urPTMdb/TeaProt: Upstream and Downstream Proteomics Analysis

Abstract

We have developed the underrepresented post-translational modification (PTM) database (urPTMdb), a PTM gene set database to accelerate the discovery of enriched protein modifications in experimental data. urPTMdb provides curated lists of proteins reported to be substrates of underrepresented modifications. Their enrichment in proteomics datasets can reveal unexpected PTM regulations. urPTMdb can be implemented in existing workflows, or used in TeaProt, an online Shiny tool that integrates upstream transcription factor enrichment analysis with downstream pathway analysis through an easy-to-use interactive interface. TeaProt annotates user-uploaded data with drug–gene interactions, subcellular localizations, phenotypic functions, gene–disease associations, and enzyme–gene interactions. TeaProt enables gene set enrichment analysis (GSEA) to discover enrichments in gene sets from various resources, including MSigDB, CHEA, and urPTMdb. We demonstrate the utility of urPTMdb and TeaProt through the analysis of a previously published Western diet-induced remodeling of the tongue proteome, which revealed altered cellular processes associated with energy metabolism, interferon alpha/gamma response, adipogenesis, HMGylation substrate enrichment, and transcription regulation through PPARG and CEBPA. Additionally, we analyzed the interactome of ADP-ribose glycohydrolase TARG1, a key enzyme that removes mono-ADP-ribosylation. This analysis identified an enrichment of ADP-ribosylation, ribosomal proteins, and proteins localized in the nucleoli and endoplasmic reticulum. TeaProt and urPTMdb are accessible at https://tea.coffeeprot.com/.

Introduction

The field of omics-based research has expanded tremendously due to advances in instrumentation, leading to the generation of increasing amounts of experimental data. Concurrently, advances in bioinformatics tools and databases have been essential in allowing the discovery of biological insights into diseases and experimental treatments. The introduction of the gene set enrichment analysis (GSEA)¹ allowed the widespread discovery of biologically relevant enrichment patterns in differential expression data, by comparing experimental data with predefined lists of targets known as gene sets. Collections of gene sets have been defined for biological pathways, genomic location, subcellular localization, interactions with transcription factors, and associations with tissues or diseases. Furthermore, this type of analysis is not limited to gene sets, but has been expanded to include phosphosite-specific signatures² and lipid sets.³ This flexibility with regards to the input gene set collection has led to the implementation of GSEA in numerous tools and packages, leading to over 31 000 citations of the original publication by Subramanian et al.¹ Maleki et al. provide an in-depth review of gene set analysis methods, list over 100 available tools, explain the different methods and significance assessments commonly used, and discuss the shortcomings of each approach.⁴ Surprisingly, collections of gene sets related to post-translational modifications (PTMs) are limited, despite the significant biological impacts of these modifications. Although gene set databases for some PTMs such as phosphorylation have been created,² for many PTMs discovered in recent years such gene sets are nonexistent. Furthermore, gene sets related to PTMs can be divided in those that contain the pathway enzymes versus gene sets that contain the PTM substrates. The former category (PTM pathway gene sets) contains a handful of known enzymes involved in the synthesis, conjugation, and catabolism of the modifying molecule. The latter category (PTM substrates) can contain hundreds of proteins known to receive a specific PTM, typically derived from experimental proteomics data.

A recent publication has highlighted the capabilities of mass spectrometry-based proteomics techniques to characterize a wide array of PTMs,⁵ many of which are exciting modifications with undiscovered roles in biological processes. As evidenced by PubMed query trends, various novel PTMs have been discovered in recent years, the functions and associations of which are only just starting to be studied (Figure 1A). For example, poly-ADP-ribosylation was found to prevent proteasomal degradation of proteins by inducing deubiquitylation,⁶ and lysine lactoylation is enriched on glycolytic enzymes, suggesting a role in the regulation of glycolysis.⁷ To accommodate future research into these exciting modifications, we generated 153 gene sets for 18 PTMs (Figure 1B). Our novel underrepresented PTM database, urPTMdb, was developed to accelerate ongoing and future research, as disease or experimental treatment associated with these PTMs may not be discovered if the analysis software does not take them into account. urPTMdb is freely available to be downloaded for use in commonly used GSEA workflows or can directly be used in our novel tool TeaProt. TeaProt is a novel tool suited for the functional analysis and integration of proteomics/transcriptomics data on the results of differential expression analyses (Figure 2). The TeaProt workflow is primarily focused on data annotation, functional enrichment, and gene set enrichment analyses based on curated pathways, transcription factors, and post-translational modifications.

