #197) as described
in Step 1 and Step 2 (Supplementary file 2, part A, Fig. A2) and
degree of separation was calculated (Supplementary file 2, part A,
Fig. A3). The new ACD was then used to calculate a second new
ACD with the next variable (Supplementary file 2, Steps 3 and 4).
This iterative process was carried on in a sequential manner, ulti-
mately reducing all 59 variables to a single variable. The final single
variable was used to calculate an ACD representing the maximum
degree of separation between GO and PO patients (Supplementary
file 2, part A, Fig. A4). The corresponding ROC was used to calculate
major variables (TP, FP, TN, and FN) of ROC curve statistics as well
as new variables unique to this study (% Separation, PPCL, NPCL,
CL) between the GO and PO groups (See the description of steps in
Supplementary File 2, part B along with Fig. B1 and Table B1).
2.4. Data analysis
MCL patient serum samples (n = 21) with known, good or poor
clinical outcomes (Supplementary file 1, Table 1) were analyzed
using the ACB2000 system. Each experiment was performed in trip-
licate to generate the raw data set. Raw data files of all samples
were further processed to identify ionization signature patterns
restricted to either good or poor outcome by eliminating the vari-
ables with low ACD or ionization signatures that were shared
between the two groups. This processing by the first component
of the VSR-ROC algorithm identified 59 variables. The second com-
ponent of the VSR-ROC algorithm reduced these 59 variables to a
single variable representing the best possible separation between
GO and PO groups. The unique ionization signatures generated after
application of the VSR-ROC algorithm were assigned to either the
good or poor outcome group based on known clinical outcome.
Then same set of MCL serum samples (n = 21) was tested to verify
the ability of the VSR-ROC algorithm to identify patients with GO
and PO.
2.5. Statistical analysis
Data are summarized as mean ± standard error of the mean
(SEM) for relative intensity (RI) along with the probability (%P)
of patients experiencing favorable prognostic factors so as to be
labeled as good outcome (GO) or patients experiencing poor prog-
nostic factors and therefore having a poor outcome (PO). The
correlation between the clinical features and clinical outcome for
the cohort was assessed using contingency Table. Statistical anal-
ysis was performed using GraphPad Prism software. The details of
the statistical tests are indicated in the figure legends.
3. Results
3.1. Py-GC-DMS system generates a three-dimensional ionization
signature based on DMS signal intensity
Three-dimensional DMS signals were plotted as a function of
compensation voltage (x axis), retention time (y axis), and an
ion abundance (DMS signal intensity; z axis). A 2.5 ␮l sample of
patient serum was sufficient to generate 120 points that subse-
quently, separated into 60 negative and 60 positive ionized spectral
points. Typical waterfall display consistent with 3D-x, y, z format
(Fig. 3A) shows that ionized spectra represent ionized materials
from patient serum sample in continuous spectrum format that

Fig. 3. The display pattern generated by the ACB 200 Py-GC-DMS detection system. (A) Typical three-dimensional (3D) waterfall display of raw data with an average rate of
1.5 spectra/s demonstrates significant signatures in both the negative ion-space (right yellow axis) and the positive ion-space (left green axis). (B) A two-dimensional (2D)
pixel format obtained from 3D raw data representing 572 pixels as variables being incorporated into the multivariate algorithm for calculations. (For interpretation of the
references to colour in this figure legend, the reader is referred to the web version of this article.)
is divided into positive and negative areas. These same raw data
points can be converted into a 2-D ((x, y), z) pixel array format as
shown in Fig. 3B where each peak represents a sum of data points
quantifying an average intensity of pixels. All data output from
the triplicate analyses were quantitatively analyzed by Py-GC-DMS
system and presented as relative intensity for each patient serum
sample used in this study (Supplementary file 1, Table 2).

