Illuminating the Dark Cancer Phosphoproteome Through a Machine-Learned Co-Regulation Map of 26,280 Phosphosites.

Mass spectrometry-based phosphoproteomics offers a comprehensive view of protein phosphorylation, but limited knowledge about the regulation and function of most phosphosites restricts our ability to extract meaningful biological insights from phosphoproteomics data. To address this, we combine machine learning and phosphoproteomic data from 1,195 tumor specimens spanning 11 cancer types to construct CoPheeMap, a network mapping the co-regulation of 26,280 phosphosites. Integrating network features from CoPheeMap into a machine learning model, CoPheeKSA, we achieve superior performance in predicting kinase-substrate associations. CoPheeKSA reveals 24,015 associations between 9,399 phosphosites and 104 serine/threonine kinases, including many unannotated phosphosites and under-studied kinases. We validate the accuracy of these predictions using experimentally determined kinase-substrate specificities. By applying CoPheeMap and CoPheeKSA to phosphosites with high computationally predicted functional significance and cancer-associated phosphosites, we demonstrate the effectiveness of these tools in systematically illuminating phosphosites of interest, revealing dysregulated signaling processes in human cancer, and identifying under-studied kinases as putative therapeutic targets.

Kinase inhibitor pulldown assay (KiP) for clinical proteomics.

Protein kinases are frequently dysregulated and/or mutated in cancer and represent essential targets for therapy. Accurate quantification is essential. For breast cancer treatment, the identification and quantification of the protein kinase ERBB2 is critical for therapeutic decisions. While immunohistochemistry (IHC) is the current clinical diagnostic approach, it is only semiquantitative. Mass spectrometry-based proteomics offers quantitative assays that, unlike IHC, can be used to accurately evaluate hundreds of kinases simultaneously. The enrichment of less abundant kinase targets for quantification, along with depletion of interfering proteins, improves sensitivity and thus promotes more effective downstream analyses. Multiple kinase inhibitors were therefore deployed as a capture matrix for kinase inhibitor pulldown (KiP) assays designed to profile the human protein kinome as broadly as possible. Optimized assays were initially evaluated in 16 patient derived xenograft models (PDX) where KiP identified multiple differentially expressed and biologically relevant kinases. From these analyses, an optimized single-shot parallel reaction monitoring (PRM) method was developed to improve quantitative fidelity. The PRM KiP approach was then reapplied to low quantities of proteins typical of yields from core needle biopsies of human cancers. The initial prototype targeting 100 kinases recapitulated intrinsic subtyping of PDX models obtained from comprehensive proteomic and transcriptomic profiling. Luminal and HER2 enriched OCT-frozen patient biopsies subsequently analyzed through KiP-PRM also clustered by subtype. Finally, stable isotope labeled peptide standards were developed to define a prototype clinical method. Data are available via ProteomeXchange with identifiers PXD044655 and PXD046169.

IDPpub: Illuminating the Dark Phosphoproteome Through PubMed Mining.

Global phosphoproteomics experiments quantify tens of thousands of phosphorylation sites. However, data interpretation is hampered by our limited knowledge on functions, biological contexts, or precipitating enzymes of the phosphosites. This study establishes a repository of phosphosites with associated evidence in biomedical abstracts, using deep learning-based natural language processing techniques. Our model for illuminating the dark phosphoproteome through PubMed mining (IDPpub) was generated by fine-tuning BioBERT, a deep learning tool for biomedical text mining. Trained using sentences containing protein substrates and phosphorylation site positions from 3000 abstracts, the IDPpub model was then used to extract phosphorylation sites from all MEDLINE abstracts. The extracted proteins were normalized to gene symbols using the National Center for Biotechnology Information gene query, and sites were mapped to human UniProt sequences using ProtMapper and mouse UniProt sequences by direct match. Precision and recall were calculated using 150 curated abstracts, and utility was assessed by analyzing the CPTAC (Clinical Proteomics Tumor Analysis Consortium) pan-cancer phosphoproteomics datasets and the PhosphoSitePlus database. Using 10-fold cross validation, pairs of correct substrates and phosphosite positions were extracted with an average precision of 0.93 and recall of 0.94. After entity normalization and site mapping to human reference sequences, an independent validation achieved a precision of 0.91 and recall of 0.77. The IDPpub repository contains 18,458 unique human phosphorylation sites with evidence sentences from 58,227 abstracts and 5918 mouse sites in 14,610 abstracts. This included evidence sentences for 1803 sites identified in CPTAC studies that are not covered by manually curated functional information in PhosphoSitePlus. Evaluation results demonstrate the potential of IDPpub as an effective biomedical text mining tool for collecting phosphosites. Moreover, the repository (http://idppub.ptmax.org), which can be automatically updated, can serve as a powerful complement to existing resources.

