The Functional Comparison of Eukaryotic Proteomes: Implications for Choosing an Appropriate Model Organism to Probe Human Biology

Abstract

Phenotypic differences between species are, in significant part, determined by their proteomic diversity. The link between proteomic and phenotypic diversity can be best understood in the context of the various pathways and biological processes in which proteins participate. While the conservation pattern for individual proteins across species is expected to follow the phylogenetic relationships among the species, the diversity patterns of individual pathways may not: certain pathways may be much more conserved among distantly related species than two closely related species, owing to the ecological histories of the species. Thus, a pathway-centric analysis of proteome conservation and diversity has important implications for the appropriate choice of a model organism when investigating specific aspects of human biology. Exploiting the complete genome sequences and protein-coding gene annotations, here we perform a comprehensive gene-set-centric analysis of proteomic diversity between humans and 54 eukaryotic organisms, resulting in a catalog of organisms that are most similar to humans in terms of specific pathways, processes, expression patterns, and diseases. We corroborate our findings using species-specific mass spectrometry data.

Our analysis provides a general framework to identify conserved and unique pathways in a group of organisms and a resource to prioritize appropriate model systems to study a specific biological system in a reference organism such as humans.

1. Introduction

Research in model organisms is a powerful tool to understand human biology. In general, proposing a model organism for a given biological process should be informed by similarities between organisms’ proteomes at various spatial, structural, and functional granularities [1]. Such comparisons enable a rational choice of the “average” organism to study a given process and extrapolate its functioning to the other organisms of the group [1]. This can reduce the number of organisms that one has to study in order to understand the molecular biology of a given process across the tree of life [2].

Broadly, the interspecies phenotypic conservation and differences are determined by the overall gene complement of a species and the spatiotemporal regulatory program of those genes. Many of the differences observed between closely related species are due to the regulation of gene expression and protein activity, probably encoded in the noncoding portion of the genome [3, 4]. Nevertheless, a significant part of the interspecies differences, especially in distant species, is due to differences in the complement of encoded proteins. Because proteins work cooperatively in functional pathways and biological processes [5], the phenotypic differences and commonality between species can be better understood via a “pathway”- or “biological process”-centric comparative analysis of genomes, as opposed to a protein-centric or protein family centric approach. However, there is a lack of comprehensive pathway-centric comparative analysis of eukaryotic proteomes.

The proteome of an organism can be approximately identified from its whole genome sequence and various in silico annotation approaches. However, the assignment of proteins in the proteome to specific tissues, organs, or any other physiologically or pathologically relevant set of proteins relies on experimental proteomics. Tandem Mass Spectrometry (MS/MS) proteomic datasets can be roughly grouped into distinct classes, defined in Box 1. However, analyzing such data is challenging because many peptide signals do not map to the reference proteome. This is either because they contain mutations, are novel peptides, or are alternative splicing products. While various strategies have been proposed to address these issues, those approaches are limited in various ways, including computational inefficiencies and errors when applied to large-scale studies [6–9].

Box 1: Contexts in Which the Proteome Is Studied.

Complete proteome:

An entire set of proteins coded in a genome of an organism, including alternative isoforms. The various terminologies below refer to various subsets of the complete proteome (or set of proteins encoced in a genome of an invidual organism):

Cellular proteome:

Proteins expressed in specific types of cells or tissues in specific environmental contexts.

Functional proteome:

Proteins for which functional information is available.

Enzymatic proteome:

Proteins with known enzymatic activity.

Receptor proteome:

Proteins that are known to be integral membrane proteins that recognize environmental signals and initiate signal transduction.

Ligand proteome:

Proteins that bind to receptors and trigger signal transduction.

Localized proteome:

Proteins present in specific subcellular compartments and components.

Transcriptional regulatory proteome:

Proteins that regulate transcription activity of genes via DNA binding.