Table 2
Summary of clinical information on patients and relative intensity of the corre-
sponding sample detected with ACB2000 Py-GC-DMS.
PatientNo. Chemo-naïve, Y/N Treatment Clinical Outcome Status Relative Intensity
1 Y H-CVAD GO Alive 11.6
7 Y P-05165 GO Alive 49.9
8 N H-CVAD GO Alive 27.45
10 Y None GO Dead 79.8
11 Y None GO Alive 50.55
12 N H-CVAD GO Alive 14.55
13 N H-CVAD PO Dead 14.13
15 Y H-CVAD PO Dead 14.25
18 N F/FC PO Dead 65.3
20 N CHOP PO Alive 86.6
21 Y F/FC PO Dead 45.9
3.2. Relative signal intensity of ionization from each outcome
group is similar but ionized pattern is different
An individual relative intensity value for each MCL case was gen-
erated by averaging signal intensities of variables with good signal
(>9% of optimized signal) from triplicate runs. The average RI val-
ues for GO and PO groups were calculated by averaging the relative
intensity of all MCL patients sorted according to their treatment
outcomes (GO vs PO) (Supplementary file 1, Table 2). These average
RI values were not significantly different between GO (33.33 ± 5.62,
N = 12) and PO (40.50 ± 8.629, N = 9) where p = 0.4139, suggesting
that RI values alone cannot differentiate clinical outcomes. The
average RI value was determined for newly diagnosed chemo naïve
(CN)-MCL patients prior to HCVAD (Table 2, boxed data from GO
and PO groups). The CN-MCL patients with GO had an average RI
of 32.85 ± 6.675, N = 6, and CN-MCL patients with PO had an aver-
age RI of 29.19 ± 9.257, N = 5, p = 0.3743 indicating that differences
between GO and PO groups were not significant (Fig. 4B).
3.3. In silico modification of ionized signatures and optimization
of the multivariate algorithm increases accuracy for predicting
clinical outcome
Analysis of ionization signatures permitted identification of a
total of 572 pixels as independent variables specific to each ioniza-
tion signal on the output grid that were unique to good or poor
outcome groups but were not shared between the two groups.
Therefore, the signature output from each GO and PO sample
was only represented by group-specific ionization points. Receiver
Operating Characteristic curve analysis was performed on these
572 pixels, and the 59 pixels with the highest ACD were selected
(Supplementary File, Table A2). These 59 variables and correspond-
ing ionization signatures were further optimized by application
of the VSR-ROC algorithm (Supplementary file 2: Part A and B).
The steps (Supplementary file 2: Part B) were repeated with all 59
selected pixels (variables) which reduced the set of variables to a
single, aggregate variable for each outcome group. The final fre-
quency plot demonstrated the best possible separation of patient
populations with good versus poor clinical outcome (Supplemen-
tary file 2, Fig. A4). The degree of separation between good and poor
outcomes from the aggregate variable was 90%, and showed signif-
icant differences in the pattern and location of ionized signatures
A
True
G
O
False
PO
True
PO
False
G
O
0
10
20
30
40
50
60
70
80
90
100
*** **
Probability(%)
B
True
G
O
False
PO
True
PO
False
G
O
0
10
20
30
40
50
60
70
80
90
100
** NS
Probability(%)
Fig. 6. The ionization signature pattern specific for GO and PO groups is potentially
capable of accurately discriminating between good and poor outcome groups among
MCL patients. (A) The probability that a sample from any patient with a GO signature
pattern irrespective of prior treatment history would truly belong to the GO group
(true GO); the probability that a sample with a PO signature pattern would truly
belong to the PO group (true PO); the probability that a sample with GO signature
would be falsely considered under the PO group (false PO) and the probability that
a sample would be falsely considered under the GO group (false GO) were 88%,
89%, 12% and 11%, respectively. The significance of differences between true GO and
false PO groups, and true PO and false GO was analyzed using two-tailed paired t-
test; **p < 0.01, ***p < 0.001. (B) The probability that a sample from a chemo-naive
patient would be a true GO, false PO, true PO and false GO was 92%, 80%, 8% and
20%, respectively. Significance was analyzed using two-tailed paired t-test. NS, non-
significant; **p < 0.01. Error bars indicate SEM.
between the good and poor outcome groups that could be visually
identified (Fig. 5A and B).
3.4. A new algorithm and VSR-ROC statistics on ionization
signatures identify good and poor clinical outcome MCL patients
with high predictability and accuracy
The VSR-ROC algorithm used to construct a database of ioniza-
tion pixels with best ACD was tested against a set of MCL samples
with known clinical outcomes. Based on the analysis of the ionized-
signature pattern of this set of MCL samples, the probability (%P)
of obtaining a unique ionization signature pattern specific for good
outcome or poor outcome groups was calculated using percent GO
and percent PO from Supplementary file 1, Table 3. The probability
that a sample from a good outcome patient would show the ion-
ization signatures specific for good outcome (defined as true GO)
was 88% (p < 0.0009) for all GO patients irrespective of prior treat-
ment (Fig. 6A). The probability that a sample from a poor outcome
patient would show the ionized signature specific for poor out-