Figure 1.

Underrepresented PTMs. (A) PubMed publications per year discussing underrepresented PTMs. (B) Overview of urPTMdb. (C) Number of gene sets per PTM included in urPTMdb. (D) Gene set Jaccard index network of urPTMdb. The Jaccard index indicates the similarity between two gene sets, where the connected nodes have an index >0.15.

Figure 2.

Overview of the TeaProt workflow. Differential expression results from external tools are entered and validated. Input data are pre-processed and annotated using information on subcellular localizations, transcription factors, and interactions with other molecules. Analyses are divided into quality control and enrichment analyses. The distribution of p-values or fold-changes (1) can be inspected prior to the other analyses. Volcano plots aid the identification of proteins that are differentially expressed (2). The annotated data are visualized in contingency tables to highlight enriched groups of proteins/transcripts present in the data (3). The gene set enrichment analyses include analyses based on MSigDB (4), CHEA (5), and the novel underrepresented PTM database, urPTMdb (6).

In some cases, the novelty of an analytical tool is not defined by the methods it uses, but by the annotations or underlying databases that were previously unavailable. For example, despite the widespread use of overrepresentation and gene set enrichment analyses in other tools, CHEA3⁸ enabled transcription factor enrichment analyses (TFEA) by combining public data from various sources (ChIP-seq, coexpression, literature) and thus generating a comprehensive novel resource. Evidently, the value of a tool depends not only on the analyses that it can perform, but also on advances in the databases that allow the discovery of biological alterations that could otherwise not be detected. To this end, urPTMdb combines proteomics search results and summarizes public proteomics-based data of underrepresented PTMs into a novel resource that is freely available and can be expanded as additional primary data are published.

Methods

PubMed Query Analysis

To identify the number of publications published per year, mentioning underrepresented PTMs of interest, we performed PubMed queries using the following terms: (lysine & Kac) | “lysine acetylation”, ADP-ribosylation, AMPylation, “citrullination”, (lysine & Kcr) | “lysine crotonylation”, “DSSylation”, “FAT10ylation”, (lysine & Kglu) | “lysine glutarylation”, “HMGylation”, (lysine & Khib) | “lysine hydroxyisobutyrylation” | “lysine 2-hydroxyisobutyrylation”, “ISGylation”, (lysine & lactoylation) | lactoyllys, “protein lipidation” | “peptide lipidation”, (lipoylation & protein), “neddylation”, (prolyl & hydroxylation) | (proline & hydroxylation), (lysine & Ksu) | “lysine succinylation”, “UFMylation”. The resulting “results by year” graphs displayed by PubMed were downloaded and combined into a single graph. For Figure 1A, data points between 2007–2021 were displayed.

