The authors have declared that no competing interests exist.
Conceived and designed the experiments: GVCF DS AB MS JWM RB PAK BM RTN WRM. Performed the experiments: DS AB MS PAK CHB BM WRM. Analyzed the data: GVCF AM ZH NO MT DL RB RTN. Contributed reagents/materials/analysis tools: GVCF DS AB MS JWM CHB WRM. Wrote the paper: GVCF AM KKYL BM RTN WRM.
¶ RTN and WRM are joint senior authors on this work.
Recent technical advances in the field of quantitative proteomics have stimulated a large number of biomarker discovery studies of various diseases, providing avenues for new treatments and diagnostics. However, inherent challenges have limited the successful translation of candidate biomarkers into clinical use, thus highlighting the need for a robust analytical methodology to transition from biomarker discovery to clinical implementation. We have developed an end-to-end computational proteomic pipeline for biomarkers studies. At the discovery stage, the pipeline emphasizes different aspects of experimental design, appropriate statistical methodologies, and quality assessment of results. At the validation stage, the pipeline focuses on the migration of the results to a platform appropriate for external validation, and the development of a classifier score based on corroborated protein biomarkers. At the last stage towards clinical implementation, the main aims are to develop and validate an assay suitable for clinical deployment, and to calibrate the biomarker classifier using the developed assay. The proposed pipeline was applied to a biomarker study in cardiac transplantation aimed at developing a minimally invasive clinical test to monitor acute rejection. Starting with an untargeted screening of the human plasma proteome, five candidate biomarker proteins were identified. Rejection-regulated proteins reflect cellular and humoral immune responses, acute phase inflammatory pathways, and lipid metabolism biological processes. A multiplex multiple reaction monitoring mass-spectrometry (MRM-MS) assay was developed for the five candidate biomarkers and validated by enzyme-linked immune-sorbent (ELISA) and immunonephelometric assays (INA). A classifier score based on corroborated proteins demonstrated that the developed MRM-MS assay provides an appropriate methodology for an external validation, which is still in progress. Plasma proteomic biomarkers of acute cardiac rejection may offer a relevant post-transplant monitoring tool to effectively guide clinical care. The proposed computational pipeline is highly applicable to a wide range of biomarker proteomic studies.
Novel proteomic technology has led to the generation of vast amounts of biological data and the identification of numerous potential biomarkers. However, computational approaches to translate this information into knowledge capable of impacting clinical care have been lagging. We propose a computational proteomic pipeline for biomarker studies that is founded on the combination of advanced statistical methodologies. We demonstrate our approach through the analysis of data obtained from heart transplant patients. Heart transplantation is the gold standard treatment for patients with end-stage heart failure, but is complicated by episodes of immune rejection that can adversely impact patient outcomes. Current rejection monitoring approaches are highly invasive, requiring a biopsy of the heart. This work aims to reduce the need for biopsies, and demonstrate the power and utility of computational approaches in proteomic biomarker discovery. Our work utilizes novel high-throughput proteomic technology combined with advanced statistical techniques to identify blood markers that guide the decision as to whether a biopsy is warranted, reduce the number of unnecessary biopsies, and ultimately diagnose the presence of rejection in heart transplant patients. Additionally, the proposed computational methodologies can be applied to a range of proteomic biomarker studies of various diseases and conditions.
After the first successful human-to-human heart transplant in 1967, cardiac transplantation became the primary therapy for patients with end-stage heart failure due to dilated cardiomyopathy or ischemic heart disease. Improvements in immunosuppressive drug therapies have significantly increased the number of successful transplants, yet episodes of acute rejection and progression of chronic rejection remain major factors that negatively impact long term graft survival. Acute rejection is predominantly considered to be an immunological reaction in response to the major and minor histocompatibility antigens recognized as ‘foreign’ by the graft recipient. This process triggers the subsequent activation, migration and infiltration of immune cells such as T- and B-lymphocytes, which can ultimately lead to cellular- and antibody-mediated destruction of the heart allograft tissue
Since proteins may serve as molecular indicators (i.e., biomarkers) of cardiac allograft rejection, plasma proteomics offers an attractive and promising avenue for the development of diagnosis tools for cardiac transplantation
Computational proteomics is a new and expanding field of research which primarily focuses on data management and mass-spectra analysis for the discovery phase of biomarker studies
We complement previous work by proposing a computation pipeline powered by extensive statistical analysis for all stages of quantitative proteomics biomarker studies (
The 3-stage computational pipeline enables an initial untargeted exploration of the plasma proteome resulting in a list of potential biomarkers, followed by the validation of a set of candidate biomarkers that emphasizes the combination of candidate protein biomarkers into a classifier score with clinical utility. The bottom panel outlines the main steps of the computational pipeline that provide a systematic process from discovery to validation to clinical implementation of plasma protein biomarkers.