A B
G
O PO
0
10
20
30
40
50
NS
Relativeintensity
G
O PO
0
10
20
30
40
50 NS
Relativeintensity
Fig. 4. Analysis of correlation between relative intensity (RI) of signals and clinical outcomes. (A) Comparison of RI of GO and PO groups irrespective of previous treatment
history. (B) Comparison of RI of GO and PO groups in patients who were chemo-naïve prior to starting treatment. Each value is the average of RI from 12 GO and 9 PO samples;
each sample was run in triplicate. No significant differences in the GO and PO outcome groups were found. Significance was analyzed using two-tailed paired t-test. Error
bars indicate SEM.
Fig. 5. Ionization signature-pattern of MCL samples. Ionization signatures for an MCL patient with good outcome (A) and a patient with poor outcome (B) have distinct
patterns.
Table 3
Classification output (analysis) of MCL training test set.
Patient No. Intensity >9% %dY-R %dY-G % PO % GO Conf-NR Conf-R ClinicalOutcome
1 11.6 0 51 0 100 0 100 GO
2 47.5 5 41.5 21.5 78.5 4.5 95.5 GO
3 22.0 7 21 25 75 17 83 GO
4 29.4 0 67 0 100 9 91 GO
5 30.6 0 67.5 0 100 7 93 GO
6 56.0 0 56 0 100 15 85 GO
7 49.9 0 38 0 100 8.5 91.5 GO
8 38.8 0 85 0 100 0 100 GO
9 29.2 0 37 0 100 0 100 GO
10 79.8 0 9 0 100 0 100 GO
11 50.55 0 55 0 100 0 100 GO
12 14.55 75 0 100 0 95.5 4.5 GO
13 14.13333 34.66667 0 100 0.66666667 88.66667 11.33333 PO
14 56.95 47.5 0 100 0 100 0 PO
15 14.25 24 0 100 0 100 0 PO
16 14.6 39.5 0 100 0 100 0 PO
17 45.9 0 65 0 100 0 100 PO
18 24.3 58 0 100 0 92 8 PO
19 65.3 64 0 100 0 100 0 PO
20 42.5 14 0 100 0 81 19 PO
21 86.6 92 0 100 0 99 1 PO

