Progress on Identifying and Characterizing the Human Proteome: 2018 Metrics from the HUPO Human Proteome Project

INTRODUCTION

The Human Proteome Project (HPP) of the Human Proteome Organization (www.hupo.org) has provided a framework for international communication, collaboration, quality assurance, data sharing, and acceleration of progress in the global proteomics community since its announcement in 2010 and launch in 2011. The HPP has two over-arching goals: (1) completing the human protein parts list tied to predicted protein-coding genes as curated by neXtProt and updated annually in this HPP Metrics publication; and (2) integrating proteomics with genomics and other omics for use throughout the biomedical/life sciences community, led by the Biology and Disease-driven HPP. The parts list starts with at least one HPP Guidelines-compliant identification of a protein product matching the predicted sequences and expands through detection and characterization of the functions of splice variants, sequence variants, post-translational modifications, and protein-protein interactions. The omics integration has been facilitated through development of targeted proteomics, including the SRMAtlas and DIA/SWATH-MS, combined with bibliometric analyses that identify popular proteins widely studied in biomedical research.

There are 50 research teams worldwide organized by chromosome, mitochondria, biological processes, and disease categories plus resource pillar groups for affinity-based protein capture, mass spectrometry, knowledge bases, and, most recently, pathology. This Perspective introduces the sixth annual HPP special issue of the Journal of Proteome Research^1–6, with emphasis on the identification of neXtProt PE2,3,4 missing proteins (MPs) led by the Chromosome-centric C-HPP.

PROGRESS ON THE HUMAN PROTEOME PARTS LIST

neXtProt release 2018–01-17 (www.nextprot.org) and PeptideAtlas v2018–01b (www.peptideatlas.org), using HPP Guidelines for Interpretation of MS Data v2.1⁷ (hupo.org/Guidelines), provided the baseline for HPP investigators and other scientists around the world to prepare manuscripts on MPs and other topics for this 2018 special issue. Full datasets and metadata are shared with the community through the ProteomeXchange Consortium repositories PRIDE, PASSEL/SRM, MassIVE, jPOST, and iProX as described in 2016⁴. ProteomeXchange⁸ as of 2018–05-15 had 5248 publicly-released data sets, of which 2165 are from human samples, up from 3496 and 1478, respectively, one year ago (http://proteomecentral.proteomexchange.org/).

neXtProt release 2018–01-17 updated its validation of PE1 proteins to 17,470, an increase of 462 from release 2017–01-23 (see Table 1 for definitions and data). In 2013–2014 we excluded the “uncertain/dubious” genes of PE5 in our denominator of predicted proteins to be detected³. Thus, the number of missing proteins (MP = PE2+3+4) has been reduced to 2186 from 2579 a year ago, 2949 two years ago, and 5511 in 2012. The PE1 proteins now constitute 89% of the total PE 1,2,3,4 predicted proteins.

Table 1.

neXtProt protein existence evidence levels from 2012 to 2018 showing progress in identifying PE1 proteins and PeptideAtlas canonical proteins. More stringent guidelines imposed in 2016.

PE Level	Feb 2012	Sept 2013	Oct 2014	April 2016	Jan 2017	Jan 2018
1: Evidence at protein level	13,975	15,646	16,491	16,518	17,008	17,470 a
2: Evidence at transcript level	5205	3570	2647	2290	1939	1660		2186 Missing Proteinsb
3: Inferred from homology	218	187	214	565	563	452
4: Predicted	88	87	87	94	77	74
5: Uncertain or dubious	622	638	616	588	572	574
Human PeptideAtlas canonical proteins	12,509	13,377	14,928	14,629	15,173	15,798

Open in a new tab

^a

Percent of predicted proteins classified as PE1 by neXtProt = PE1/PE1+2+3+4 = 89%.

^b

Missing Proteins PE 2+3+4 = 2186, down from 2579 in neXtProt v2017-01.

