Simultaneous targeted and discovery-driven clinical proteotyping using hybrid-PRM/DIA.

BACKGROUND: Clinical samples are irreplaceable, and their transformation into searchable and reusable digital biobanks is critical for conducting statistically empowered retrospective and integrative research studies. Currently, mainly data-independent acquisition strategies are employed to digitize clinical sample cohorts comprehensively. However, the sensitivity of DIA is limited, which is why selected marker candidates are often additionally measured targeted by parallel reaction monitoring. METHODS: Here, we applied the recently co-developed hybrid-PRM/DIA technology as a new intelligent data acquisition strategy that allows for the comprehensive digitization of rare clinical samples at the proteotype level. Hybrid-PRM/DIA enables enhanced measurement sensitivity for a specific set of analytes of current clinical interest by the intelligent triggering of multiplexed parallel reaction monitoring (MSxPRM) in combination with the discovery-driven digitization of the clinical biospecimen using DIA. Heavy-labeled reference peptides were utilized as triggers for MSxPRM and monitoring of endogenous peptides. RESULTS: We first evaluated hybrid-PRM/DIA in a clinical context on a pool of 185 selected proteotypic peptides for tumor-associated antigens derived from 64 annotated human protein groups. We demonstrated improved reproducibility and sensitivity for the detection of endogenous peptides, even at lower concentrations near the detection limit. Up to 179 MSxPRM scans were shown not to affect the overall DIA performance. Next, we applied hybrid-PRM/DIA for the integrated digitization of biobanked melanoma samples using a set of 30 AQUA peptides against 28 biomarker candidates with relevance in molecular tumor board evaluations of melanoma patients. Within the DIA-detected approximately 6500 protein groups, the selected marker candidates such as UFO, CDK4, NF1, and PMEL could be monitored consistently and quantitatively using MSxPRM scans, providing additional confidence for supporting future clinical decision-making. CONCLUSIONS: Combining PRM and DIA measurements provides a new strategy for the sensitive and reproducible detection of protein markers from patients currently being discussed in molecular tumor boards in combination with the opportunity to discover new biomarker candidates.

Proteomic Profiling of Advanced Melanoma Patients to Predict Therapeutic Response to Anti-PD-1 Therapy.

PURPOSE: Despite high clinical need, there are no biomarkers that accurately predict the response of patients with metastatic melanoma to anti-PD-1 therapy. EXPERIMENTAL DESIGN: In this multicenter study, we applied protein depletion and enrichment methods prior to various proteomic techniques to analyze a serum discovery cohort (n = 56) and three independent serum validation cohorts (n = 80, n = 12, n = 17). Further validation analyses by literature and survival analysis followed. RESULTS: We identified several significantly regulated proteins as well as biological processes such as neutrophil degranulation, cell-substrate adhesion, and extracellular matrix organization. Analysis of the three independent serum validation cohorts confirmed the significant differences between responders (R) and nonresponders (NR) observed in the initial discovery cohort. In addition, literature-based validation highlighted 30 markers overlapping with previously published signatures. Survival analysis using the TCGA database showed that overexpression of 17 of the markers we identified correlated with lower overall survival in patients with melanoma. CONCLUSIONS: Ultimately, this multilayered serum analysis led to a potential marker signature with 10 key markers significantly altered in at least two independent serum cohorts: CRP, LYVE1, SAA2, C1RL, CFHR3, LBP, LDHB, S100A8, S100A9, and SAA1, which will serve as the basis for further investigation. In addition to patient serum, we analyzed primary melanoma tumor cells from NR and found a potential marker signature with four key markers: LAMC1, PXDN, SERPINE1, and VCAN.

Proteogenetic drug response profiling elucidates targetable vulnerabilities of myelofibrosis.

Myelofibrosis is a hematopoietic stem cell disorder belonging to the myeloproliferative neoplasms. Myelofibrosis patients frequently carry driver mutations in either JAK2 or Calreticulin (CALR) and have limited therapeutic options. Here, we integrate ex vivo drug response and proteotype analyses across myelofibrosis patient cohorts to discover targetable vulnerabilities and associated therapeutic strategies. Drug sensitivities of mutated and progenitor cells were measured in patient blood using high-content imaging and single-cell deep learning-based analyses. Integration with matched molecular profiling revealed three targetable vulnerabilities. First, CALR mutations drive BET and HDAC inhibitor sensitivity, particularly in the absence of high Ras pathway protein levels. Second, an MCM complex-high proliferative signature corresponds to advanced disease and sensitivity to drugs targeting pro-survival signaling and DNA replication. Third, homozygous CALR mutations result in high endoplasmic reticulum (ER) stress, responding to ER stressors and unfolded protein response inhibition. Overall, our integrated analyses provide a molecularly motivated roadmap for individualized myelofibrosis patient treatment.