come (defined as true PO) was 89% (p < 0.0081) for all PO patients
irrespective of treatment (Fig. 6A). Similarly, the predictability for
samples from chemo-naïve patients treated with HCVAD regimen
being true GO and PO was 92% (p = 0.004) and 80% (p = 0.1043),
respectively (Fig. 6 B). Thus, in the analysis of this set of patients,
the VSR-ROC algorithm differentiated ionization spectra that were
specific to good (11/12) or poor (8/9) clinical outcomes with accu-
racy levels of 92% and 89%, respectively, at 70% confidence. Thus,
we anticipate that the spectral analysis will provide 100% accuracy
for predicting clinical outcome if the confidence level is lowered.
4. Discussion
Cancer is the second most common cause of death in the U.S.,
and accounts for nearly one of every four deaths [38]. Although
there have been advancements in cancer diagnosis and manage-
ment, clinical trials and research studies indicate the need for
innovative personalized treatment regimens, especially because
not all patients respond to treatment or else succumb to multi-
ple relapses. Unfortunately, the conventional scientific approach
alone is minimally effective, and the research communities need to
consider alternate and innovative approaches that will supplement
and support the clinical findings [39]. Discovery of biomarkers for
predicting clinical outcomes is gaining momentum, but a robust
analytical tool that can support biomarker-based conclusions as
an independent secondary method would enhance the accuracy
of diagnosis, and would be especially helpful for differentiating
patients who are at high risk for poor clinical outcome. The avail-
ability of such a tool would facilitate patient selection, would
accelerate delivery of precision medicine, and would likely result
in better cancer care.
Pyrolysis is a form of thermolysis that couples extremely high
temperatures and an oxygen-depleted environment for optimal
vaporization of samples. Pyrolyzed analytes are further sepa-
rated by Py-GC–MS, and the resulting ionized signature represents
a unique biochemical fingerprint of that specific sample. This
technology has previously been used to test bio-specimens from
humans for various purposes. Three decades of Py-GC–MS research
have improved the minimal volume required for accurate analyses
[40–42], leading to a variety of uses. For example, Py-GC–MS has
been used as a diagnostic tool to discriminate between leukemic
and normal white blood cells [43,44]; screening bacterial species
[45]; analysis of DL-lactin/glycolic acid composition for orthope-
dic use [46], and structural analysis of neuromelanin in Parkinson’s
disease [47]. More recent uses of Py-GC–MS include detection of
volatile organic compounds in exhaled breath [48] and isolation of
malignant B cells from patients with chronic lymphocytic leukemia
(CLL) for prognostic studies [31]. Similar to these applications, our
data show that this technology is likely to be useful for predicting
clinical outcomes from analysis of MCL patient sera, which would
have implications for selecting the most appropriate therapeutic
approach.
We have selected a liquid tumor model to test and validate
the use of the Py-GC-DMS system. Because of its pathological het-
erogeneity and complexity, MCL typically carries a poor prognosis
with a propensity to develop drug resistance, primary treatment
failure, and early relapse, which results in shortened survival rang-
ing from a few months to four years [49,50]. To provide better
alternative treatment options, patients who are at high risk to
experience a poor outcome must be identified early and with
high accuracy; however, this is a difficult task with current clin-
ical methods. Current clinical markers for MCL include cyclin
D1/D2/D3, translocation status t(11;14)(q13;q32), Ki67 and SOX11.
These parameters, in combination with flow cytometric analysis for
immuno-phenotypic patterns, cell morphology, and MIPI scores are
used for diagnostic and prognostic purposes [51,52]. For example,
SOX11 is the most recent clinical diagnostic marker for mantle cell
lymphoma, and is capable of detecting both cyclin D1-positive and -
negative MCL cases [53]. Similarly, clinical parameters such as MIPI
score and cell morphology, although routinely used to determine
the prognosis for MCL patients, are of marginal prognostic signifi-
cance, and lack a high degree of sensitivity or specificity [18]. Even
in our cohort, we observed that patients with a low MIPI score were
as likely to die as were patients with a high MIPI score. Similarly,
high beta-2 microglobulin, leukemic, and blastoid status failed to
correlate significantly with clinical outcome (Supplementary file 1,
Table 1 and Fig. 1).
Our data showed that the multivariate algorithm based on Py-
GC–MS approach can potentially be applied with current diagnostic
and prognostic methods for improving accuracy for stratification
of patients and predicting clinical outcome with high accuracy
and reliability. Optimization through multiple repetitions using
modified multivariate statistical analysis algorithms, including a
two-component VSR-ROC algorithm, was carried out to increase
the accuracy of the ionization signatures to produce a most robust
and consistent high-intensity ionization pattern, obviating the
need to maintain a large set of signatures. After optimization, typi-
cal and visually distinguishable patterns of ionization signatures of
MCL patients belonging to good and poor clinical outcomes were
obtained (Fig. 5A and B). The results of our study indicate that the
optimized VSR-ROC algorithm and the visual display of ionization
signatures can be used in combination with clinical data sets to
differentiate patient samples according to their likely outcome.
The optimized ionization signatures specific for GO and PO
groups were then used to determine the ability to predict the clin-
ical outcome on a second set of patients (Supplementary file 1,
Table 3). Our results showed that the ionization signatures can be
further stratified to support prediction of clinical outcome for all
treatments combined or for patients receiving HCVAD treatment.
In this dataset, the probability between groups of identifying true
non-responders/PO and false responders/GO was found to be non-
significant, mainly due to small sample size (n = 5) (Fig. 6A and B);
however, a larger cohort with greater than 100 MCL biospecimen
would provide more false negative and positive datasets.
We expect that the high accuracy of prediction (>90%, p < 0.005)
from the current data will not be significantly reduced as the size
of cohort is increased if the clinical outcome criteria remain the
same as used in this study. Specific ionized fragments that dif-
ferentiated good versus poor outcome groups could be further
purified via high resolution chromatographic methods and this
remains as our future goal. The purpose of this preliminary study
was to explore the utility of Py-GC-DMS-derived ionization sig-
natures based biomarkers to predict the outcome of patients as
good and poor. Although, this study has implemented small cohort
of samples which gave out large number of variables (572 sig-
nals) initially, using VSR-ROC algorithm, we were able to further
reduce them down to 59 variables capable of demonstrating the
best possible separation of patient populations with good versus
poor clinical outcome. We understand that use of small sample
size often results into large number of variables and often over-
fitting of the data occurs leading to including of variables due to
noise and some not demonstrating the real difference [54]. In this
study, we incorporated VS-ROC algorithm based statistical anal-
ysis but other statistical methods especially Partial least squares
discriminant analysis (PLSDA) coupled with various cross valida-
tion to reliably minimize the error rate and improve the accuracy
of the prediction [54]. We are aware of such spurious correla-
tions occurring especially in situations where limited number is
present and will take into consideration optimal statistical strate-
gies by vigorously employing cross validation methods/models to
avoid common mistakes associated with statistical analysis with