Posttranslational modified proteome:

Proteins that are known to undergo chemical modifications of their amino acids, such as phosphorylation and ubiquitination.

Interacting proteome:

Proteins known to physically interact with other proteins.

Pathways proteome:

Proteins that are part of known biological pathways and processes.

A more comprehensive approach to identify novel peptides is proteogenomics. Proteogenomics is an area of research at the interface of proteomics and genomics that maps peptides identified in MS/MS experiments to nucleic acid regions that potentially code for proteins. MS/MS-based analysis in proteogenomics contributes to the measurement of peptides arising from expressed proteins, allowing for direct verification of coding regions of a genomic sequence. In addition to verifying protein-coding genes, proteogenomics confirms the expression of alternatively spliced variants in the eukaryotic genome. The broad goal of proteogenomics is to produce a complete quantitative species- and context-specific map of the proteome, including posttranslation modifications (PTM).

Based on the functional annotation of the genes and proteins as well as their gene expression profiles in multiple species, various comparative analyses have been performed: co-expression activity of genes based on orthology [10], differential expression analysis among species [11], and cross-species mapping of posttranslation modifications (PTM) [12] For instance, a recent comparative proteomic analysis suggested that proteins involved in the mammalian reproductive system have rapidly evolved under sexual selection [13]. Such comparative analyses would benefit from improved orthology mapping between pairs of species. Furthermore, improvements in the mapping of MS/MS spectra would also help identify paralog sequences in the reference genome [14] and help identify novel genes [15], both of which are critical to comparative analyses.

While a gene-centric comparative approach helps identify conserved genes and the evolution of individual genes and gene families, as mentioned above, to understand the conservation and diversification of biological processes and pathways, a pathway- or process-centric comparative approach is more suitable. In a pathway-centric comparative proteomic study, one must consider individual protein functions as well as the processes, pathways, and tissues in which each protein is involved, thus defining sets of proteins for such levels of functional abstraction. Given the variable accuracy and completeness across organisms in genome and proteome annotation, one also needs a way to transfer functional annotation from well-annotated to less well-annotated genomes.

This chapter describes a workflow to perform “pathway”--centric proteomic comparisons and applies that workflow to compare the human proteome to the proteomes of 54 other eukaryotes at different levels of functional abstraction. We use these comparisons to rank the nonhuman eukaryotes by their appropriateness as model organism candidates to study a particular functional context and to identify conservation and diversity between human and another 54 eukaryotic species relative to the particular function.

2. Materials

2.1. Proteome Sequences

The complete proteomes of Homo sapiens (20,125 proteins), Pan troglodytes (24,732 proteins), and Macaca mulatta (23,272 proteins) were downloaded from NCBI. The complete proteome of Gorilla gorilla (27,335 proteins) was downloaded from Ensembl (version 66.31). The complete proteomes of 51 eukaryotic organisms with fully sequenced genomes (Table 1) were downloaded from the KEGG database (December 2009), cross-referenced, and complemented with the corresponding proteomes found in the NCBI database.

Table 1.

Human-centric comparison of proteomes for 54 eukaryotes with fully sequenced genomes