Mapping the dynamic high-density lipoprotein synapse.

BACKGROUND AND AIMS: Heterogeneous high-density lipoprotein (HDL) particles, which can contain hundreds of proteins, affect human health and disease through dynamic molecular interactions with cell surface proteins. How HDL mediates its long-range signaling functions and interactions with various cell types is largely unknown. Due to the complexity of HDL, we hypothesize that multiple receptors engage with HDL particles resulting in condition-dependent receptor-HDL interaction clusters at the cell surface. METHODS: Here we used the mass spectrometry-based and light-controlled proximity labeling strategy LUX-MS in a discovery-driven manner to decode HDL-receptor interactions. RESULTS: Surfaceome nanoscale organization analysis of hepatocytes and endothelial cells using LUX-MS revealed that the previously known HDL-binding protein scavenger receptor B1 (SCRB1) is embedded in a cell surface protein community, which we term HDL synapse. Modulating the endothelial HDL synapse, composed of 60 proteins, by silencing individual members, showed that the HDL synapse can be assembled in the absence of SCRB1 and that the members are interlinked. The aminopeptidase N (AMPN) (also known as CD13) was identified as an HDL synapse member that directly influences HDL uptake into the primary human aortic endothelial cells (HAECs). CONCLUSIONS: Our data indicate that preformed cell surface residing protein complexes modulate HDL function and suggest new theragnostic opportunities.

Ex vivo drug response heterogeneity reveals personalized therapeutic strategies for patients with multiple myeloma.

Multiple myeloma (MM) is a plasma cell malignancy defined by complex genetics and extensive patient heterogeneity. Despite a growing arsenal of approved therapies, MM remains incurable and in need of guidelines to identify effective personalized treatments. Here, we survey the ex vivo drug and immunotherapy sensitivities across 101 bone marrow samples from 70 patients with MM using multiplexed immunofluorescence, automated microscopy and deep-learning-based single-cell phenotyping. Combined with sample-matched genetics, proteotyping and cytokine profiling, we map the molecular regulatory network of drug sensitivity, implicating the DNA repair pathway and EYA3 expression in proteasome inhibitor sensitivity and major histocompatibility complex class II expression in the response to elotuzumab. Globally, ex vivo drug sensitivity associated with bone marrow microenvironmental signatures reflecting treatment stage, clonality and inflammation. Furthermore, ex vivo drug sensitivity significantly stratified clinical treatment responses, including to immunotherapy. Taken together, our study provides molecular and actionable insights into diverse treatment strategies for patients with MM.

Integrated multi-omics reveals anaplerotic rewiring in methylmalonyl-CoA mutase deficiency.

Methylmalonic aciduria (MMA) is an inborn error of metabolism with multiple monogenic causes and a poorly understood pathogenesis, leading to the absence of effective causal treatments. Here we employ multi-layered omics profiling combined with biochemical and clinical features of individuals with MMA to reveal a molecular diagnosis for 177 out of 210 (84%) cases, the majority (148) of whom display pathogenic variants in methylmalonyl-CoA mutase (MMUT). Stratification of these data layers by disease severity shows dysregulation of the tricarboxylic acid cycle and its replenishment (anaplerosis) by glutamine. The relevance of these disturbances is evidenced by multi-organ metabolomics of a hemizygous Mmut mouse model as well as through identification of physical interactions between MMUT and glutamine anaplerotic enzymes. Using stable-isotope tracing, we find that treatment with dimethyl-oxoglutarate restores deficient tricarboxylic acid cycling. Our work highlights glutamine anaplerosis as a potential therapeutic intervention point in MMA.

Decoding Functional High-Density Lipoprotein Particle Surfaceome Interactions.