biomarker related research in future studies with large cohort [55].
Such strategies will definitely help us to achieve higher probability
of accurate prediction for clinical outcome for the disease under
consideration.
In summary, we find several advantages of Py-GC-DMS-based
ionization signatures for supporting current clinical MCL mark-
ers and subsequently for any disease under consideration as an
independent analytical tool for guiding choices of individualized
treatment for patients. The analytic methods employed in our
study (1) required only 2.5 ␮l of serum per run, (2) took less than
90 seconds for detection time, (3) eliminated any special handling
and chemical reagents, (4) provided automation of all data pro-
cessing, and (5) have the inherent capacity for addition of new
signatures to the existing database to improve accuracy. Our study
suggests that Py-GC-DMS ionization signatures in combination
with existing clinical markers could support clinical decisions for
treatment strategies, risk stratification, and reducing the cost of
cancer care. As a predictive tool, this system promises to provide
information that will help avoid unnecessary treatment expenses
for patients with a high likelihood of being poor responders, provid-
ing for early consideration of adjuvant therapies for such high-risk
patients, and offering a convenient and rapid method for predicting
the response of an individual patient to a therapeutic agent.
5. Conclusion
The Py-GC-DMS system can be readily adapted to existing clin-
ical markers to support stratification of patients for diagnosis,
prognosis, and prediction of clinical outcomes in MCL and possi-
bly in other cancers. The algorithm can be extrapolated to include
more ionization profiles. The systematic analysis of ionization sig-
natures can provide prognostic information of high specificity and
sensitivity relevant to the patient selection process. The ionization
profiling can be incorporated into current clinical decision-making
as an independent technological approach, and this technology can
potentially improve point-of-screening and point-of-care clinical
protocols.
Competing interests
The authors declare that they have no competing interests.
Authors’ contributions
AI analyzed data and drafted the manuscript. PB and YR carried
out AC2000 experiments and generated preliminary version of the
manuscript. SN, WM, AL and APS participated in analyses of data. AP
and AG participated in discussion. KSS designed and directed the
project, supervised all experiments, analyzed research data, edit
and wrote the manuscript. All authors read and approved the final
manuscript.
Acknowledgements
We thank the John Theurer Cancer Center and Hackensack Uni-
versity Medical Center for supporting funding and preparation of
this manuscript. We also thank the Elyssa and Jack Schecter Family
Fund, Leon Lowenstein Foundation, Walter & Louise Sutcliffe Foun-
dation and Newman’s Own Foundation for supporting research at
the Genomics and Biomarkers Program of HackensackUMC.
Appendix A. Supplementary data
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.jaap.2016.02.019.
References
[1] D.A. Howell, A.G. Smith, A. Jack, R. Patmore, U. Macleod, E. Mironska, E.
Roman, Time-to-diagnosis and symptoms of myeloma, lymphomas and
leukaemias: a report from the Haematological Malignancy Research Network,
BMC Hematol. 13 (2013) 9.
[2] D. Banerjee, Recent advances in the pathobiology of Hodgkin’s Lymphoma:
potential impact on diagnostic predictive, and therapeutic strategies, Adv.
Hematol. 2011 (2011) 439456.
[3] S. Sethi, S. Ali, P.A. Philip, F.H. Sarkar, Clinical advances in molecular
biomarkers for cancer diagnosis and therapy, Int. J. Mol. Sci. 14 (2013)
14771–14784.
[4] M. Raffeld, E.S. Jaffe, bcl-1, t(11;14), and mantle cell-derived lymphomas,
Blood 78 (1991) 259–263.
[5] M.E. Williams, S.H. Swerdlow, C.L. Rosenberg, A. Arnold, Centrocytic
lymphoma: a B-cell non-Hodgkin’s lymphoma characterized by chromosome
11 bcl-1 and PRAD 1 rearrangements, Curr. Top. Microbiol. Immunol. 182
(1992) 325–329.
[6] Y. Yatabe, R. Suzuki, K. Tobinai, Y. Matsuno, R. Ichinohasama, M. Okamoto, M.
Yamaguchi, J. Tamaru, N. Uike, Y. Hashimoto, Y. Morishima, T. Suchi, M. Seto,
S. Nakamura, Significance of cyclin D1 overexpression for the diagnosis of
mantle cell lymphoma: a clinicopathologic comparison of cyclin D1-positive