Meanwhile, the number of canonical proteins in PeptideAtlas increased during 2017 by 625 to reach 15,798, or 98% of the neXtProt figure of 16,092 PE1 proteins based on MS (see Table 1).

Figure 1 provides a detailed depiction of the changes within neXtProt by PE categories during the year from release 2017–01-23 to release 2018–01-17. There were 91 proteins added to neXtProt, due to their addition to UniProtKB/Swiss-Prot; 44 were qualified as PE1, while 40 were added as PE2,3, or 4 MPs, and 7 as PE5. Among the new PE1 proteins are four sORFs with strong proteomics evidence - ASDURF (NX_L0R819), NBDY (NX_A0A0U1RRE5) and SLC35A4 and MIEF1 upstream open reading frame proteins (NX_L0R6Q1 and NX_L0R8F8); the recently characterized LINC00961-encoded SPAR polypeptide (NX_A0A1B0GVQ0); and three chimeric proteins with biological activity – IQCJ-SCHIP1 (NX_B3KU38), CSB-PGBD3 (NX_P0DP91) and BARGIN (NX_Q6ZT62). Among the new PE3 proteins is the sORF Minion (NX_A0A1B0GTQ4), recently characterized in mouse but still lacking transcriptomic or proteomic evidence in human. Further curation led to deletion of 10 PE1, 3 PE2,3,4 missing proteins, and 7 PE5 dubious entries. The net effect was expansion of neXtProt entries by 71 proteins. The biggest change by far is the movement of 431 PE2,3,4 proteins into PE1, due to additional MS evidence reflected in the large increase of canonical proteins in PeptideAtlas and other high quality data at protein level. ZNF804A, GAGE12G and CEACAM19 were previously validated as PE1 due to PPI and characterization data, but these data have been removed from UniProtKB/Swiss-Prot, resulting in their downgrade to MP status. Finally, four entries were lifted from PE5 status to PE2,3,4, while six MPs were downgraded to PE5 status; none moved from PE5 to PE1.

UniProtKB/Swiss-Prot and neXtProt integrate manually curated protein-protein interaction (PPI) data from the IntAct database⁹. These data are primarily based on Yeast Two Hybrid (Y2H) methods, supplemented by affinity purification/mass spectrometry (AP/MS), phage display, and co-immunoprecipitation. To validate a protein as PE1 based on PPI data, UniProtKB/Swiss-Prot and neXtProt use a subset of PPI data from IntAct (labeled as Gold in neXtProt). This subset is built by IntAct using a scoring system detailed at https://www.ebi.ac.uk/intact/pages/faq/faq.xhtml#4 with weighting of different kinds of experimental evidence. These criteria ensure that only experimental data are used and there are always at least two experimental datasets to reach the threshold score to trigger a PE1 upgrade. Moreover, the upgrade cannot be triggered if the protein pair only co-occurs in larger complexes as detected by AP/MS, because there must be at least one indication that the proteins are in physical contact. Currently, 530 PE1 protein validations are based on “Gold” PPI, up from 372 a year ago. Y2H methods utilize artificially-expressed bait proteins to detect protein interaction partners. Generally, this approach does not identify the tissues of expression or guide researchers to a choice of biological specimens to study; however, preys may be selected from a library of transcripts expressed in a particular tissue.

Using the neXtProt “Interactions” view, https://www.nextprot.org/entry/NX_O60479/interactions, users can look for the Gold Protein-Protein interactions for each of the proteins of interest to see the number of experiments behind each interaction. Clicking on the “evidence” button will link to the IntAct page with details about the experimental datasets used.