Deep Learning Prediction Boosts Phosphoproteomics-Based Discoveries Through Improved Phosphopeptide Identification.

Shotgun phosphoproteomics enables high-throughput analysis of phosphopeptides in biological samples. One of the primary challenges associated with this technology is the relatively low rate of phosphopeptide identification during data analysis. This limitation hampers the full realization of the potential offered by shotgun phosphoproteomics. Here we present DeepRescore2, a computational workflow that leverages deep learning-based retention time and fragment ion intensity predictions to improve phosphopeptide identification and phosphosite localization. Using a state-of-the-art computational workflow as a benchmark, DeepRescore2 increases the number of correctly identified peptide-spectrum matches by 17% in a synthetic dataset and identifies 19% to 46% more phosphopeptides in biological datasets. In a liver cancer dataset, 30% of the significantly altered phosphosites between tumor and normal tissues and 60% of the prognosis-associated phosphosites identified from DeepRescore2-processed data could not be identified based on the state-of-the-art workflow. Notably, DeepRescore2-processed data uniquely identifies EGFR hyperactivation as a new target in poor-prognosis liver cancer, which is validated experimentally. Integration of deep learning prediction in DeepRescore2 improves phosphopeptide identification and facilitates biological discoveries.

Kinase Inhibitor Pulldown Assay Identifies a Chemotherapy Response Signature in Triple-negative Breast Cancer Based on Purine-binding Proteins.

Triple-negative breast cancer (TNBC) constitutes 10%-15% of all breast tumors. The current standard of care is multiagent chemotherapy, which is effective in only a subset of patients. The original objective of this study was to deploy a mass spectrometry (MS)-based kinase inhibitor pulldown assay (KIPA) to identify kinases elevated in non-pCR (pathologic complete response) cases for therapeutic targeting. Frozen optimal cutting temperature compound-embedded core needle biopsies were obtained from 43 patients with TNBC before docetaxel- and carboplatin-based neoadjuvant chemotherapy. KIPA was applied to the native tumor lysates that were extracted from samples with high tumor content. Seven percent of all identified proteins were kinases, and none were significantly associated with lack of pCR. However, among a large population of "off-target" purine-binding proteins (PBP) identified, seven were enriched in pCR-associated samples (P < 0.01). In orthogonal mRNA-based TNBC datasets, this seven-gene "PBP signature" was associated with chemotherapy sensitivity and favorable clinical outcomes. Functional annotation demonstrated IFN gamma response, nuclear import of DNA repair proteins, and cell death associations. Comparisons with standard tandem mass tagged-based discovery proteomics performed on the same samples demonstrated that KIPA-nominated pCR biomarkers were unique to the platform. KIPA is a novel biomarker discovery tool with unexpected utility for the identification of PBPs related to cytotoxic drug response. The PBP signature has the potential to contribute to clinical trials designed to either escalate or de-escalate therapy based on pCR probability. Significance: The identification of pretreatment predictive biomarkers for pCR in response to neoadjuvant chemotherapy would advance precision treatment for TNBC. To complement standard proteogenomic discovery profiling, a KIPA was deployed and unexpectedly identified a seven-member non-kinase PBP pCR-associated signature. Individual members served diverse pathways including IFN gamma response, nuclear import of DNA repair proteins, and cell death.

Proteogenomic data and resources for pan-cancer analysis.