MCL and cyclin D1-negative MCL-like B-cell lymphoma, Blood 95 (2000)
2253–2261.
[7] C. Sarkozy, C. Terré, F. Jardin, I. Radford, C. Roche-Lestienne, D. Penther, C.
Bastard, S. Rigaudeau, S. Pilorge, F. Morschhauser, D. Bouscary, R. Delarue, H.
Farhat, P. Rousselot, O. Hermine, H. Tilly, S. Chevret, S. Castaigne, Complex
karyotype in mantle cell lymphoma is a strong prognostic factor for the time
to treatment and overall survival, independent of the MCL international
prognostic index, Genes. Chromosomes Cancer 53 (2014) 106–116.
[8] L. Hernández, S. Beà, M. Pinyol, G. Ott, T. Katzenberger, A. Rosenwald, F. Bosch,
A. López-Guillermo, J. Delabie, D. Colomer, E. Montserrat, E. Campo, CDK4 and
MDM2 gene alterations mainly occur in highly proliferative and aggressive
mantle cell lymphomas with wild-type INK4a/ARF locus, Cancer Res. 65
(2005) 2199–2206.
[9] D. Hui, T. Reiman, J. Hanson, R. Linford, W. Wong, A. Belch, R. Lai,
Immunohistochemical detection of cdc2 is useful in predicting survival in
patients with mantle cell lymphoma, Mod. Pathol. 18 (2005) 1223–1231.
[10] W. Klapper, E. Hoster, O. Determann, I. Oschlies, J. van der Laak, F. Berger,
H.W. Bernd, J. Cabeç adas, E. Campo, S. Cogliatti, M.L. Hansmann, P.M. Kluin, R.
Kodet, Y.A. Krivolapov, C. Loddenkemper, H. Stein, P. Möller, T.E. Barth, K.
Müller-Hermelink, A. Rosenwald, G. Ott, S. Pileri, E. Ralfkiaer, G. Rymkiewicz,
J.H. van Krieken, H.H. Wacker, M. Unterhalt, W. Hiddemann, M. Dreyling,
Ki-67 as a prognostic marker in mantle cell lymphoma—consensus guidelines
of the pathology panel of the European MCL Network, J. Hematopathol. 2
(2009) 103–111.
[11] M. Meggendorfer, W. Kern, C. Haferlach, T. Haferlach, S. Schnittger, SOX11
overexpression is a specific marker for mantle cell lymphoma and correlates
with t(11;14) translocation, CCND1 expression and an adverse prognosis,
Leukemia 27 (2013) 2388–2391.
[12] J. Slotta-Huspenina, I. Koch, L. de Leval, G. Keller, M. Klier, K. Bink, M. Kremer,
M. Raffeld, F. Fend, L. Quintanilla-Martinez, The impact of cyclin D1 mRNA
isoforms, morphology and p53 in mantle cell lymphoma: p53 alterations and
blastoid morphology are strong predictors of a high proliferation index,
Haematologica 97 (2012) 1422–1430.
[13] C. Yoo, D.H. Yoon, S. Kim, J. Huh, C.S. Park, C.J. Park, S.W. Lee, C. Suh, Serum
beta-2 microglobulin as a prognostic biomarker in patients with mantle cell
lymphoma, Hematol. Oncol. (2015), http://dx.doi.org/10.1002/hon.2188.
[14] T.E. Witzig, M.J. Maurer, T.M. Habermann, B.K. Link, I.N. Micallef, G.S.
Nowakowski, S.M. Ansell, J.P. Colgan, D.J. Inwards, L.F. Porrata, S.N. Markovic,
P.B. Johnston, Y. Lin, C. Thompson, M. Gupta, J.A. Katzmann, J.R. Cerhan,
Elevated monoclonal and polyclonal serum immunoglobulin free light chain
as prognostic factors in B- and T-cell non-Hodgkin lymphoma, Am. J. Hematol.
89 (2014) 1116–1120.
[15] M.B. Sonbol, M.J. Maurer, M.J. Stenson, C. Allmer, B.R. LaPlant, G.J. Weiner,
W.R. Macon, J.R. Cerhan, T.E. Witzig, M. Gupta, Elevated soluble IL-2R␣ IL-8,
and MIP-1␤ levels are associated with inferior outcome and are independent
of MIPI score in patients with mantle cell lymphoma, Am. J. Hematol. 89
(2014) E223–E227.
[16] H.J. Dong, L.T. Zhou, C. Fang, L. Fan, D.X. Zhu, Y.H. Wang, J.Y. Li, W. Xu, TP53
mutation is not an independent prognostic factor in patients with mantle cell
lymphoma at advanced stage, Med. Oncol. 29 (2012) 2166–2173.
[17] N. Vogt, W. Klapper, Variability in morphology and cell proliferation in
sequential biopsies of mantle cell lymphoma at diagnosis and relapse: clinical
correlation and insights into disease progression, Histopathology 62 (2013)
334–342.
[18] M. Todorovic, B. Balint, B. Andjelic, D. Stanisavljevic, N.K. Kurtovic, Z.
Radisavljevic, B. Mihaljevic, Outcome prediction of advanced mantle cell
lymphoma by international prognostic index versus different mantle cell
lymphoma indexes: one institution study, Med. Oncol. 29 (2012) 2212–2219.
[19] G.G. Guilbault, G. Palleschi, G. Lubrano, Non-invasive biosensors in clinical
analysis, Biosens. Bioelectron. 10 (1995) 379–392.
[20] L. Lorenzelli, A. Benvenuto, A. Adami, V. Guarnieri, B. Margesin, V. Mulloni, D.
Vincenzi, Development of a gas chromatography silicon-based microsystem
in clinical diagnostics, Biosens. Bioelectron. 20 (2005) 1968–1976.