The pie chart (Figure 2) shows the nature of the evidence data for PE1 proteins, as well as the numbers in other categories as of neXtProt release 2018–01-17. There are 16,092 PE1 proteins identified with MS data compliant with HPP Guidelines, of which 98% are canonical in PeptideAtlas. There are 1378 additional PE1 proteins identified with other kinds of protein evidence: 99 by Edman sequencing, 176 by disease mutations, 75 from 3D structures, 530 by protein-protein interactions, 58 with Ab-based techniques, 170 from PTMs or proteolytic processing, and 270 from other biochemical studies. PE2,3,4 missing proteins are divided into those with no MS data (1798) and those with insufficient or unconfirmed MS data (388, down from 453 in 2017 and 485 in 2016), primarily due to the application in 2016 of the more stringent HPP Guidelines⁷ for accepting MS-based identifications. Nearly 100 of those excluded in 2016 have been restored as additional data have been reported and reviewed.

EXPANSION OF PEPTIDEATLAS FROM 2017 to 2018

The canonical proteins in PeptideAtlas increased from 15,173 to 15,798 during the year from v2017–01 to v2018–01. Of the 40 MS datasets that were added to the PeptideAtlas Human Build during 2017, eight provided 528 of the 625 proteins that were newly validated as canonical (Figure 3). Often these datasets and publications provided a second uniquely mapping non-nested peptide with length ≥9 amino acids, raising the corresponding protein to canonical status. In some cases, the datasets provided both peptides. The greatest contributions came from studies that enriched for proteins that had not been well-represented previously, including SUMOylated proteins, membrane proteins, and zinc finger proteins. PeptideAtlas would contain 273 fewer canonical proteins if PTM-containing peptides were excluded. PTMs are the primary focus of the HPP MS resource pillar, including an ongoing community project with a specially-prepared sample of 96 phosphopeptides. Both neXtProt and PeptideAtlas have growing content of PTMs as noted by Omenn et al⁵. A major advance is the introduction of MSFragger for ultra-fast identification of post-translational and chemical modifications of peptides (Kong et al¹⁰). An extensive review of human proteoforms has recently been published¹¹.

THE FATE OF MISSING PROTEINS NOMINATED FOR neXtProt REVIEW IN THE JPR 2017 PAPERS

The editorial for the JPR 2017 special issue¹² highlighted six papers that used a variety of promising methods to find Missing Proteins, with a total of 32 identified for validation as PE1: 15 from Li et al¹³ using Triton X-100 solubilization plus ProteoMiner hexapeptide-covered beads as an enrichment/equalization strategy for low-abundance proteins with kidney, bladder, liver, and colorectal specimens, and confirmation with PRM; 12 from Carapito et al¹⁴ on the sperm proteome with PRM confirmation; 3 from Wang et al¹⁵ using a multi-protease strategy on testis; 1 from Peng¹⁶ from the kidney phosphoproteome; 1 from Meyfour et al¹⁷ based on biological studies of a Y chromosome protein in cardiac development; plus 41 cautiously forwarded from Elguoshy et al¹⁸ using the stranded peptides approach described below. Twenty MP reported by Li et al⁸, including 5 without PRM validation that had not been counted in the editorial, are now classified as canonical in PeptideAtlas and validated as PE1 in neXtProt. Only 4 from Carapito et al¹⁴ were validated as PE1, because the review of PRM data at PeptideAtlas was incomplete, as was the consideration of immunochemistry results using antibodies obtained from HPA. From the list of proteins reported by Wang et al,¹⁵ only P0DMU9 was validated as PE1, and that was based on unrelated data for 5 proteotypic peptides not reported by Wang et al¹⁵. Beta-defensin 123 (Q8N688) may be a candidate for a policy discussion of implementation of the Guidelines exceptions clause⁷, because there are excellent spectra for all of the three potential proteotypic tryptic peptides, but two are only 8 aa in length. SIM1 from Peng et al¹⁶ was validated. TBL1Y proposed by Meyfour et al¹⁷ was validated, but based on other MS data. The Elguoshy et al¹⁸ data, which lacked PXD identifiers, could not be considered by PeptideAtlas and neXtProt, but 16 of their proposed proteins were coincidentally promoted to PE1 from other datasets.

In sum, through various paths, 43 of the 73 MP candidates recommended from the 2017 JPR special issue have to date qualified as PE1 by neXtProt. It takes a village!