urPTMdb

urPTMdb is a novel database of gene sets covering currently underrepresented PTMs. A total of 18 PTMs were selected on the basis of their presence in proteomic data repositories, whilst being underrepresented in gene set databases. Specifically, we generated gene sets for proteins modified by adenosine diphosphate ribose (ADP-ribosylation), covalently attached adenosine monophosphate (AMPylation), citrullination, deleted in split hand/split foot 1 (DSS1) conjugation (DSSylation), Human Leukocyte Antigen (HLA)-F adjacent transcript 10 (FAT10) conjugation (FAT10ylation), 3-hydroxyl-3-methylglutarylation (HMGylation), interferon stimulated gene 15 (ISG15) conjugation (ISGylation), lipidation (myristoylation and prenylation), lipoylation, lysine acetylation (Kac), lysine crotonylation (Kcr), lysine glutarylation (Kglu), lysine 2-hydroxyisobutyrylation (Khib), lysine lactoylation (LactoylLys), lysine succinylation (Ksu), neuronal precursor cell-expressed developmentally down-regulated protein 8 (NEDD8) conjugation (neddylation), prolyl hydroxylation and Ubiquitin-fold modifier 1 (UFM1) conjugation (UFMylation) (Figure 1C). These PTMs were selected because of being currently understudied (Figure 1A), an absence of relevant gene sets in existing gene set collections, and the availability of experimental proteomics data. Datasets related to the PTMs were collected from the Proteomics Identifications Database (PRIDE)⁹ and Mass Spectrometry Interactive Virtual Environment (MASSIVE). The collected datasets are shown in Supplementary Table S1. Datasets were processed in R, and the scripts to generate the gene sets have been deposited to www.github.com/JeffreyMolendijk/urPTMdb. Additionally, literature/pathway gene sets were manually created on the basis of the known enzymes involved in each modification or retrieved from MSigDB.¹ These gene sets were collated into a collection (urptmdb_latest.gmt) that is included for direct use in TeaProt or available to be downloaded from https://tea.coffeeprot.com/ and www.github.com/JeffreyMolendijk/urPTMdb. The urPTMdb file follows the conventions of the “gene matrix transposed file format” (.gmt) that is commonly used to store collections of gene sets. Following this format we report the gene set names in the first column, a description in the second column, and the relevant genes in all subsequent columns. Furthermore, species-specific human or mouse versions of urPTMdb were generated by converting gene symbols to their respective homologues using HomoloGene. Note that these databases still contain gene sets annotated with either “hs” or “mm” to indicate the species in which the experiments were performed, but the reported gene symbols are species-specific. Generating a network of gene sets, where nodes are connected on the basis of the Jaccard index (Figure 1D), indicates that gene sets of the same PTM, or of a similar type (e.g., ubiquitin-like modifiers), have significant overlap in the included proteins. A drawback of the Jaccard index is a failure to visualize similar gene sets with large differences in the number of genes contained in each set. Upon inspection of the lone citrullination gene set (PXD026238_mm_Citrullination_sites), it appears that the difference in size is cause for the low Jaccard index with other citrullination gene sets. On the basis of this finding we decided to generate another plot based on the Szymkiewicz–Simpson coefficient rather than the Jaccard index. The Jaccard index is calculated as the ratio of the intersection and the union of the two sets, whereas the Szymkiewicz–Simpson coefficient is the ratio of the intersection and the number of genes in the smaller gene set. A small gene set that is completely present in a larger second gene set is expected to get a low Jaccard index, but a high Szymkiewicz–Simpson coefficient. The gene set that was previously isolated (PXD026238_mm_Citrullination_sites) is connected to five other citrullination gene sets with a Szymkiewicz–Simpson coefficient ≥0.6. Interactive Jaccard and Szymkiewicz–Simpson networks can be inspected on the urPTMdb tab of the TeaProt web server.

TeaProt

Web Server Implementation

TeaProt is implemented as a Shiny application using the R programming language for the backend. TeaProt is deployed on the Nectar Research Cloud and Melbourne Research Cloud, utilizing hypervisors built on AMD EPYC 2 (base CPU clock speed 2.0 GHz, burst clock speed 3.35 GHz) and running Ubuntu 18.04.

Privacy and Security

Any uploaded tables are only available to be analyzed if the user is connected to the TeaProt server. No user data are retained following session termination. Secure HTTPS connections are used to transfer data to and from the TeaProt server.

Local Installation

The code required to run TeaProt is hosted at www.github.com/ryip10903/TeaProt. Creating a copy of the TeaProt allows users to run the application locally, without accessing the web server. After installing the R packages required by TeaProt, users can directly launch the Shiny application to start analyzing their own data.

Databases and Packages

TeaProt converts UniProt and ENSEMBL identifiers to gene symbols using the AnnotationDbi package in combination with the org.Hs.eg.db (human, version 3.14.0) and org.Mm.eg.db (mouse, version 3.14.0) packages. Subcellular localizations were retrieved from the Cell Atlas¹⁰ and were further annotated with ancestor localizations to reduce overly specific localizations.¹¹ Genotype–phenotype associations were downloaded from the International Mouse Phenotyping Consortium (IMPC, www.mousephenotype.org).¹² Transcription factor data were downloaded from CHEA3.⁸ Gene-disease associations were retrieved from DisGeNet (v7)¹³ and gene-enzyme annotations from BRENDA (BRENDA is available at www.brenda-enzymes.org).¹⁴ Drug–gene interactions were retrieved from the drug–gene interaction database (DGIdb).¹⁵ Gene set enrichment analyses are performed using the fgsea package (version 1.20.0).¹⁶

Documentation and Tutorial

A demo dataset of Western diet-induced remodeling of the tongue proteome is included with TeaProt.¹⁷ This dataset and a detailed tutorial are available on the “start” page of TeaProt.