[21] B. Buszewski, D. Grzywinski, T. Ligor, T. Stacewicz, Z. Bielecki, J. Wojtas,
Detection of volatile organic compounds as biomarkers in breath analysis by
different analytical techniques, Bioanalysis 5 (2013) 2287–2306.
[22] T. Niwa, Metabolic profiling with gas chromatography-mass spectrometry
and its application to clinical medicine, J. Chromatogr. 379 (1986) 313–345.
[23] S.B. Milne, T.P. Mathews, D.S. Myers, P.T. Ivanova, H.A. Brown, Sum of the
parts: mass spectrometry-based metabolomics, Biochemistry 52 (2013)
3829–3840.
[24] M. Phillips, R.N. Cataneo, B.A. Ditkoff, P. Fisher, J. Greenberg, R. Gunawardena,
C.S. Kwon, O. Tietje, C. Wong, Prediction of breast cancer using volatile
biomarkers in the breath, Breast Cancer Res. Treat. 99 (2006) 19–21.
[25] M. Phillips, N. Altorki, J.H. Austin, R.B. Cameron, R.N. Cataneo, J. Greenberg, R.
Kloss, R.A. Maxfield, M.I. Munawar, H.I. Pass, A. Rashid, W.N. Rom, P. Schmitt,
Prediction of lung cancer using volatile biomarkers in breath, Cancer Biomark.
3 (2007) 95–109.
[26] M. Phillips, N. Altorki, J.H. Austin, R.B. Cameron, R.N. Cataneo, R. Kloss, R.A.
Maxfield, M.I. Munawar, H.I. Pass, A. Rashid, W.N. Rom, P. Schmitt, J. Wai,
Detection of lung cancer using weighted digital analysis of breath biomarkers,
Clin. Chim. Acta 393 (2008) 76–84.
[27] H. Wu, R. Xue, C. Lu, C. Deng, T. Liu, H. Zeng, Q. Wang, X. Shen, Metabolomic
study for diagnostic model of oesophageal cancer using gas
chromatography/mass spectrometry, J. Chromatogr. B. Analyt. Technol.
Biomed. Life Sci. 877 (2009) 3111–3117.
[28] A. Ikeda, S. Nishiumi, M. Shinohara, T. Yoshie, N. Hatano, T. Okuno, T. Bamba,
E. Fukusaki, T. Takenawa, T. Azuma, M. Yoshida, Serum metabolomics as a
novel diagnostic approach for gastrointestinal cancer, Biomed. Chromatogr.
26 (2012) 548–558.
[29] H. Song, J.S. Peng, Y. Dong-Sheng, Z.L. Yang, H.L. Liu, Y.K. Zeng, X.P. Shi, B.Y. Lu,
Serum metabolic profiling of human gastric cancer based on gas
chromatography/mass spectrometry, Braz. J. Med. Biol. Res. 45 (2012) 78–85.
[30] T. Kobayashi, S. Nishiumi, A. Ikeda, T. Yoshie, A. Sakai, A. Matsubara, Y. Izumi,
H. Tsumura, M. Tsuda, H. Nishisaki, N. Hayashi, S. Kawano, Y. Fujiwara, H.
Minami, T. Takenawa, T. Azuma, M. Yoshida, A novel serum
metabolomics-based diagnostic approach to pancreatic cancer, Cancer
Epidemiol. Biomark. Prev. 22 (2013) 571–579.
[31] G.M. Hayes, R. Busch, J. Voogt, I.M. Siah, T.A. Gee, M.K. Hellerstein, N.
Chiorazzi, K.R. Rai, E.J. Murphy, Isolation of malignant B cells from patients
with chronic lymphocytic leukemia (CLL) for analysis of cell proliferation:
validation of a simplified method suitable for multi-center clinical studies,
Leuk. Res. 34 (2010) 809–815.
[32] A.B. Kanu, H.H. Hill Jr., Ion mobility spectrometry detection for gas
chromatography, J. Chromatogr. A. 1177 (2008) 12–27.
[33] T.P. Conrads, M. Zhou, E.F. Petricoin 3rd, L. Liotta, T.D. Veenstra, Cancer
diagnosis using proteomic patterns, Expert Rev. Mol. Diagn. 3 (2003) 411–420.
[34] S.F. Mitsuma, M.K. Mansour, J.P. Dekker, J. Kim, M.Z. Rahman, A. Tweed-Kent,
P. Schuetz, Promising new assays and technologies for the diagnosis and
management of infectious diseases, Clin. Infect. Dis. 56 (2013) 996–1002.
[35] J.G. Wilkes, F. Rafii, J.B. Sutherland, L.G. Rushing, D.A. Buzatu, Pyrolysis mass
spectrometry for distinguishing potential hoax materials from bioterror
agents, Rapid Commun. Mass Spectrom. 20 (2006) 2383–2386.
[36] W. Cheung, Y. Xu, C.L. Thomas, R. Goodacre, Discrimination of bacteria using
pyrolysis-gas chromatography-differential mobility spectrometry
(Py-GC-DMS) and chemometrics, Analyst 134 (2009) 557–563.
[37] M.W. Maswadeh , S.A. Peters. Multivariable and multigroup Receiver
Operating Characteristics curve analyses for qualitative and quantitative
analysis. http://www.dtic.mil/dtic/tr/fulltext/u2/a554489.pdf 2012 (accessed
29.04.15).
[38] National Cancer Institute: http://www.cancer.gov/ (accessed 29.04.15).
[39] K.S. Suh, Discovering Biomarkers to Predict Clinical Outcome. In: Third annual