With the pace in MP identification seeming to decline as the “more easily detected proteins” had been identified by conventional strategies, the C-HPP and HPP announced at the Sun Moon Lake HPP Workshop after the HUPO World Congress in Taiwan in September 2016 a “MP50 Challenge” to accelerate identification of 50 PE2,3,4 missing proteins per chromosome team over two years. While some teams are still in planning phases, work from the entire proteomics community has raised 952 MPs to PE1 status since neXtProt release 2016–01-11 (see Table 1), including the notable completion of the mitochondrial chromosome with its 15^th protein by the Italian team¹⁹. Barriers include limited temporal and spatial expression of proteins important in fetal development, disease responses, rare cell types, and difficulties to access tissues. Of course, the more stringent guidelines introduced in 2016 appropriately raised the bar for MP identification from MS.

Two new situations have arisen with regard to the standardized reanalysis of datasets by PeptideAtlas. First, as illustrated above, some labs use PRM instead of SRM for comparison of natural and synthetic peptides. PeptideAtlas, which several years ago developed PASSEL for extensive SRM data, has now begun to process PRM data. The second development was the strategy of the Chromosome 7 team (Baker et al²⁰) and the Chromosome X team (Elguoshy et al¹⁸) to identify what we have named “stranded peptides” in major databases that could be combined to make a pair of proteotypic peptides for individual missing proteins and then use the reported spectra from the original work (preferably deposited in PRIDE or other accessible repository with a PXD identifier) to compare with the spectra available in SRMAtlas²¹ for the corresponding synthetic peptide. This work will be facilitated by development of a Universal Spectrum Identifier, soon to be released by PeptideAtlas and PSI. Peptides in PeptideAtlas have been validated. There may be many stranded peptides from different studies in GPMdb, MassIVE, Proteomics DB, or other resources. There is no automated connector between GPMdb and either PeptideAtlas or neXtProt; they use different reference genomes and have other incompatible features. Thus, use of stranded peptides from GPMdb requires search within GPMdb for the original source and links to ProteomeXchange PXD identifiers and to original spectra and metadata. Making the case, including careful scrutiny of the original MS spectra, is the responsibility of authors; requests to original MS data generators seeking their re-submission of original data to ProteomeXchange may be useful in some cases.

SEEKING PE2,3,4 MISSING PROTEINS IN MAJOR PROTEIN FAMILIES

Hydrophobic proteins are estimated to account for 924 of the 2186 MPs (Zhang et al²²). Thus, techniques to extract and solubilize these proteins remain a key need. Many of these proteins are members of six major protein families, as shown in Figure 4. NeXtProt releases for 2013, 2016, 2017, and 2018 were downloaded for Chromosomes 1–22, X, Y, and mitochondria; PE2–4 MPs were sorted alphabetically according to protein families or groupings and confirmed through UniProtKB, Pfam, and GeneCards. Miscellaneous “uncharacterized proteins” were excluded. Figure 4 summarizes the progression of findings for the top six PE2,3,4 missing protein families: (i) olfactory receptors (ORs), (ii) non-OR transmembrane proteins, (iii) zinc finger proteins, (iv) homeobox proteins, (v) keratin-associated proteins, and (vi) coiled-coil proteins. Membrane proteins annotated as olfactory receptors and as other transmembrane proteins (including non-OR GPCRs, taste receptors and solute carrier proteins) are now the two most abundant PE2–4 families. Non-OR transmembrane proteins have overtaken zinc-finger proteins, as a significant number of zinc-finger proteins were re-classified as PE1 between 2017 and 2018 (see Adhikari et al²³). The only top-six protein family where no progress has been made to identify proteins by MS is ORs; taste receptors also show little progress (Adhikari et al²³). Rare ORs (OR1D2, OR3A4) have reached PE1 status through protein-protein interaction analyses (see Siddiqui et al²⁴). There are no credible OR detections in PeptideAtlas.