The National Cancer Institute's Clinical Proteomic Tumor Analysis Consortium (CPTAC) investigates tumors from a proteogenomic perspective, creating rich multi-omics datasets connecting genomic aberrations to cancer phenotypes. To facilitate pan-cancer investigations, we have generated harmonized genomic, transcriptomic, proteomic, and clinical data for >1000 tumors in 10 cohorts to create a cohesive and powerful dataset for scientific discovery. We outline efforts by the CPTAC pan-cancer working group in data harmonization, data dissemination, and computational resources for aiding biological discoveries. We also discuss challenges for multi-omics data integration and analysis, specifically the unique challenges of working with both nucleotide sequencing and mass spectrometry proteomics data.

The Breast Cancer Proteome and Precision Oncology.

The goal of precision oncology is to translate the molecular features of cancer into predictive and prognostic tests that can be used to individualize treatment leading to improved outcomes and decreased toxicity. Success for this strategy in breast cancer is exemplified by efficacy of trastuzumab in tumors overexpressing ERBB2 and endocrine therapy for tumors that are estrogen receptor positive. However, other effective treatments, including chemotherapy, immune checkpoint inhibitors, and CDK4/6 inhibitors are not associated with strong predictive biomarkers. Proteomics promises another tier of information that, when added to genomic and transcriptomic features (proteogenomics), may create new opportunities to improve both treatment precision and therapeutic hypotheses. Here, we review both mass spectrometry-based and antibody-dependent proteomics as complementary approaches. We highlight how these methods have contributed toward a more complete understanding of breast cancer and describe the potential to guide diagnosis and treatment more accurately.

Deep learning prediction boosts phosphoproteomics-based discoveries through improved phosphopeptide identification.

Shotgun phosphoproteomics enables high-throughput analysis of phosphopeptides in biological samples, but low phosphopeptide identification rate in data analysis limits the potential of this technology. Here we present DeepRescore2, a computational workflow that leverages deep learning-based retention time and fragment ion intensity predictions to improve phosphopeptide identification and phosphosite localization. Using a state-of-the-art computational workflow as a benchmark, DeepRescore2 increases the number of correctly identified peptide-spectrum matches by 17% in a synthetic dataset and identifies 19%-46% more phosphopeptides in biological datasets. In a liver cancer dataset, 30% of the significantly altered phosphosites between tumor and normal tissues and 60% of the prognosis-associated phosphosites identified from DeepRescore2-processed data could not be identified based on the state-of-the-art workflow. Notably, DeepRescore2-processed data uniquely identifies EGFR hyperactivation as a new target in poor-prognosis liver cancer, which is validated experimentally. Integration of deep learning prediction in DeepRescore2 improves phosphopeptide identification and facilitates biological discoveries.

OmicsEV: a tool for comprehensive quality evaluation of omics data tables.

SUMMARY: RNA-Seq and mass spectrometry-based studies generate omics data tables with measurements for tens of thousands of genes across all samples in a study. The success of a study relies on the quality of these data tables, which is determined by both experimental data generation and computational methods used to process raw experimental data into quantitative data tables. We present OmicsEV, an R package for the quality evaluation of omics data tables. For each data table, OmicsEV uses a series of methods to evaluate data depth, data normalization, batch effect, biological signal, platform reproducibility and multi-omics concordance, producing comprehensive visual and quantitative evaluation results that help assess the data quality of individual data tables and facilitate the identification of the optimal data processing method and parameters for the omics study under investigation. AVAILABILITY AND IMPLEMENTATION: The source code and the user manual of OmicsEV are available at https://github.com/bzhanglab/OmicsEV, and the source code is released under the GPL-3 license.

Proteogenomic Markers of Chemotherapy Resistance and Response in Triple-Negative Breast Cancer.