OrgName	Phylum	Domain	h-Oc	h-Ac	NHP	nh-Oc	nh-Ac
Homo sapiens (human)	Vertebrates	ANIMAL	20,125	0	–	–	–
Gorilla gorilla	Vertebrates		19,510	615	27,473	26,106	1367
Mus musculus (mouse)	Vertebrates		19,274	851	29,311	24,841	4470
Pan troglodytes (chimpanzee)	Vertebrates		19,270	855	25,175	20,809	4366
Rattus norvegicus (rat)	Vertebrates		19,132	993	26,157	23,597	2560
Macaca mulatta (rhesus monkey)	Vertebrates		19,088	1037	23,998	21,206	2792
Bos taurus (cow)	Vertebrates		18,214	1911	22,368	18,365	4003
Canis familiaris (dog)	Vertebrates		18,205	1920	19,835	18,813	1022
Danio rerio (zebrafish)	Vertebrates		17,946	2179	27,234	25,039	2195
Monodelphis domestica (opossum)	Vertebrates		17,625	2500	19,157	16,943	2214
Xenopus tropicalis (western clawed frog)	Vertebrates		15,778	4347	8741	8623	118
Drosophila melanogaster (fruit fly)	Insects		14,526	5599	13,827	9434	4393
Xenopus laevis (African clawed frog)	Vertebrates		14,473	5652	10,894	10,174	720
Drosophila pseudoobscura pseudoobscura	Insects		14,468	5657	16,071	9819	6252
Aedes aegypti (yellow fever mosquito)	Insects		14,214	5911	15,428	10,769	4659
Strongylocentrotus purpuratus (purple sea urchin)	Echinoderms		13,955	6170	28,881	19,243	9638
Caenorhabditis elegans (Nematode)	Nematodes		13,425	6700	20,185	9443	10,742
Brugia malayi (filaria)	Nematodes		12,928	7197	11,371	6802	4569
Neurospora crassa	Ascomycetes	FUNGI	8911	11,214	9824	4113	5711
Aspergillus nidulans	Ascomycetes		8848	11,277	9541	4691	4850
Laccaria bicolor	Basidiomycetes		8727	11,398	18,215	4774	13,441
Aspergillus oryzae	Ascomycetes		8722	11,403	12,074	5282	6792
Cryptococcus neoformans B-3501A	Basidiomycetes		8653	11,472	6500	3767	2733
Debaryomyces hansenii	Ascomycetes		8164	11,961	6324	3453	2871
Schizosaccharomyces pombe (fission yeast)	Ascomycetes		8153	11,972	5003	3392	1611
Pichia stipitis	Ascomycetes		8078	12,047	5816	3414	2402
Candida glabrata	Ascomycetes		7743	12,382	5191	3117	2074
Fusarium graminearum	Ascomycetes		7015	13,110	11,656	4094	7562
Cryptococcus neoformans JEC21	Basidiomycetes		6607	13,518	6273	3015	3258
Ustilago maydis	Basidiomycetes		6598	13,527	6538	2970	3568
Magnaporthe grisea	Ascomycetes		6235	13,890	14,010	3802	10,208
Kluyveromyces lactis	Ascomycetes		5958	14,167	5335	2553	2782
Saccharomyces cerevisiae (budding yeast)	Ascomycetes		5918	14,207	5881	2829	3052
Candida albicans	Ascomycetes		4043	16,082	14,629	2093	12,536
Encephalitozoon cuniculi	Microsporidians		3215	16,910	1996	879	1117
Arabidopsis thaliana (Thale cress)	Dicotyledons	PLANTS	9561	10,564	27,361	12,364	14,997
Oryza sativa japonica (Japanese rice)	Monocotyledons		7375	12,750	26,937	9184	17,753
Ostreococcus lucimarinus	Green algae		7029	13,096	7603	4010	3593
Cyanidioschyzon merolae	Red algae		7028	13,097	5013	2932	2081
Chlamydomonas reinhardtii	Green algae		6275	13,850	14,416	4480	9936
Monosiga brevicollis	Choanoflagellates	PROTISTS	8829	11,296	9171	4926	4245
Dictyostelium discoideum (cellular slime mold)	Cellular slime molds		7685	12,440	13,437	5067	8370
Trypanosoma cruzi	Euglenozoa		6781	13,344	19,607	7070	12,537
Entamoeba histolytica	Entamoeba		6570	13,555	8162	4094	4068
Trypanosoma brucei	Euglenozoa		6544	13,581	8712	3666	5046
Tetrahymena thermophila	Alveolates		6307	13,818	24,770	6631	18,139
Plasmodium yoelii	Alveolates		6227	13,898	7353	2291	5062
Trichomonas vaginalis	Parabasalids		5058	15,067	59,679	7717	51,962
Leishmania major	Euglenozoa		4683	15,442	8265	3015	5250
Plasmodium falciparum 3D7	Alveolates		4327	15,798	5263	1791	3472
Cryptosporidium parvum	Alveolates		4313	15,812	3805	1615	2190
Cryptosporidium hominis	Alveolates		4075	16,050	3885	1403	2482
Theileria annulata	Alveolates		3879	16,246	3792	1473	2319
Theileria parva	Alveolates		3586	16,539	4061	1453	2608
Giardia lamblia	Diplomonads		2971	17,154	6502	1594	4908