FINDING EVIDENCE FOR FUNCTIONAL ANNOTATION OF UNCHARACTERIZED neXtProt PE1 PROTEINS

A comprehensive understanding of the human proteome requires not just the “parts” list and their interactions, but deep knowledge of their functions in health and disease. Notably, according to neXtProt release 2018–01-17, 1937 PE1,2,3,4 proteins lack specific functional annotation, including 1260 uncharacterized PE1 proteins (uPE1) (https://tinyurl.com/upe1proteins) and 677 uncharacterized missing proteins (PE2+PE3+E4). C-HPP investigators agreed in September 2017 at the HPP meeting in Dublin to launch a project focused on characterization of the functions of proteins and proteoforms, not just stringent identification of their expression³³. Deep-dive biological studies are strongly encouraged; an example is Na et al³⁴. According to Paik et al³³, 14 C-HPP teams have committed to begin work on selected uPE1 proteins. The Chromosome 17 team is exploiting a computational approach using I-TASSER and COFACTOR algorithms for prediction of protein functions³⁵.

2018 UPDATE OF THE HUMAN PROTEIN ATLAS

At the end of 2017, the Human Protein Atlas (HPA) released version 18 (based on Ensembl version: 88.38), which included 26,009 antibodies, targeting proteins from almost 17,000 human genes (~87% of the human protein-coding genes). The HPA now presents three major atlases: The Tissue Atlas³⁶, the Cell Atlas³⁷, and a Pathology Atlas³⁸. The Tissue Atlas added data for caudate nucleus and thymus. The Cell Atlas was expanded by data from RNA sequencing of 8 cell lines and increased the panel for immunofluorescence staining to 26 cell lines, as well as introducing cleavage furrow as an annotated structure. The Pathology Atlas integrates mRNA expression levels from 17 cancer types and 8000 patients hosted by The Cancer Genome Atlas³⁹, links the expression of protein-encoding genes to the overall survival time for each patient, and complements these insights with protein level data from immunohistochemistry. Based on Kaplan-Meier analysis, elevated relative mRNA expression of 6800 genes correlated with poor prognosis in at least one of the analyzed cancer types, while elevated relative mRNA expression of about 6100 genes was linked to good prognosis.

The HPA portal also integrated the latest guidelines regarding the validation of antibodies, using “enhanced validation” criteria as defined by its International Working Group⁴⁰. A total of 10,540 antibodies in the HPA v18 (40%), targeting 6,787 human proteins, now have enhanced validation data from analyses of cells or tissues with immunocytochemistry, immunohistochemistry, and Western blots. The five validation procedures are: (i) before and after knock-down of target genes (denoted genetic validation), (ii) induced overexpression or fluorescent tagging of proteins (recombinant expression validation), (iii) comparison of staining pattern with two antibodies targeting different epitopes (independent antibody validation), (iv) antibody-free methods (orthogonal validation), and (v) relating the staining pattern and determined protein size with a capture MS method (capture MS validation). As illustrated with a set of 197 antibodies, this validation process is complex and painstaking, including recognition of significant batch-to-batch variation⁴¹.

NOTABLE THRUSTS IN THE USE OF PROTEOMICS FOR BIOLOGICAL AND DISEASE STUDIES, AN UPDATE FROM THE B/D-HPP

One of the main goals of the Biology and Disease-driven Human Proteome Project (B/D-HPP) is to reveal the molecular basis of physiological/pathological processes by identifying the driver proteins involved. To guide organ and biofluid studies, B/D HPP initiatives have been encouraged to utilize lists of “popular proteins” (highly cited proteins associated with the organ or other topic of interest) to generate functional hypotheses and to pave the way for new clinical applications of targeted proteomics. Two web tools have been developed to perform systematic bibliographic searches and rank the most cited proteins under specific topics (Lam et al⁴²; Yu KS et al⁴³). The popular protein approach has been used in two studies demonstrating the principal role of reconfiguration of one carbon metabolism in the liver during hepatocarcinogenesis⁴⁴ and creating a targeted assay to monitor B-type natriuretic peptidoforms that might be biomarkers for diagnosing and monitoring heart failure⁴⁵.