We demonstrate the power of our methodology in a proteomic biomarker study in the context of cardiac transplantation, with a goal towards the development of a more accurate and less invasive blood test for monitoring graft survival. Our work identified a panel of five candidate plasma proteins that clearly discriminates acute cardiac allograft rejection from non-rejection. These biomarker proteins distribute broadly among three relevant biological processes: cellular and humoral immune responses, acute phase inflammatory pathways and lipid metabolism. Of the five candidate biomarkers, we corroborated four using two independent platforms. A classifier score based on these four corroborated proteins measured by multiple reaction monitoring mass-spectrometry (MRM-MS) demonstrated that plasma protein biomarkers have significant potential in serving as a reliable, minimally-invasive, inexpensive, and timely diagnostic tool for acute cardiac allograft rejection. Our results advance the approaches to diagnosis with respect to cardiac transplantation biomarker, as well as the computational methodologies tailored for a wide range of proteomic biomarker studies.
A synopsis of the computational pipeline proposed in this study is illustrated in
Recent technological advances in quantitative proteomics have enabled the untargeted quantitation and identification of hundreds to thousands of proteins simultaneously from complex samples such as human plasma. The aim of the discovery pipeline is to create a list of candidate markers from an extensive set of proteins identified and measured within each sample.
The first step of the discovery pipeline is to assemble the data generated by (untargeted) quantitative proteomics approaches to perform further statistical analysis. As broadly reviewed by Nesvizhskii and Aebersold
Because for many cases there is insufficient evidence to determine which protein(s) from each group was originally present in the analyzed samples, a comprehensive exploration of the data needs to link and compare protein groups, instead of single protein identities, across multiple experimental runs. Thus, we used an algorithm, called Protein Group Code Algorithm (PGCA), to pre-process protein summaries organized using protein groups (manuscript in preparation). PGCA creates global protein groups from connected groups identified across the different runs and assigns a protein group code (PGC) to each global group. Using this PGC to link groups across multiple runs enables the analysis of interesting proteins that are identified within groups with an unstable composition across runs (further details are given as supporting materials in
Motivated by the diagnosis study in cardiac transplantation, this study was designed to ensure a rigorous case-control analysis at the event time. Thus, the discovery cohort (training set) was constructed by selecting one sample per patient to maintain usual assumptions of independence between samples in statistical tests. Samples in the case and control groups were matched 1 to 2 by time post-transplant and, where possible, age and gender. A power calculation was used to determine the number of samples statistically needed within the case and the control groups (
A common observation in most shotgun proteomic studies is that, even when the same sample is processed multiple times, not all peptides are detected in every experimental run (undersampling)
An important step of the computational analysis is the selection of candidate protein markers through univariate statistical tests. It is important to note that low sample variance estimates associated with small sample sizes can increase the number of false positives in many classical tests. Thus, when the study sample size is small, we recommend the use of an empirical Bayes method (eBayes) that estimates protein-specific variances by pooling information from other proteins to construct moderated
Another step of fundamental importance in any discovery study is quality assessment of the identified markers. Several quality control parameters were examined in accordance with guidelines for proteomic data publication
The quality of the results was also assessed from a statistical perspective, by examining the potential existence of important confounding factors. As previously demonstrated by Culhane and Quackenbush, the signature of the identified panel may be influenced by the experimental design
As some statistical methodologies do not allow the presence of missing values, the imputation of missing values can be a critical step of the computational pipeline. We used a k-nearest neighbor (kNN) approach
The relative levels of the candidate biomarkers were combined into a single classifier score generated by a Linear Discriminant Analysis (LDA) classifier to demonstrate the joint performance of the identified candidate markers. As LDA seeks a linear combination of markers that best discriminates both groups
Although the GlobalAncova analysis performs a simultaneous global assessment for all candidate PGCs, it does not test the influence of potential confounders in the aggregated classifier score. Thus, we also examined the potential effect of the confounding factors on the LDA score by looking at the Pearson correlation between the score and the potential confounders.