Trends in Cancer Diagnostics: Early Detection and Optimum Therapy, August
24–25, (2011).
[40] G. Chiavari, S. Prati, Analytical pyrolysis as diagnostic tool in the investigation
of works of art, Chromatographia 58 (2003) 543–554.
[41] E. Ribechini, J. Pérez-Arantegui, M.P. Colombini, Gas chromatography/mass
spectrometry and pyrolysis-gas chromatography/mass spectrometry for the
chemical characterisation of modern and archaeological figs (Ficus carica), J.
Chromatogr. A 1218 (2011) 3915–3922.
[42] A.S. Le Hô, M. Regert, O. Marescot, C. Duhamel, J. Langlois, T. Miyakoshi, C.
Genty, M. Sablier, Molecular criteria for discriminating museum Asian
lacquerware from different vegetal origins by pyrolysis gas
chromatography/mass spectrometry, Anal. Chim. Acta 710 (2012) 9–16.
[43] S.M. Huff, H.L.C. Meuzelaar, D.L. Pope, C.R. Kjeldsberg, Characterization of
leukemic and normal white blood cells by Curie-point pyrolysis-mass
spectrometry: I. numerical evaluation of the results of a pilot study, J. Anal.
Appl. Pyrolysis 3 (1981) 95–109.
[44] M. Gunasekaran, W.T. Hughes, R.B. Wilber, Rapid diagnosis of systemic
candidiasis in children with cancer by pyrolysis gas liquid chromatography,
Mycopathologia 84 (1983) 17–19.
[45] M.A. Sugden, H.M. Talbot, P. Farrimond, Flash pyrolysis—a rapid method for
screening bacterial species for the presence of bacteriohopanepolyols, Org.
Geochem. 36 (2005) 975–979.
[46] K. Urakami, A. Higashi, K. Umemoto, M. Godo, C. Watanabe, K. Hashimoto,
Compositional analysis of copoly (dl-lactic/glycolic acid) (PLGA) by
pyrolysis-gas chromatography/mass spectrometry combined with one-step
thermally assisted hydrolysis and methylation in the presence of
tetramethylammonium hydroxide, Chem. Pharm. Bull. 49 (2001) 203–205.
[47] A. Dzierzega-Lecznar, S. Kurkiewicz, K. Stepien, E. Chodurek, P. Riederer, M.
Gerlach, Structural investigations of neuromelanin by pyrolysis-gas
chromatography/mass spectrometry, J. Neural. Transm. 113 (2006) 729–734.
[48] M. Westhoff, P. Litterst, L. Freitag, W. Urfer, S. Bader, J.I. Baumbach, Ion
mobility spectrometry for the detection of volatile organic compounds in
exhaled breath of patients with lung cancer: results of a pilot study, Thorax
64 (2009) 744–748.
[49] P. Jares, E. Campo, Advances in the understanding of mantle cell lymphoma,
Br. J. Haematol. 142 (2008) 149–165.
[50] M. Dreyling, V. Amador, M. Callanan, M. Jerkeman, S. Le Gouill, C. Pott, S. Rule,
F. Zaja, Update on the molecular pathogenesis and targeted approaches of
mantle cell lymphoma: summary of the 12th annual conference of the
European Mantle Cell Lymphoma Network, Leuk. Lymphoma. 56 (2015)
866–876.
[51] J.M. Vose, Mantle cell lymphoma: 2013 Update on diagnosis,
risk-stratification, and clinical management, Am. J. Hematol. 88 (2013)
1082–1088.
[52] M. Dreyling, S. Ferrero, N. Vogt, W. Klapper, European Mantle Cell Lymphoma
Network, New paradigms in mantle cell lymphoma: is it time to risk-stratify
treatment based on the proliferative signature? Clin. Cancer Res. 20 (2014)
5194–5206.
[53] A. Mozos, C. Royo, E. Hartmann, D. De Jong, C. Baro, A. Valera, K. Fu, D.D.
Weisenburger, J. Delabie, S.S. Chuang, E.S. Jaffe, C. Ruiz-Marcellan, S. Dave, L.
Rimsza, R. Braziel, R.D. Gascoyne, F. Solé, A. López-Guillermo, D. Colomer, L.M.
Staudt, A. Rosenwald, G. Ott, P. Jares, E. Campo, SOX11 expression is highly
specific for mantle cell lymphoma and identifies the cyclin D1-negative
subtype, Haematologica 94 (2009) 1555–1562.
[54] J.A. Westerhuis, H.C.J. Hoefsloot, S. Smit, D.J. Vis, A.K. Smilde, E.J.J. van Velzen,
J.P.M. van Duijnhoven, F.A. van Dorsten, Assessment of PLSDA cross
validation, Metabolomics 4 (2008) 81–89.
[55] D.I. Broadhurst, D.B. Kell, Statistical strategies for avoiding false discoveries in
metabolomics and related experiments, Metabolomics 2 (2006) 171–196.

Publication - AC Birox

Recommended

Recommended

More Related Content

What's hot

What's hot (17)

Similar to Publication - AC Birox

Similar to Publication - AC Birox (20)

Publication - AC Birox