Microscaled proteogenomics was deployed to probe the molecular basis for differential response to neoadjuvant carboplatin and docetaxel combination chemotherapy for triple-negative breast cancer (TNBC). Proteomic analyses of pretreatment patient biopsies uniquely revealed metabolic pathways, including oxidative phosphorylation, adipogenesis, and fatty acid metabolism, that were associated with resistance. Both proteomics and transcriptomics revealed that sensitivity was marked by elevation of DNA repair, E2F targets, G2-M checkpoint, interferon-gamma signaling, and immune-checkpoint components. Proteogenomic analyses of somatic copy-number aberrations identified a resistance-associated 19q13.31-33 deletion where LIG1, POLD1, and XRCC1 are located. In orthogonal datasets, LIG1 (DNA ligase I) gene deletion and/or low mRNA expression levels were associated with lack of pathologic complete response, higher chromosomal instability index (CIN), and poor prognosis in TNBC, as well as carboplatin-selective resistance in TNBC preclinical models. Hemizygous loss of LIG1 was also associated with higher CIN and poor prognosis in other cancer types, demonstrating broader clinical implications. SIGNIFICANCE: Proteogenomic analysis of triple-negative breast tumors revealed a complex landscape of chemotherapy response associations, including a 19q13.31-33 somatic deletion encoding genes serving lagging-strand DNA synthesis (LIG1, POLD1, and XRCC1), that correlate with lack of pathologic response, carboplatin-selective resistance, and, in pan-cancer studies, poor prognosis and CIN. This article is highlighted in the In This Issue feature, p. 2483.

A proteogenomic portrait of lung squamous cell carcinoma.

Lung squamous cell carcinoma (LSCC) remains a leading cause of cancer death with few therapeutic options. We characterized the proteogenomic landscape of LSCC, providing a deeper exposition of LSCC biology with potential therapeutic implications. We identify NSD3 as an alternative driver in FGFR1-amplified tumors and low-p63 tumors overexpressing the therapeutic target survivin. SOX2 is considered undruggable, but our analyses provide rationale for exploring chromatin modifiers such as LSD1 and EZH2 to target SOX2-overexpressing tumors. Our data support complex regulation of metabolic pathways by crosstalk between post-translational modifications including ubiquitylation. Numerous immune-related proteogenomic observations suggest directions for further investigation. Proteogenomic dissection of CDKN2A mutations argue for more nuanced assessment of RB1 protein expression and phosphorylation before declaring CDK4/6 inhibition unsuccessful. Finally, triangulation between LSCC, LUAD, and HNSCC identified both unique and common therapeutic vulnerabilities. These observations and proteogenomics data resources may guide research into the biology and treatment of LSCC.

Proteogenomic insights into the biology and treatment of HPV-negative head and neck squamous cell carcinoma.

We present a proteogenomic study of 108 human papilloma virus (HPV)-negative head and neck squamous cell carcinomas (HNSCCs). Proteomic analysis systematically catalogs HNSCC-associated proteins and phosphosites, prioritizes copy number drivers, and highlights an oncogenic role for RNA processing genes. Proteomic investigation of mutual exclusivity between FAT1 truncating mutations and 11q13.3 amplifications reveals dysregulated actin dynamics as a common functional consequence. Phosphoproteomics characterizes two modes of EGFR activation, suggesting a new strategy to stratify HNSCCs based on EGFR ligand abundance for effective treatment with inhibitory EGFR monoclonal antibodies. Widespread deletion of immune modulatory genes accounts for low immune infiltration in immune-cold tumors, whereas concordant upregulation of multiple immune checkpoint proteins may underlie resistance to anti-programmed cell death protein 1 monotherapy in immune-hot tumors. Multi-omic analysis identifies three molecular subtypes with high potential for treatment with CDK inhibitors, anti-EGFR antibody therapy, and immunotherapy, respectively. Altogether, proteogenomics provides a systematic framework to inform HNSCC biology and treatment.

Proteogenomic Landscape of Breast Cancer Tumorigenesis and Targeted Therapy.

The integration of mass spectrometry-based proteomics with next-generation DNA and RNA sequencing profiles tumors more comprehensively. Here this "proteogenomics" approach was applied to 122 treatment-naive primary breast cancers accrued to preserve post-translational modifications, including protein phosphorylation and acetylation. Proteogenomics challenged standard breast cancer diagnoses, provided detailed analysis of the ERBB2 amplicon, defined tumor subsets that could benefit from immune checkpoint therapy, and allowed more accurate assessment of Rb status for prediction of CDK4/6 inhibitor responsiveness. Phosphoproteomics profiles uncovered novel associations between tumor suppressor loss and targetable kinases. Acetylproteome analysis highlighted acetylation on key nuclear proteins involved in the DNA damage response and revealed cross-talk between cytoplasmic and mitochondrial acetylation and metabolism. Our results underscore the potential of proteogenomics for clinical investigation of breast cancer through more accurate annotation of targetable pathways and biological features of this remarkably heterogeneous malignancy.