As widely discussed in the literature, any list of candidate markers identified in a discovery stage must be validated in a large and independent cohort of patients before its clinical utility assessment. To bridge the gap between discovery and clinical technologies, the validation stage is usually performed in an independent platform which provides a timely and cost-effective approach to measure all samples. To overcome the dependence on antibody availability, we developed an MRM-MS assay to complete the validation stage. However, similar analytical steps would have been taken if another independent platform was used.
The first step is to corroborate that the results from the discovery are successfully translated to the proteomic technologies required in the validation stage. This technical corroboration was first examined by a correlation analysis among the protein levels in common samples measured by different proteomic platforms. To control the influence of outliers in the results, these correlations were estimated using the Spearman correlation coefficient.
The second analytical step is to corroborate that the previously identified differentiation of protein levels between case and control samples is still present when the candidate markers are measured using the new platform(s). The number of samples required for this corroboration needs to be determined based on the estimated variation associated with the new platform (
Despite the general consensus that a panel of biomarkers will be required to classify new samples in a clinical setting, a fundamental analytical step often neglected during the validation stage is the examination of the complementary contribution of each candidate marker to classify new samples
In this study, different classifiers were built using MRM-MS measurements that sequentially incorporate the corroborated proteins to evaluate their complementary contribution to the classification performance. Although these multivariate classifiers were constructed by Linear Discriminant Analysis (LDA) in our biomarker study, alternative methodologies might be considered, including Support Vector Machines, Elastic Net logistic regression analysis, and Random Forests, among others
The examination of the contribution of each biomarker to the classifier performance can be used to select a final classifier (i.e., a final biomarker panel with the corresponding model) to be tested on an external cohort of patients. Since an independent test cohort was not available at this point, all classification performance measures of the panel were estimated by a stratified 6-fold cross-validation (more details are given as supporting material in
In the last step of the validation stage, the classifier score must be validated in a large external cohort of patients. Common performance measures include sensitivity, specificity, and area under the receiver operating curve (AUC)
The final translation of proteomic results from the validation to the clinical implementation stage requires careful examination of many factors, including the development of assays suitable for clinical laboratories, considerations from health economics, as well as approval of regulatory agencies (e.g., Food and Drug Administration, Conformité Européenne mark)
A critical next step towards clinical implementation is to develop and validate an assay suitable for clinical deployment according to Clinical Laboratory and Standards Institute's protocols to measure the proteins in the identified panel.
After the assay migration step, the biomarker
Lastly, the classifier needs to be tested in an independent and large cohort with clinical complexity. Different performance measures may be evaluated and emphasized depending on the “fit-to-purpose” of the study, including sensitivity, specificity, negative and/or positive predictive value.
A brief summary of the materials and methods used in the proteomic biomarker study of cardiac transplantation are outlined here and further details are given as supporting material in
This study was approved by the Human Research Ethics Board of the University of British Columbia. All patients enrolled in this study signed consent forms.
A prospective, longitudinal study, approved by the Human Research Ethics Board of the University of British Columbia, was conducted on 63 patients, with signed consent, who received a cardiac transplant at St. Paul's Hospital, Vancouver, British Columbia between March 2005 and February 2008. Of these 63 patients, 44 were included in the acute rejection cohort and were thus part of this study. Patient demographic characteristics are summarized in
Pre-transplant and protocol heart tissue biopsies were blindedly reviewed by multiple cardiac pathologists and classified according to the current International Society for Heart and Lung Transplantation (ISHLT) grading scale
An important component of the experimental design for an untargeted platform like iTRAQ is the choice of the reference sample used to ensure interpretable results across different runs
All blood samples were processed following rigorously defined standard operating procedures
Four out of five markers in the panel were assayed either by enzyme-linked immune-sorbent assay (ELISA), for adiponectin (ADIPOQ, R&D Systems, Minneapolis, MN), factor X (FX, Diapharma, West Chester, OH), and β2-microglobulin (B2M, standard clinical laboratory), or by immunonephelometric assay (INA) for serum ceruloplasmin (CP, standard clinical laboratory). ELISA/INA assays were not available for phospholipid transfer protein precursor (PLTP). The same pooled plasma control used for iTRAQ, and patient plasma samples were assayed in duplicate or triplicate. An analogous pooled serum control sample was used for CP. Data from ELISA kits was analyzed on a VersaMax Tunable Microplate Reader (Molecular Devices, Sunnyvale, CA).