High-density lipoprotein (HDL) is a mixture of complex particles mediating reverse cholesterol transport (RCT) and several cytoprotective activities. Despite its relevance for human health, many aspects of HDL-mediated lipid trafficking and cellular signaling remain elusive at the molecular level. During HDL's journey throughout the body, its functions are mediated through interactions with cell surface receptors on different cell types. To characterize and better understand the functional interplay between HDL particles and tissue, we analyzed the surfaceome-residing receptor neighborhoods with which HDL potentially interacts. We applied a combination of chemoproteomic technologies including automated cell surface capturing (auto-CSC) and HATRIC-based ligand-receptor capturing (HATRIC-LRC) on four different cellular model systems mimicking tissues relevant for RCT. The surfaceome analysis of EA.hy926, HEPG2, foam cells, and human aortic endothelial cells (HAECs) revealed the main currently known HDL receptor scavenger receptor B1 (SCRB1), as well as 155 shared cell surface receptors representing potential HDL interaction candidates. Since vascular endothelial growth factor A (VEGF-A) was recently found as a regulatory factor of transendothelial transport of HDL, we next analyzed the VEGF-modulated surfaceome of HAEC using the auto-CSC technology. VEGF-A treatment led to the remodeling of the surfaceome of HAEC cells, including the previously reported higher surfaceome abundance of SCRB1. In total, 165 additional receptors were found on HAEC upon VEGF-A treatment representing SCRB1 co-regulated receptors potentially involved in HDL function. Using the HATRIC-LRC technology on human endothelial cells, we specifically aimed for the identification of other bona fide (co-)receptors of HDL beyond SCRB1. HATRIC-LRC enabled, next to SCRB1, the identification of the receptor tyrosine-protein kinase Mer (MERTK). Through RNA interference, we revealed its contribution to endothelial HDL binding and uptake. Furthermore, subsequent proximity ligation assays (PLAs) demonstrated the spatial vicinity of MERTK and SCRB1 on the endothelial cell surface. The data shown provide direct evidence for a complex and dynamic HDL receptome and that receptor nanoscale organization may influence binding and uptake of HDL.

Use of MS-GUIDE for identification of protein biomarkers for risk stratification of patients with prostate cancer.

BACKGROUND: Non-invasive liquid biopsies could complement current pathological nomograms for risk stratification of prostate cancer patients. Development and testing of potential liquid biopsy markers is time, resource, and cost-intensive. For most protein targets, no antibodies or ELISAs for efficient clinical cohort pre-evaluation are currently available. We reasoned that mass spectrometry-based prescreening would enable the cost-effective and rational preselection of candidates for subsequent clinical-grade ELISA development. METHODS: Using Mass Spectrometry-GUided Immunoassay DEvelopment (MS-GUIDE), we screened 48 literature-derived biomarker candidates for their potential utility in risk stratification scoring of prostate cancer patients. Parallel reaction monitoring was used to evaluate these 48 potential protein markers in a highly multiplexed fashion in a medium-sized patient cohort of 78 patients with ground-truth prostatectomy and clinical follow-up information. Clinical-grade ELISAs were then developed for two of these candidate proteins and used for significance testing in a larger, independent patient cohort of 263 patients. RESULTS: Machine learning-based analysis of the parallel reaction monitoring data of the liquid biopsies prequalified fibronectin and vitronectin as candidate biomarkers. We evaluated their predictive value for prostate cancer biochemical recurrence scoring in an independent validation cohort of 263 prostate cancer patients using clinical-grade ELISAs. The results of our prostate cancer risk stratification test were statistically significantly 10% better than results of the current gold standards PSA alone, PSA plus prostatectomy biopsy Gleason score, or the National Comprehensive Cancer Network score in prediction of recurrence. CONCLUSION: Using MS-GUIDE we identified fibronectin and vitronectin as candidate biomarkers for prostate cancer risk stratification.

Reproducible Determination of High-Density Lipoprotein Proteotypes.