Results and Discussion

Case Study 1: Analyzing Tongue Proteomics Data Using TeaProt

To illustrate the use of TeaProt, we re-analyzed a previously published dataset of the alterations in the tongue proteome induced by the Western diet enriched in carbohydrates and lipids including cholesterol.¹⁷ The differential expression results obtained by comparing the control diet with the Western diet were used in this case study and are also available as a demo dataset in TeaProt. The first step in the workflow involves validating the user-uploaded data and performing annotation based on the parameters set by the user. The aforementioned parameters include the selection of the columns corresponding to the gene identifier, p-value and fold-change. Furthermore, the species which the data are generated from and the p-value and fold-change significance cutoffs need to be specified. TeaProt visualizes the distributions of p-values and fold-changes in the uploaded data (Figure 3A,B) and generates an interactive volcano plot to highlight the proteins with the greatest changes between conditions (Figure 3C). Finally, the data are filtered and annotated with databases.

Figure 3.

Tongue proteomics analysis. Tongue proteomics data from Dutt et al. were prepared to analyze the differential expression results of Western diet vs chow diet. Distributions of (A) p-values and (B) log₂ fold changes between treatment groups. (C) Volcano plot highlighting proteins with an absolute log₂ fold change >1 and p < 0.05. (D) A Pearson’s Chi-squared test based on protein annotations (subcellular localization) indicates whether specific annotations are primarily found in significantly (p < 0.05) upregulated, downregulated, or nonsignificant (NS) proteins. Only localizations with residuals > 2 and < −2 in the upregulated group are shown. The data in the figure are colored by Pearson residuals and sized by the absolute Pearson residuals. (E) GSEA using the MSigDB Hallmark collection reveals six significantly enriched pathways (p_adj < 0.05). (F) GSEA using the CHEA3 Literature collection reveals two significantly enriched gene sets (p_adj < 0.05). (G) GSEA enrichment and volcano (H) plots for the urPTMdb set “PXD005895_mm_HMGylation_substrates”. Proteins found in the gene set are highlighted in the volcano plot.

TeaProt currently supports annotations for Homo sapiens and Mus musculus. Data are annotated with information from the drug–gene interaction database,¹⁵ Cell Atlas,¹⁰ IMPC,¹² BRENDA,¹⁴ and DisGeNet.¹³ The annotation tab visualizes the distribution of annotations from various sources.

The annotated data are analyzed to detect enrichments in the upregulated or downregulated protein targets, in accordance with user-defined significance cutoffs. Cell Atlas annotations reveal whether the altered proteins are found in specific organelles (Figure 3D, Supplementary Table S2), where the data points refer to the Pearson residuals (Residual = (Observed – Expected)/Expected^0.5) as derived from a Chi-square test of independence. In this dataset, the proteins upregulated as a result of the Western diet are enriched in proteins localized in the mitochondria, consistent with the previously described changes in energy metabolism induced by Western diets across a range of tissues.¹⁸⁻²¹