20,125 human proteins were considered. Columns: NHP—total number of proteins in nonhuman proteomes. h-O_c— number of human proteins in O_c against the proteome of the target organism. nh-O_c—number of target proteins in O_c with the human proteome. h-A_c—number of human proteins absent in the proteome of the target organism. nh-A_c— number of target proteins absent in the human proteome

3. Methods

3.1. Homology Analysis

BLAST+ version 2.2.26 was downloaded from NCBI and used locally. All protein sequences were formatted using FormatDB. A pipeline that identifies various levels of homology (orthologs, domain orthologs, homologues) and classifies two proteomes relative to each other based on these relationships was developed and implemented using Python 2.7, Numpy (Numerical Python), Scipy, and Matplotlib.

3.2. Proteome Comparison and Classification

Comparison between the complete proteomes of any two organisms was performed as described previously [1]. In short, the proteomes are first BLASTed against each other. Afterwards, ortholog and homolog pairs are separated based on e-value, percentage identity, and aligned fraction. Figure 1 details the different groups of proteins generated from this analysis. Each protein of the total proteome from a human HP is blasted against the entire proteome of the target organism NHP. The set of ortholog ($O_c$) clusters are defined as all pairs that are matched with $e-\mathit{value}<{10}^{-30}$ and an identity greater than 30% with coverage of alignment of over 80% of the human protein length. Homolog (H_c) clusters are defined as all pairs of proteins that are matched with an identity ≤30% with ${10}^{-30}<e-\mathit{value}<{10}^{-04}$. Absent clusters ($A_c$) are defined as all proteins aligned with $e-\mathit{value}>{10}^{-04}$. Complete protein sets in the $A_c$ cluster further classify unique protein sets, each from a human and a nonhuman proteome, and construct sets as $b-A_c$ (unique in human) and $nb-A_c$ (unique in nonhuman).

Fig. 1.

Integrative workflow of sequence-based orthology analysis and proteomics: The human proteome is BLASTed against the proteome of each organism from Table 1. $O_c$ are clusters of human–nonhuman protein pairs whose alignment has $-\mathit{value}\le {10}^{-30}$, and sequence identity > 30%over at least 80% of the alignment. $H_c$ are clusters of human–nonhuman protein pairs whose alignment has ${10}^{-30}<e-\mathit{value}\le {10}^{-4}$ and sequence identity ≤ 30%. The $h-A_c$ are clusters of proteins unique to the human proteome that do not meet any of the previous quantitative criteria for $O_C$ and $H_C$. The $nh-A_C$ are clusters of nonhuman proteins that do not meet any of the previous quantitative criteria for $O_C$ and $H_C$

Because more than one protein may satisfy criteria under the homolog or ortholog clusters, we select the closest pair of the categories by assigning a score, “F-score,” to that of many-to-many related proteins, quantifying the similarity between a pair of proteins. We select those pairs of proteins that maximize the “F-score” independently within the $O_c$ and $H_c$ clusters. The “F-score” is defined as follows:

$$F-\mathit{score}=(F1+F2)-F3$$ (1)

$$F1=1-((S-I)/S)$$ (2)

In Eq. 2, “S” represents the number of similar residues over the alignment, and “I” represents the number of identical residues. Both these values are outputs of BLAST. F1 is always between 0 and 1.

$$F2=(AL-G1-G2)/PL$$ (3)

In Eq. 3, “G1” is the number of gaps within the aligned region of the query sequence, ‘G2’ is the number of gaps within the aligned region of the target sequence, “AL” represents the length of the alignment, and “PL” is the total length of the query sequence. $F2$ is always between 0 and 1.

$$F3=(G1/L1)+(G2/L2)$$ (4)

In Eq. 4, “L1” is the length of the query sequence, and “L2” is the full length of the target sequence. F3 is always between 0 and 1.