Microscaled proteogenomic methods for precision oncology.

Cancer proteogenomics promises new insights into cancer biology and treatment efficacy by integrating genomics, transcriptomics and protein profiling including modifications by mass spectrometry (MS). A critical limitation is sample input requirements that exceed many sources of clinically important material. Here we report a proteogenomics approach for core biopsies using tissue-sparing specimen processing and microscaled proteomics. As a demonstration, we analyze core needle biopsies from ERBB2 positive breast cancers before and 48-72 h after initiating neoadjuvant trastuzumab-based chemotherapy. We show greater suppression of ERBB2 protein and both ERBB2 and mTOR target phosphosite levels in cases associated with pathological complete response, and identify potential causes of treatment resistance including the absence of ERBB2 amplification, insufficient ERBB2 activity for therapeutic sensitivity despite ERBB2 amplification, and candidate resistance mechanisms including androgen receptor signaling, mucin overexpression and an inactive immune microenvironment. The clinical utility and discovery potential of proteogenomics at biopsy-scale warrants further investigation.

WebGestalt 2019: gene set analysis toolkit with revamped UIs and APIs.

WebGestalt is a popular tool for the interpretation of gene lists derived from large scale -omics studies. In the 2019 update, WebGestalt supports 12 organisms, 342 gene identifiers and 155 175 functional categories, as well as user-uploaded functional databases. To address the growing and unique need for phosphoproteomics data interpretation, we have implemented phosphosite set analysis to identify important kinases from phosphoproteomics data. We have completely redesigned result visualizations and user interfaces to improve user-friendliness and to provide multiple types of interactive and publication-ready figures. To facilitate comprehension of the enrichment results, we have implemented two methods to reduce redundancy between enriched gene sets. We introduced a web API for other applications to get data programmatically from the WebGestalt server or pass data to WebGestalt for analysis. We also wrapped the core computation into an R package called WebGestaltR for users to perform analysis locally or in third party workflows. WebGestalt can be freely accessed at http://www.webgestalt.org.

Identification and characterization of a supraclavicular brown adipose tissue in mice.

A fundamental challenge to our understanding of brown adipose tissue (BAT) is the lack of an animal model that faithfully represents human BAT. Such a model is essential for direct assessment of the function and therapeutic potential of BAT depots in humans. In human adults, most of the thermoactive BAT depots are located in the supraclavicular region of the neck, while mouse studies focus on depots located in the interscapular region of the torso. We recently discovered BAT depots that are located in a region analogous to that of human supraclavicular BAT (scBAT). Here, we report that the mouse scBAT depot has morphological characteristics of classical BAT, possesses the potential for high thermogenic activity, and expresses a gene signature that is similar to that of human scBAT. Taken together, our studies reveal a mouse BAT depot that represents human BAT and provides a unique tool for developing new translatable approaches for utilizing human scBAT.

MEF2C regulates outflow tract alignment and transcriptional control of Tdgf1.