High-density lipoprotein (HDL) is a heterogeneous mixture of blood-circulating multimolecular particles containing many different proteins, lipids, and RNAs. Recent advancements in mass spectrometry-based proteotype analysis show promise for the analysis of proteoforms across large patient cohorts. In order to create the required spectral libraries enabling these data-independent acquisition (DIA) strategies, HDL was isolated from the plasma of more than 300 patients with a multiplicity of physiological HDL states. HDL proteome spectral libraries consisting of 296 protein groups and more than 786 peptidoforms were established, and the performance of the DIA strategy was benchmarked for the detection of HDL proteotype differences between healthy individuals and a cohort of patients suffering from diabetes mellitus type 2 and/or coronary heart disease. Bioinformatic interrogation of the data using the generated spectral libraries showed that the DIA approach enabled robust HDL proteotype determination. HDL peptidoform analysis enabled by using spectral libraries allowed for the identification of post-translational modifications, such as in APOA1, which could affect HDL functionality. From a technical point of view, data analysis further shows that protein and peptide quantities are currently more discriminative between different HDL proteotypes than peptidoforms without further enrichment. Together, DIA-based HDL proteotyping enables the robust digitization of HDL proteotypes as a basis for the analysis of larger clinical cohorts.

Diagnostics and correction of batch effects in large-scale proteomic studies: a tutorial.

Advancements in mass spectrometry-based proteomics have enabled experiments encompassing hundreds of samples. While these large sample sets deliver much-needed statistical power, handling them introduces technical variability known as batch effects. Here, we present a step-by-step protocol for the assessment, normalization, and batch correction of proteomic data. We review established methodologies from related fields and describe solutions specific to proteomic challenges, such as ion intensity drift and missing values in quantitative feature matrices. Finally, we compile a set of techniques that enable control of batch effect adjustment quality. We provide an R package, "proBatch", containing functions required for each step of the protocol. We demonstrate the utility of this methodology on five proteomic datasets each encompassing hundreds of samples and consisting of multiple experimental designs. In conclusion, we provide guidelines and tools to make the extraction of true biological signal from large proteomic studies more robust and transparent, ultimately facilitating reliable and reproducible research in clinical proteomics and systems biology.

Mapping specificity, cleavage entropy, allosteric changes and substrates of blood proteases in a high-throughput screen.

Proteases are among the largest protein families and critical regulators of biochemical processes like apoptosis and blood coagulation. Knowledge of proteases has been expanded by the development of proteomic approaches, however, technology for multiplexed screening of proteases within native environments is currently lacking behind. Here we introduce a simple method to profile protease activity based on isolation of protease products from native lysates using a 96FASP filter, their analysis in a mass spectrometer and a custom data analysis pipeline. The method is significantly faster, cheaper, technically less demanding, easy to multiplex and produces accurate protease fingerprints. Using the blood cascade proteases as a case study, we obtain protease substrate profiles that can be used to map specificity, cleavage entropy and allosteric effects and to design protease probes. The data further show that protease substrate predictions enable the selection of potential physiological substrates for targeted validation in biochemical assays.

The Tumor Profiler Study: integrated, multi-omic, functional tumor profiling for clinical decision support.

The application and integration of molecular profiling technologies create novel opportunities for personalized medicine. Here, we introduce the Tumor Profiler Study, an observational trial combining a prospective diagnostic approach to assess the relevance of in-depth tumor profiling to support clinical decision-making with an exploratory approach to improve the biological understanding of the disease.

Standardization and harmonization of distributed multi-center proteotype analysis supporting precision medicine studies.

Cancer has no borders: Generation and analysis of molecular data across multiple centers worldwide is necessary to gain statistically significant clinical insights for the benefit of patients. Here we conceived and standardized a proteotype data generation and analysis workflow enabling distributed data generation and evaluated the quantitative data generated across laboratories of the international Cancer Moonshot consortium. Using harmonized mass spectrometry (MS) instrument platforms and standardized data acquisition procedures, we demonstrate robust, sensitive, and reproducible data generation across eleven international sites on seven consecutive days in a 24/7 operation mode. The data presented from the high-resolution MS1-based quantitative data-independent acquisition (HRMS1-DIA) workflow shows that coordinated proteotype data acquisition is feasible from clinical specimens using such standardized strategies. This work paves the way for the distributed multi-omic digitization of large clinical specimen cohorts across multiple sites as a prerequisite for turning molecular precision medicine into reality.

MassIVE.quant: a community resource of quantitative mass spectrometry-based proteomics datasets.

MassIVE.quant is a repository infrastructure and data resource for reproducible quantitative mass spectrometry-based proteomics, which is compatible with all mass spectrometry data acquisition types and computational analysis tools. A branch structure enables MassIVE.quant to systematically store raw experimental data, metadata of the experimental design, scripts of the quantitative analysis workflow, intermediate input and output files, as well as alternative reanalyses of the same dataset.