Hence, “F-score” is always between 0 and 2, and the pair of proteins with the highest F-score in each $O_{\mathrm{c}}$ or $H_c$ cluster is considered to be the functional orthologs or homologs, respectively.

We have benchmarked this procedure by BLASTing the human genome against itself and the baker’s yeast genome against itself. In every single case, the highest F-score is that of a protein with itself. Unlike the e-value, the F-score provides a measure that is symmetric between two proteomes. The highest F-score pair for an $O_{\mathrm{c}}$ between two organisms is always the same, whether the target genome is S or T.

Next, we used the $O_c$ and $A_c$ clusters for the analysis of functional evolution.

3.3. Functional ReAnnotation of the Human Proteome

To attribute biological processes, molecular functions, and cellular localization to each cluster, we downloaded the GOSLIM data for human and baker’s yeast from the GO database [16, 17]. We also downloaded data from KEGG that associates genes with KEGG metabolic circuits in fully sequenced genomes [18] and attributed pathway terms to each of the clusters.

3.4. Calculation of the Hamming Distance

The Hamming Distance (HD) between the vector $\overrightarrow{V1}$ of protein function associated with a specific process or pathway in humans and the vector $\overrightarrow{V2}$ of corresponding protein functions in another organism gives a measure of how different the two vectors are. It is calculated using the formula:

$$HD=\sum _{i=1}^n\left(1-{\delta }_i\right)$$

where ${\delta }_i$ is the Kronecker delta. ${\delta }_i$ is 1 if the elements in position i of both vectors are orthologs and 0 otherwise. The smaller the distance, the more similar the two vectors are, and the more similar the set of genes executing a specific process in both organisms. HD can be normalized (NHD) by dividing it by the maximum HD between the corresponding vectors of all organisms. Consequently, the smaller the NHD is, the more likely the other nonhuman eukaryote is a good model to study the relevant process or pathways and generalize the results for the other organism. The vectors we define for each pathway include all proteins that could participate in that pathway in all organisms in the KEGG database. This ensures that the comparison we are making accounts for differences between the pathway in humans and that in other organisms and vice-versa. All calculations were performed using Mathematica [19].

3.5. Proteomics Analysis

The species-specific experimental mass-spec data can be downloaded from different public repositories and transformed to mzML format using the MSConvertGUI tool integrated in ProteoWizard 3.0 [20]. Next, one can search spectra from each of the samples against the corresponding species-specific library of pre-identified spectra to which peptide sequences have been assigned through the SpectraST search engine incorporated in Trans Proteomic Pipeline (TPP) software. For this, one can download human and other eukaryotes-specific libraries from the PeptideAtlas website (http://www.peptideatlas.org/speclib/) in split format. Each species-specific sample in mzML can be input into the TPP-incorporated SpectraST tool for spectra searching and for downstream analysis of the matched spectra with protein sequences. For mapping identified peptides on protein sequences, one can use input FASTA files made of protein sequences, each from O_c clusters identified during comparison between human and nonhuman eukaryotes proteomes.

Further, one can validate the identified matched peptides using PeptideProphet and refine peptide-level analysis using iProphet search engines. For the selection of true-positive spectra, one can choose various statistical models available in the TPP SpectraST tool, such as the NRS model, number of sibling ions (NSI) model, number of sibling modifications (NSM) model, number of sibling searches (NSS) model, number of sibling experiments (NSE) model, and number of sibling peptides (NSP) model. In addition, one can normalize the sample peptides against the pI and hydrophobicity information about library peptides. Ultimately, the SpectraST-based search produces output in pepXML and protXML formats.