TeaProt also enables easy gene set enrichment analysis using gene sets from various sources including the MSigDB¹ to characterize the downstream cellular process regulated, and the transcription factor sets from CHEA3⁸ to characterize potential upstream transcriptional regulators. Performing the GSEA analysis using the MSigDB Hallmark collection reveals six significantly enriched gene sets related to interferons, oxidative phosphorylation, fatty acid metabolism, adipogenesis, and androgen response (Figure 3E). Furthermore, gene sets relating to the transcription factors Peroxisome proliferator- activated receptor gamma (PPARG) and CCAAT/enhancer-binding protein alpha (CEBPA) were significantly enriched in the dataset (Figure 3F). Both PPARG and CEBPA have previously been characterized to play important roles in the metabolism of lipids, confirming the ability of TeaProt to identify known up-stream regulators.²² Performing GSEA using urPTMdb gene sets reveals an enrichment of HMGylation substrates in the Western diet-fed mice, as the gene set with the lowest q-value (normalized enrichment score (NES): 2.25, adjusted p-value 1.1 × 10^–8) (Figure 3G) followed by lysine glutarylation (NES: 2.13, adjusted p-value 6.1 × 10^–6) (data not shown). HMGylation is a fascinating nonenzymatic PTM that has been validated in recent years by Wagner et al.²³ HMGylation is similar to glutarylation with significant overlap between the HMGylome and Glutarylome.²³ The authors found that proteins involved in ketogenesis and leucine degradation and the TCA cycle are among the most HMGylated proteins, suggesting that the HMG-CoA generated in these pathways reacts nonenzymatically and locally with nearby proteins.²³ These findings support the previously described enrichment of proteins related to oxidative phosphorylation and lipid metabolism (Figure 3E) and a related increase in protein HMGylation (Figure 3G). Generating a volcano plot where the HMGylation gene set is highlighted indicates that the top altered proteins are Aldob, Ech1, Cpt2, and Acadl (Figure 3H). Altogether, these enrichment results show consistent alterations in the tongue protein composition associated with lipid metabolism. Furthermore, the analysis using the urPTMdb generated a new hypothesis on the possible regulation of HMGylation following consumption of Western diet.

Case Study 2: Using urPTMdb to Identify the Effects of Underrepresented PTM Alterations

The use of urPTMdb in GSEA allows the discovery of proteome-wide PTM alterations induced by a specific intervention or experimental design, by comparisons to published proteomics datasets. Here we show an application of urPTMdb by analyzing the TARG1 tandem affinity purification proteome in olaparib treated HEK293 cells.²⁴ TARG1 is an ADP-ribose glycohydrolase that hydrolyzes mono-ADP-ribose attached to protein glutamate residues. The ARTD1/2 inhibitor olaparib was added to reduce poly-ADP-ribose (PAR) formation as a result of cell lysis-induced DNA shearing. We downloaded Supplementary Table S2 to compare the TARG1-TAP condition with the control TAP-tag condition using TeaProt and urPTMdb. The downloaded data were reprocessed by filtering by number of missing values (<6), imputation, log₂ transforming the data, and finally calculating the fold change and p-value for each protein. In the TeaProt analyses, we considered a protein with a log₂ fold change >1 and p-value <0.05 to be significantly altered. Inspecting the distributions of p-values and fold changes reveals a skewed distribution with more positive fold changes than negative fold changes, which is generally expected for the enrichment of proteins following affinity purification (Figure 4A,B). Using these aforementioned criteria, 207 out of 1039 measured proteins were increased in the TARG1-TAP condition (Figure 4C). As shown in the volcano plot, the largest fold change (10 000×) was detected for the protein TARG1.

Figure 4.

Underrepresented PTM GSEA. GSEA were performed on the TARG1 tandem affinity purification of olaparib treated HEK293 cells published by Bütepage et al. Distributions of (A) p-values and (B) log₂ fold changes between treatment groups. (C) Volcano plot highlighting proteins with a log₂ fold change >1 and p < 0.05. (D) A Pearson’s Chi-squared test based on protein annotations (subcellular localization) indicates whether specific annotations are primarily found in significantly (p < 0.05) upregulated, downregulated, or nonsignificant (NS) proteins. Only localizations with positive residuals in the upregulated group are shown. The data in the figure are colored by Pearson residuals and sized by the absolute Pearson residuals. (E) GSEA using the urPTMdb sets highlighting the top 5 significantly enriched pathways (p_adj <0.05). (F) GSEA enrichment and volcano (G) plots for the urPTMdb set “PXD028902_hs_ADPribosylation_substrates”. Proteins found in the gene set are highlighted in the volcano plot.