A Proteogenomic Resource Enabling Integrated Analysis of Listeria Genotype-Proteotype-Phenotype Relationships.

Listeria monocytogenes is an opportunistic foodborne pathogen responsible for listeriosis, a potentially fatal foodborne disease. Many different Listeria strains and serotypes exist, but a proteogenomic resource that bridges the gap in our molecular understanding of the relationships between the Listeria genotypes and phenotypes via proteotypes is still missing. Here, we devised a next-generation proteogenomics strategy that enables the community to rapidly proteotype Listeria strains and relate this information back to the genotype. Based on sequencing and de novo assembly of the two most commonly used Listeria model strains, EGD-e and ScottA, we established two comprehensive Listeria proteogenomic databases. A genome comparison established core- and strain-specific genes potentially responsible for virulence differences. Next, we established a DIA/SWATH-based proteotyping strategy, including a new and robust sample preparation workflow, that enables the reproducible, sensitive, and relative quantitative measurement of Listeria proteotypes. This reusable and publicly available DIA/SWATH library covers 70% of open reading frames of Listeria and represents the most extensive spectral library for Listeria proteotype analysis to date. We used these two new resources to investigate the Listeria proteotype in states mimicking the upper gastrointestinal passage. Exposure of Listeria to bile salts at 37 °C, which simulates conditions encountered in the duodenum, showed significant proteotype perturbations including an increase of FlaA, the structural protein of flagella. Given that Listeria is known to lose its flagella above 30 °C, this was an unexpected finding. The formation of flagella, which might have implications on infectivity, was validated by parallel reaction monitoring and light and scanning electron microscopy. flaA transcript levels did not change significantly upon exposure to bile salts at 37 °C, suggesting regulation at the post-transcriptional level. Together, these analyses provide a comprehensive proteogenomic resource and toolbox for the Listeria community enabling the analysis of Listeria genotype-proteotype-phenotype relationships.

Structure-function relationships of HDL in diabetes and coronary heart disease.

High-density lipoproteins (HDL) contain hundreds of lipid species and proteins and exert many potentially vasoprotective and antidiabetogenic activities on cells. To resolve structure-function-disease relationships of HDL, we characterized HDL of 51 healthy subjects and 98 patients with diabetes (T2DM), coronary heart disease (CHD), or both for protein and lipid composition, as well as functionality in 5 cell types. The integration of 40 clinical characteristics, 34 nuclear magnetic resonance (NMR) features, 182 proteins, 227 lipid species, and 12 functional read-outs by high-dimensional statistical modeling revealed, first, that CHD and T2DM are associated with different changes of HDL in size distribution, protein and lipid composition, and function. Second, different cellular functions of HDL are weakly correlated with each other and determined by different structural components. Cholesterol efflux capacity (CEC) was no proxy of other functions. Third, 3 potentially novel determinants of HDL function were identified and validated by the use of artificially reconstituted HDL, namely the sphingadienine-based sphingomyelin SM 42:3 and glycosylphosphatidylinositol-phospholipase D1 for the ability of HDL to inhibit starvation-induced apoptosis of human aortic endothelial cells and apolipoprotein F for the ability of HDL to promote maximal respiration of brown adipocytes.

HATRIC-based identification of receptors for orphan ligands.

Cellular responses depend on the interactions of extracellular ligands, such as nutrients, growth factors, or drugs, with specific cell-surface receptors. The sensitivity of these interactions to non-physiological conditions, however, makes them challenging to study using in vitro assays. Here we present HATRIC-based ligand receptor capture (HATRIC-LRC), a chemoproteomic technology that successfully identifies target receptors for orphan ligands on living cells ranging from small molecules to intact viruses. HATRIC-LRC combines a click chemistry-based, protein-centric workflow with a water-soluble catalyst to capture ligand-receptor interactions at physiological pH from as few as 1 million cells. We show HATRIC-LRC utility for general antibody target validation within the native nanoscale organization of the surfaceome, as well as receptor identification for a small molecule ligand. HATRIC-LRC further enables the identification of complex extracellular interactomes, such as the host receptor panel for influenza A virus (IAV), the causative agent of the common flu.