For the selection of significant signatures of peptides, one can input the pepXML into SkyLine tools [21] and compare those against sequences from randomly selected proteins in the same FASTA files that one had used during the previous peptide searching step. Ultimately, one can select the proteins whose peptides are mapped at FDR < 0.1 and consider these proteins as expressed candidates in the corresponding species-specific MS/MS data.

3.6. Results

3.6.1. Analysis Overview

We systematically compared the complete human proteome to that of 54 other eukaryotes with fully sequenced genomes using BLAST and separated the results into four main groups (Fig. 1):

1. Cluster of orthologs ($O_c$), which includes proteins that have strong similarity to a specific human protein over the entire length of both proteins. Protein pairs in these clusters are deemed the most likely functional orthologs between the two proteomes.

2. Cluster of homologs ($H_c$), which includes proteins that only have low sequence similarity to a specific human protein. Protein pairs in these clusters are deemed less functionally similar between the two proteomes.

3. Clusters of human-only proteins ($h-A_c$), which include proteins in the human proteome that have no match to any proteins in the proteome of a compared eukaryote.

4. Clusters of nonhuman proteins ($nh-A_c$), which include proteins in the proteome of a eukaryote that have no match to any proteins in the compared human proteome.

3.6.2. Large Scale Proteome Comparisons

Table 1 summarizes the $O_c,H_c,h-A_c$, and $nh-A_c$ results of the comparison between the proteome of humans and that of each of the other 54 eukaryotic species. As one would expect, the proteome of humans is most similar to that of other vertebrates. The vertebrate with a proteome having the least similarity to that of humans is the African clawed frog. Among all human proteins, 28% have no strong sequence match in the frog proteome, whereas only 7% of the frog proteins are absent in humans. On the opposite end of the scale, and to our surprise, gorillas have a proteome that is most similar to that of humans. Among all human proteins, 3% have no sequence match to the proteome of gorillas, while 5% of the gorilla proteins are absent in the human proteome. The corresponding numbers in Chimpanzees, which are phylogenetically closer to humans than Gorillas, are 4% and 17%, respectively (see Table 1).

The non-vertebrate species with a proteome that is closest to that of humans is Drosophila melanogaster, while the least similar proteome is that of Brugia malayi (filaria). Only 64% of the human proteins have at least one homolog in B. malayi, while 40% of this protein are absent in humans.

Our results broadly follow the known phylogeny: in the decreasing order of proteome similarity with humans, the eukaryotic clades are: Animals, Fungi, Plants, and Protista. Giardia lamblia is an eukaryotic organism with a proteome that is the most dissimilar to that of humans, with only 15% of its proteins being present in humans. Overall, only 1250 human proteins are consistently conserved throughout the 54 nonhuman eukaryotes. In addition, 180 proteins are unique to humans and absent in all other eukaryotes.

3.6.3. Similarities and Differences in Process-Associated Proteomes Between Other Eukaryotes and Humans

$O_c$ and, to a least extent, $h-A_c$ clusters identify proteins with similar functions. By focusing on these clusters and mapping their proteins to functional gene groups, one can compare functional similarity and divergence between humans and other eukaryotes. Proteins in the $O_c$ cluster are more likely to share the same function in different organisms. Thus, these clusters permit studying functional conservation between organisms with higher confidence level, identifying the best alternative organism(s) in which to study a given process that cannot be studied for various reasons in humans.