Approximately 86% of the uploaded proteins were annotated with one or more subcellular localizations. Inspection of the subcellular localization contingency table reveals that a relatively large number of up-regulated proteins are located in the endoplasmic reticulum (16 expected, 42 detected) (Figure 4D, Supplementary Table S3). The number of expected annotations per group (upregulated, downregulated, not significant) refers to the number of annotations we expect to find in the absence of group enrichment, assuming the same proportion of annotated proteins in each group. Out of these 42 upregulated endoplasmic reticulum (ER) proteins, 38 are ribosomal proteins. Similarly, more than the expected number of nucleolar proteins were found (17 expected, 37 detected) among the up-regulated proteins. These findings are in agreement with those found by the authors and previous publications; TARG1 was shown to be primarily located in the nucleoli and the ribosomal proteins are the main interacting partners in the TARG1 interactome.^24,25 ADP-ribosylation of ribosomal proteins is a key mechanism of inhibiting polysome assembly though eIF6 binding to ribosomes.²⁶ The authors note that mono-ADP-ribosylation of ribosomal proteins promotes protein homeostasis by altering protein synthesis and preventing toxic protein aggregation.²⁶

The TARG1-interactome showed significant enrichment of proteins with known ADP-ribosylation sites (Figure 4E) where most ADP-ribosylation substrate proteins are more abundant in the TARG1-TAP condition. The top five most positively enriched PTM sets, when sorted by normalized enrichment scores (NES), are all related to ADP-ribosylation and are consistent with the results from multiple previous studies (PXD028902, PXD024233, PXD020589). These findings are consistent with the known functions of TARG1 in mono-ADP ribosylation hydrolysis, explaining why many TARG1-interacting proteins are those with known ADP-ribosylation sites. Visualizing the enrichment results for the set “PXD028902_hs_ADPribosylation_substrates” (Figure 4F) reveals that most matched proteins have low ranks (positive fold changes). The matched significant proteins with the lowest p-values include ribosomal proteins, RNA helicases and a DNA topoisomerase (Figure 4G). Interestingly, the TARG1 interactome was previously shown to be highly enriched for RNA metabolic processes.²⁵ Additionally, PARP-1-mediated ADP-ribosylation of the RNA helicase DDX21 was shown to promote ribosomal DNA (rDNA) transcription.²⁷

Altogether, these findings illustrate the links between ribosomal proteins, RNA metabolism, and ADP-ribosylation through the TARG1 interactome. Combining TeaProt with urPTMdb confidently identified alterations in proteins known to be ADP-ribosylated in this dataset, implicating this PTM in several cellular functions.

Discussion

In summary, we present urPTMdb and TeaProt for a novel and easy-to-use online proteomics/transcriptomics analysis pipeline featuring novel underrepresented gene sets to allow the discovery of downstream cellular processes, upstream transcriptional regulation, and classes of PTMs potentially regulated by a users’ intervention. The case studies presented in this study highlight the application of these tools in gaining understanding of the biological alterations and the power of including underrepresented PTMs as a novel analysis. A potential limitation of urPTMdb is that PTM substrates are reported at the protein level, rather than for individual amino acid residues. Performing PTM gene set enrichments at the protein level makes it difficult to detect modifications of which its substrates are proteins with stable abundances regardless of the treatment. Performing residue level analyses could provide further insights into the association between altered protein abundances and a particular site, or even multiple sites on the same protein. Residue level analyses can be performed by PTM signature enrichment analysis (PTM-SEA) using the PTM signatures database (PTMsigDB), although updated gene set collections are required to include other PTMs besides phosphorylation.² Further improvements can be made to urPTMdb as new experimental underrepresented PTM data are published, including the addition of novel PTMs and the generation of consensus gene sets summarizing all studies related to a PTM. We hope that urPTMdb will allow researchers to identify currently underrepresented PTM alterations in their datasets, driving future research into these exciting new fields.

Glossary

Abbreviations

PTM

post-translational modification

Kcr

lysine crotonylation

Ksu

lysine succinylation

Khib

lysine 2-hydroxyisobutyrylation

Kglu

lysine glutarylation.

Supporting Information Available

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acs.jproteome.2c00048.

• Table S1: Details of the studies included in urPTMdb (XLSX)

• Table S2: Contingency table of subcellular localization analysis related to case study 1 (XLSX)

• Table S3: Contingency table of subcellular localization analysis related to case study 2 (XLSX)

Author Contributions

TeaProt was developed by J.M. and R.Y. urPTMdb was developed by J.M. The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript.

This work was funded by an Australian National Health and Medical Research Council Ideas Grant [Grant Number APP1184363]. B.L.P. is supported by the University of Melbourne Driving Research Momentum program.

The authors declare no competing financial interest.