An integrative strategy to identify the entire protein coding potential of prokaryotic genomes by proteogenomics.

Accurate annotation of all protein-coding sequences (CDSs) is an essential prerequisite to fully exploit the rapidly growing repertoire of completely sequenced prokaryotic genomes. However, large discrepancies among the number of CDSs annotated by different resources, missed functional short open reading frames (sORFs), and overprediction of spurious ORFs represent serious limitations. Our strategy toward accurate and complete genome annotation consolidates CDSs from multiple reference annotation resources, ab initio gene prediction algorithms and in silico ORFs (a modified six-frame translation considering alternative start codons) in an integrated proteogenomics database (iPtgxDB) that covers the entire protein-coding potential of a prokaryotic genome. By extending the PeptideClassifier concept of unambiguous peptides for prokaryotes, close to 95% of the identifiable peptides imply one distinct protein, largely simplifying downstream analysis. Searching a comprehensive Bartonella henselae proteomics data set against such an iPtgxDB allowed us to unambiguously identify novel ORFs uniquely predicted by each resource, including lipoproteins, differentially expressed and membrane-localized proteins, novel start sites and wrongly annotated pseudogenes. Most novelties were confirmed by targeted, parallel reaction monitoring mass spectrometry, including unique ORFs and single amino acid variations (SAAVs) identified in a re-sequenced laboratory strain that are not present in its reference genome. We demonstrate the general applicability of our strategy for genomes with varying GC content and distinct taxonomic origin. We release iPtgxDBs for B. henselae, Bradyrhizobium diazoefficiens and Escherichia coli and the software to generate both proteogenomics search databases and integrated annotation files that can be viewed in a genome browser for any prokaryote.

Cooperative target mRNA destabilization and translation inhibition by miR-58 microRNA family in C. elegans.

In animals, microRNAs frequently form families with related sequences. The functional relevance of miRNA families and the relative contribution of family members to target repression have remained, however, largely unexplored. Here, we used the Caenorhabditis elegans miR-58 miRNA family, composed primarily of the four highly abundant members miR-58.1, miR-80, miR-81, and miR-82, as a model to investigate the redundancy of miRNA family members and their impact on target expression in an in vivo setting. We found that miR-58 family members repress largely overlapping sets of targets in a predominantly additive fashion. Progressive deletions of miR-58 family members lead to cumulative up-regulation of target protein and RNA levels. Phenotypic defects could only be observed in the family quadruple mutant, which also showed the strongest change in target protein levels. Interestingly, although the seed sequences of miR-80 and miR-58.1 differ in a single nucleotide, predicted canonical miR-80 targets were efficiently up-regulated in the mir-58.1 single mutant, indicating functional redundancy of distinct members of this miRNA family. At the aggregate level, target binding leads mainly to mRNA degradation, although we also observed some degree of translational inhibition, particularly in the single miR-58 family mutants. These results provide a framework for understanding how miRNA family members interact to regulate target mRNAs.

Ariadne's Thread: A Robust Software Solution Leading to Automated Absolute and Relative Quantification of SRM Data.

Selected reaction monitoring (SRM) MS is a highly selective and sensitive technique to quantify protein abundances in complex biological samples. To enhance the pace of SRM large studies, a validated, robust method to fully automate absolute quantification and to substitute for interactive evaluation would be valuable. To address this demand, we present Ariadne, a Matlab software. To quantify monitored targets, Ariadne exploits metadata imported from the transition lists, and targets can be filtered according to mProphet output. Signal processing and statistical learning approaches are combined to compute peptide quantifications. To robustly estimate absolute abundances, the external calibration curve method is applied, ensuring linearity over the measured dynamic range. Ariadne was benchmarked against mProphet and Skyline by comparing its quantification performance on three different dilution series, featuring either noisy/smooth traces without background or smooth traces with complex background. Results, evaluated as efficiency, linearity, accuracy, and precision of quantification, showed that Ariadne's performance is independent of data smoothness and complex background presence and that Ariadne outperforms mProphet on the noisier data set and improved 2-fold Skyline's accuracy and precision for the lowest abundant dilution with complex background. Remarkably, Ariadne could statistically distinguish from each other all different abundances, discriminating dilutions as low as 0.1 and 0.2 fmol. These results suggest that Ariadne offers reliable and automated analysis of large-scale SRM differential expression studies.