Characterization of proteoforms and PTMs is an unmet need to understand the dynamics of pathogenic processes. A novel mass spectrometry-based whole protein assay enabled quantitation of the percentage of mutant KRAS4b present in colorectal cancer tissue, and the differences on C-terminal carboxymethylation, which is critical for KRAS function⁴⁶. Understanding PTM status of drug targets and the functional implications is key to next generation therapies. Van Eyk et al⁴⁷ have shown that S-nitrosylation of GSK3B at specific residues can send the protein to the nucleus, away from its cytoplasmic location, resulting in a different repertoire of phosphorylated substrates and altering responses to drugs⁴⁷.

The detailed description of the array of peptides associated with human HLA phenotypes is of paramount importance to understand the immune system and to guide the development of next-generation vaccines and immunotherapies against autoimmunity, infectious diseases, and cancers. Mass spectrometry is the only available technology to interrogate the immunopeptidome in an accurate, systematic, unbiased manner. The Human Immunopeptidome Proteome Project (HIPP)⁴⁸ has developed the first public database of quality-controlled immunopeptidomic data generated by mass spectrometry⁴⁹. The combination of MHC isolation, peptide analysis, and exome sequencing facilitates identification of immunoglobulin neoantigens as targets for lymphoma⁵⁰ and ovarian cancer⁵¹ immunotherapy and opens new avenues for individualized immunotherapies.

One of the principal aims of the B/D-HPP is to better understand human organ physiology and pathology through comprehensive proteomic insights. To this end, eye and plasma proteomes have been updated recently. A total of 9782 non-redundant proteins (not necessarily compliant with HPP guidelines) are now in the human eye proteome database of 11 tissue compartments plus biofluids, with the highest number (6538) from vitreous humor and the lowest (827) from aqueous humor⁵². More than 122,000 peptide sequences matching to 3509 protein identifications compliant with the HPP guidelines are described in the 2017 Plasma Peptide Atlas⁵³.

The B/D HPP community has sought to identify protein biomarkers in multiple clinical fields, including Pediatrics⁵⁴ and Rheumatoid and Autoimmune Disorders such as knee osteoarthritis⁵⁵ and osteoarticular pathologies with MS⁵⁶ or protein arrays^{57, 58}, and rheumatoid arthritis with anti-citrullinated protein antibodies⁵⁸. In-depth analyses of the synaptosomal proteome led to association of specific protein expression patterns with social behavior in patients with schizophrenias⁵⁹, through a joint C-HPP (chromosome 15) and B/D HPP (Brazilian Brain initiative) effort. Similar cooperation led to a comprehensive description of the human mitochondrial proteome under standardized protocols¹⁹, to assess the pharmacological prospects of targeting specific mitochondrial proteins to selectively kill cancer cells⁶⁰. Cancer biomarker discovery has experienced significant progress, as shown by the recent studies of breast, ovarian, and colorectal cancers published by members of the Cancer B/D-HPP and NCI CPTAC initiative^61–63. Novel DIA MS-based and proteogenomic approaches have facilitated discovery of new protein species and splice variants that can be used to improve colorectal cancer screening^{64, 65} and point to therapeutic targets in breast cancers⁶⁶. Finally, realizing that food allergy is a global health concern, the pros and cons of current analytical methods for allergenic risk assessment have been reviewed by members of the Food and Nutrition B/D-HPP initiative⁶⁷; this group also evaluated state-of-the-art proteomic and metaproteomic approaches to study host-microbiome interactions⁶⁸.

As noted above, productive interaction between HPP groups is shedding light on many relevant aspects of human biology. Such cooperative multi-omics efforts should guide next steps in our endeavor to generate a comprehensive human proteome map with all functional annotations needed to decipher the code of life and support future molecular precision medicine.