Congenital heart defects are the most common birth defects in humans, and those that affect the proper alignment of the outflow tracts and septation of the ventricles are a highly significant cause of morbidity and mortality in infants. A late differentiating population of cardiac progenitors, referred to as the anterior second heart field (AHF), gives rise to the outflow tract and the majority of the right ventricle and provides an embryological context for understanding cardiac outflow tract alignment and membranous ventricular septal defects. However, the transcriptional pathways controlling AHF development and their roles in congenital heart defects remain incompletely elucidated. Here, we inactivated the gene encoding the transcription factor MEF2C in the AHF in mice. Loss of Mef2c function in the AHF results in a spectrum of outflow tract alignment defects ranging from overriding aorta to double-outlet right ventricle and dextro-transposition of the great arteries. We identify Tdgf1, which encodes a Nodal co-receptor (also known as Cripto), as a direct transcriptional target of MEF2C in the outflow tract via an AHF-restricted Tdgf1 enhancer. Importantly, both the MEF2C and TDGF1 genes are associated with congenital heart defects in humans. Thus, these studies establish a direct transcriptional pathway between the core cardiac transcription factor MEF2C and the human congenital heart disease gene TDGF1. Moreover, we found a range of outflow tract alignment defects resulting from a single genetic lesion, supporting the idea that AHF-derived outflow tract alignment defects may constitute an embryological spectrum rather than distinct anomalies.

Chemogenetic profiling identifies RAD17 as synthetically lethal with checkpoint kinase inhibition.

Chemical inhibitors of the checkpoint kinases have shown promise in the treatment of cancer, yet their clinical utility may be limited by a lack of molecular biomarkers to identify specific patients most likely to respond to therapy. To this end, we screened 112 known tumor suppressor genes for synthetic lethal interactions with inhibitors of the CHEK1 and CHEK2 checkpoint kinases. We identified eight interactions, including the Replication Factor C (RFC)-related protein RAD17. Clonogenic assays in RAD17 knockdown cell lines identified a substantial shift in sensitivity to checkpoint kinase inhibition (3.5-fold) as compared to RAD17 wild-type. Additional evidence for this interaction was found in a large-scale functional shRNA screen of over 100 genotyped cancer cell lines, in which CHEK1/2 mutant cell lines were unexpectedly sensitive to RAD17 knockdown. This interaction was widely conserved, as we found that RAD17 interacts strongly with checkpoint kinases in the budding yeast Saccharomyces cerevisiae. In the setting of RAD17 knockdown, CHEK1/2 inhibition was found to be synergistic with inhibition of WEE1, another pharmacologically relevant checkpoint kinase. Accumulation of the DNA damage marker γH2AX following chemical inhibition or transient knockdown of CHEK1, CHEK2 or WEE1 was magnified by knockdown of RAD17. Taken together, our data suggest that CHEK1 or WEE1 inhibitors are likely to have greater clinical efficacy in tumors with RAD17 loss-of-function.

Cell-autonomous activation of Hedgehog signaling inhibits brown adipose tissue development.

Although recent studies have shown that brown adipose tissue (BAT) arises from progenitor cells that also give rise to skeletal muscle, the developmental signals that control the formation of BAT remain largely unknown. Here, we show that brown preadipocytes possess primary cilia and can respond to Hedgehog (Hh) signaling. Furthermore, cell-autonomous activation of Hh signaling blocks early brown-preadipocyte differentiation, inhibits BAT formation in vivo, and results in replacement of neck BAT with poorly differentiated skeletal muscle. Finally, we show that Hh signaling inhibits BAT formation partially through up-regulation of chicken ovalbumin upstream promoter transcription factor II (COUP-TFII). Taken together, our studies uncover a previously unidentified role for Hh as an inhibitor of BAT development.

Checkpoint kinases regulate a global network of transcription factors in response to DNA damage.

DNA damage activates checkpoint kinases that induce several downstream events, including widespread changes in transcription. However, the specific connections between the checkpoint kinases and downstream transcription factors (TFs) are not well understood. Here, we integrate kinase mutant expression profiles, transcriptional regulatory interactions, and phosphoproteomics to map kinases and downstream TFs to transcriptional regulatory networks. Specifically, we investigate the role of the Saccharomyces cerevisiae checkpoint kinases (Mec1, Tel1, Chk1, Rad53, and Dun1) in the transcriptional response to DNA damage caused by methyl methanesulfonate. The result is a global kinase-TF regulatory network in which Mec1 and Tel1 signal through Rad53 to synergistically regulate the expression of more than 600 genes. This network involves at least nine TFs, many of which have Rad53-dependent phosphorylation sites, as regulators of checkpoint-kinase-dependent genes. We also identify a major DNA damage-induced transcriptional network that regulates stress response genes independently of the checkpoint kinases.