A multiplex MRM-MS assay was developed for the 5 proteins that constitute the cardiac biomarker panel. MRM-MS ion pairs for 16 selected peptides representing the 5 proteins (
A detailed description of the statistical methods was described in previous subsections. All the statistical analysis was implemented using R version 2.10.1
The first two stages of the computational pipeline, discovery and validation, were applied to a biomarker study in cardiac transplantation. An overall schematic of the number of samples, design, and proteomics data used at each stage is summarized in
Platform | Experimental Design | Cohort | Set of Proteins | |
iTRAQ | Reference design | Number of patients = 26 Number of samples = 108 (10 AR, 47 1R, 51 NR) Reference = 16 healthy | 924 PGCs, of which 43% were identified based on 2 or more peptides | |
iTRAQ | PGCs identified in at least 2/3 of case and control samples | Number of patients/samples = 20 (6 AR, 14 NR) | 127 PGCs, of which 98% were identified based on 2 or more peptides | |
iTRAQ | Case |
Number of patients/samples = 20 (6 AR, 14 NR) | 5 PGCs, of which 100% were identified based on 2 or more peptides | |
iTRAQ | Longitudinal representation | Number of patients = 26 Number of samples = 108 | Classifier score based on 5 PGCs | |
ELISA/INA | Independent samples, 25 samples (7 AR, 6 1R, 12 NR) in common with the iTRAQ samples | Number of patients/samples = 43 (13 AR, 12 1R, 18 NR) Reference = 16 healthy | 4 proteins available in ELISA or INA | |
MRM-MS | Independent samples, 23 samples (7 AR, 6 1R, 11 NR) in common with the iTRAQ and ELISA samples | Number of patients/samples = 23 (7 AR, 6 1R, 11 NR) Reference = 16 healthy | 5 proteins, 16 peptides | |
ELISA/INA | Case |
Number of patients/samples = 30 (12 AR, 18 NR) | Classifier score based on 4 corroborated proteins | |
MRM-MS | Case |
Number of patients/samples = 17 (6 AR, 11 NR) | Classifier score based on 4 corroborated proteins |
Overall schematic of the cardiac transplantation study following the computational pipeline. PGC = protein group code, AR = acute rejection, 1R = mild non-treatable rejection, NR = non-rejection.
In the discovery stage, multiplexed iTRAQ-LC-MALDI-TOF/TOF mass spectrometry was used to identify and quantitate proteins from 108 depleted plasma samples representing a time course of 20 weeks from the first 26 patients enrolled (
Following the selection criteria and the power calculation described in the supporting material (
A panel of 5 PGCs was identified with significant differential relative concentrations (robust eBayes
PGC | Gene Symbol | Fold-Change | |
6 | CP | 0.002 | +1.28 |
151 | PLTP | 0.003 | −1.56 |
188 | B2M | 0.004 | +1.46 |
84 | F10 | 0.006 | +1.27 |
92 | ADIPOQ | 0.007 | −1.31 |
Quantitative results of the discovery analysis. For each protein group code (PGC), corresponding genes (Gene Symbol) of all proteins within the groups are shown in the second column.
PGC | Gene Symbol | IPI Accession | IPI Protein Name | Uniprot | Uniprot Protein Name |
6 | CP | IPI00017601.1 | Ceruloplasmin precursor | Q1L857 P00450 A5PL27 | Ceruloplasmin (Ferroxidase; CP protein) |
151 | PLTP | IPI00643034.2 | Isoform 1 of Phospholipid transfer protein precursor | Q53H91 B3KUE5 | Phospholipid transfer protein isoform a variant; Phospholipid transfer proteinPhospholipid transfer protein, isoform CRA_c |
IPI00217778.1 | Isoform 2 of Phospholipid transfer protein precursor | P55058 | Phospholipid transfer protein (Lipid transfer protein II) | ||
IPI00022733.3 | 45 kDa protein | P55058 | Phospholipid transfer protein | ||
188 | B2M | IPI00004656.2 | β2-microglobulin | P61769 | - |
IPI00796379.1 | β2-microglobulin protein | F5H6I0 | Beta-2-microglobulin | ||
IPI00868938.1 | β2-microglobulin | A6XND9 | - | ||
84 | F10 | IPI00019576.1 | Coagulation factor X precursor | P00742 Q5JVE7 | Coagulation factor X (Stuart factor; Stuart-Prower factor; Coagulation factor X, isoform CRA_a) |
IPI00552633.2 | Coagulation factor X | Q5JVE8 | - | ||
92 | ADIPOQ | IPI00020019.1 | Adiponectin precursor | A8K660 Q15848 | Adiponectin C1Q and collagen domain containing; (30 kDa adipocyte complement-related protein; Adipocyte complement-related 30 kDa protein) |
Accession numbers and protein names from the IPI database have been updated according to UniProt database. Alternative protein names are given in parenthesis.
The quality assessment of the proteomics data demonstrated a strong confidence regarding identified protein identities, wherein 98% of the 127 analyzed PGCs and all 5 PGCs candidate biomarkers were identified based on two or more peptides (
Potential Confounders | GlobalAncova |
Correlation with Score | Acute Rejection Mean (SD) | Non-Rejection Mean (SD) |
Weight (kg) | 0.008 | −0.17 | 74.38 (17.89) | 76.76 (29.88) |
Systolic blood pressure (mmHg) | 0.015 | 0.34 | 133.67 (15.71) | 122.00 (18.13) |
BUN in blood ( mmoI/L) | 0.011 | −0.43 | 14.63 (6.12) | 11.64 (5.26) |
Creatinine in blood (umoI/L) | 0.004 | −0.42 | 145.17 (53.83) | 125.86 (55.40) |
Glucose in blood (mmoI/L) | 0.036 | −0.35 | 6.50 (1.99) | 6.06 (2.15) |
Neutrophil Number in blood (xA9/L) | 0.009 | −0.02 | 6.77 (4.58) | 6.56 (4.56) |
Cyclosporine daily dose (mg) | 0.005 | −0.18 | 175.00 (161.25) | 167.86 (195.48) |
Mycophonelate Mofetil daily dose (mg) | 0.007 | −0.38 | 2250.00 (524.40) | 1821.43 (540.91) |
Prednisone daily dose (mg) | 0.013 | 0.05 | 10.83 (8.01) | 11.79 (6.08) |
Tacrolimus daily dose (mg) | 0.014 | 0.41 | 1.67 (4.08) | 4.00 (5.22) |
The GlobalAncova analysis evaluates if the panel protein levels remain significantly differentiated between the acute rejection (AR) and the non-rejection (NR) groups after adjusting for potential confounding factors. A
To illustrate the joint performance of all candidate markers to discriminate AR from NR samples, the average LDA score was calculated for all the AR samples (n = 10) and the NR samples from NR patients (n = 40) available at each time point (
Average Linear Discriminant Analysis classifier score (Classifier score) for all available acute rejection (AR) samples (pink solid point), and non-rejection (NR) samples from NR patients (green solid triangle), at each time point. The score was centered at the LDA cut-off point so that samples with positive and negative scores are classified as “rejection” or “non-rejection”, respectively. Vertical lines represent standard errors. Means and standard error bars can be used to assess differences of the score between groups at any of the studied time points. Sample sizes available at each time point are shown in the bottom table.
Linear Discriminant Analysis classifier score (Classifier score) when patients transitioned between non-rejection (NR) and acute rejection (AR) episodes. The first consecutive AR time points were averaged (AR, pink solid point) from 7 AR patients. Non-rejection samples from the same patients, before and after AR (“NR before AR” and “NR after AR”) were averaged (pink solid triangle). The average time trend for these samples is represented with a pink solid line. A control curve (dashed green line) was constructed from 9 NR patients matched to AR patients by available time points (green solid triangle). Vertical lines represent standard errors. The asterisk (*) means that the two-sided
Further results from an initial validation performed in this stage based on 88 test iTRAQ samples not included in the discovery are shown in
The results from the iTRAQ discovery analysis were corroborated and initially validated by two independent assays: ELISA/INA (available for ADIPOQ, F10, B2M, and CP), and MRM-MS (developed for ADIPOQ, F10, B2M, CP, and PLTP). Following the results of the power calculation illustrated in
Results showed good levels of correlations for B2M, ADIPOQ, and CP (r>0.6,
The differential protein levels between AR and NR samples observed in the discovery stage were successfully translated for 3 of 4 proteins measured by ELISA/INA (B2M, ADIPOQ, and CP,
Over the last two decades, the accelerating pace of technological progress has initiated a new era in the field of clinical proteomics. In particular, plasma proteomics offers a powerful tool to examine the underlying mechanisms of various diseases and opens novel avenues for biomarkers discoveries. To date, the number and quality of technical resources available for proteomic biomarker studies are well recognized. However, the development of statistical methods to address the challenges that have arisen in the field has lagged behind, dramatically reducing the pace, quality and precision of biomarker studies. An important piece of the puzzle in clinical proteomics is to distill the information contained in the very rich data generated by new proteomic technologies via a tailored computational pipeline
In this study we propose and apply a computational pipeline that provides a systematic process to analyze proteomics data related to biomarker studies. The computational steps described in this study are consistent with, and complement those described in previous technological and analytical pipelines
There are likely additional plasma protein biomarkers that were not identified by this approach. For example, additional candidate markers may be identified using a different reference sample or an alternative proteomics platform. However, initial validation results indicate that a classifier score based on 4 corroborated biomarkers can achieve a satisfactory classification of acute rejection and non-rejection samples. If validated in a larger and external cohort of patients, the identified proteomic biomarker panel can be used to develop a more accurate and minimally invasive clinical blood test to monitor allograft rejection.
The analysis may also reveal some biomarkers previously associated with unrelated disease phenotypes, or that are not linked to cardiac transplantation. In general, looking at injury controls is a good idea and ideally one would want to include such to show that the identified panel is specific for the disease of interest. However, such comparisons would require additional carefully phenotyped cohorts, analyzed with the same analytical and technological methods on the same sample source, which for many relevant injuries are difficult or impossible to obtain. The data shown in our study do not address this point and much more work needs to be done on the comparison of acute rejection with other injuries.
The complex pathobiology of acute cardiac allograft rejection is reflected in the heterogeneity of markers identified in this study. The majority of proteins identified distribute broadly among three biological processes, consistent with the current understanding and pathogenesis of acute rejection: cellular and humoral immune responses, acute phase inflammatory pathways and lipid metabolism. Our results also highlight the anticipated distinction between the plasma proteome and that observed in tissue-based discovery studies
Transplantation elicits a host immune response that encompasses both cellular and humoral immunity, which together lead to graft tissue damage, and episodes of acute and chronic rejection. B2M is a protein associated with MHC Class I histocompatibility antigens, with increased levels reflecting allograft rejection, autoimmune or lymphoproliferative diseases as a result of increased immune activation
Acute rejection resulting from cellular infiltration of the graft leads to severe local inflammation, which has systemic consequences with a concomitant increase in circulating inflammatory markers. The acute phase response to inflammatory stimuli involves the production and release of numerous plasma proteins by the liver. CP, significantly up-regulated in AR relative to NR samples in this study, is a positive acute phase reactant. It is elevated in acute and chronic inflammatory states and elevated plasma CP is also associated with increased cardiovascular disease risk
Dyslipidemia as a consequence of immunosuppressive therapy has been reported in cardiac allograft recipients, and is a risk factor for chronic rejection
A comparison between the current panel identified for the diagnosis of cardiac allograft rejection, and that of renal allograft rejection
The plasma protein markers identified in this study have the potential to be further assessed in combinatorial analyses with Biomarkers in Transplantation (BiT) genomic and metabolomic data. Notably, numerous research groups, including the BiT group, have identified potential gene expression markers of cardiac allograft rejection using microarray and qPCR analyses of peripheral and whole blood
Taken together, the panel of protein markers identified and initially validated in this study offers a fresh approach to the diagnosis of acute cardiac rejection, providing novel avenues of investigation and potential new targets for therapeutic intervention. The computational pipeline proposed and applied in this biomarker is highly applicable to a wide range of biomarker proteomic studies.
(PDF)
(PDF)
(PDF)
(PDF)
(PDF)
(PDF)
(PDF)
(PDF)
(PDF)
(XLSX)
(PDF)
(PDF)
(PDF)
(PDF)
We thank the University of Victoria Genome BC Proteomics Centre for their expertise in conducting the iTRAQ and MRM-MS experiments.