Figure 2 summarizes how the protein complement associated with broad functional categories is conserved between humans and other eukaryotes. Overall, the conservation of the complement of proteins associated with each category follows a phylogenetic pattern. The functional category with the highest degree of conservation in its protein complement is catalytic proteins, followed by the substrate proteins functional category. Conservation of protein functions in this set is very high among animals and decreases between animals and more distant phyla, suggesting that the metabolism of animals is, in general, similar. In fact, the fraction of human proteins from the catalytic protein set that is absent from all other eukaryotes is minimal (<15% of the total human catalytic proteins).

Fig. 2.

Summary for comparison of human-specific broad functional categories associated with conserved and unique proteins respectively identified in ${O_c}^{\mathrm{*}}(\mathbf{a})$ and ${h-A_c}^{\mathrm{*}}(\mathbf{b})$ clusters: Each row summarizes the results for an organism. Each column summarizes the results for a broad functional category of the proteome. The greener the square, the more similar in the case of A (or more dissimilar in case of B), the protein set associated with the functional category of the column in the organism of the row is to the correspondent human protein set. The redder the square, the less similar in case of A (or more similar in case of B), the protein set associated with the functional category of the column in the organism of the row is to the correspondent human protein set. *(a): Cluster of orthologs ($O_c$), which includes proteins that have strong similarity to a specific human protein over the entire length of both proteins. Protein pairs in these clusters are deemed the most likely the closest functional orthologs between the two proteomes. *(b): Clusters of human-only proteins ($h-A_c$), which include proteins in the human proteome that have no match to any proteins in the proteome of a compared eukaryote

In contrast, the functional categories receptors, immunologic proteins, and ligands have the lowest degree of protein conservation between humans and other eukaryotes. This argues for the species specificity of these classes of proteins and their potential role in phenotypic diversity [22, 23].

3.6.4. Tissue- and Organ-Specific Proteome Analysis

Here we compare the fraction of the proteome that is specifically expressed in various human tissues/organs or that is secreted with the proteome of other eukaryotes. Figure 3 summarizes the results.

Fig. 3.

Summary of protein conservation for the human proteome in specific tissues: Each column summarizes the results for an organism. Each row summarizes the results for a broad functional category. The greener the square, the more similar the protein set associated with the functional category of the row in the organism of the column is to the correspondent human protein set. The redder the square, the less similar the protein set associated with the functional category of the row in the organism of the column is to the correspondent human protein set

The tissue-specific proteomes with the highest degree of conservation in eukaryotes are those associated specifically with lacrimal glands (452 proteins) and tears (459 proteins). The proteins from these sets have functional orthologs in all animals and some lower eukaryotes. Interestingly, functional orthologs for the human proteins CACNA1D, KCNN3, and PRR4 are found in all animals except for the African clawed frog. The first three proteins regulate calcium and potassium channels and their activities, while the last one appears to play a role in stabilizing the tear film, protecting the ocular surface of the eyes in humans [24], whereas in the African clawed frog, these proteins are functional in left–right asymmetry during eye development [25].

The set of proteins associated with most types of reproductive tissues in humans is highly conserved in mammals. For example, PAEP (progestagen-associated endometrial protein, Entrez gene id 5047) is conserved in all mammals. This glycoprotein contributes to making the uterine environment suitable for reproduction and is used for predicting pregnancy following an in vitro fertilization (IVF) cycle [26]. In contrast, protein sets from the urethra, human scalp, and seminal vesicle are only highly conserved in primates.

Our analysis also highlights that olfactory tract-specific proteins have a degree of conservation in a large fraction of eukaryotic organisms. Such conservation is a further indication that the mechanisms for nutrient detection and environmental recognition evolved from an original rudimentary set of proteins, as suggested by [27].

There are some interesting differences between primates with respect to the conservation of some individual tissue-specific proteins. For example, GHRL (ghrelin/obestatin prepropeptide, gene-ID: 51738) is annotated as being specifically expressed in the vena cava. This protein is absent in chimps and is present in other primates. It regulates growth hormone release and is involved in inhibiting thirst and anxiety [28]. It is also a good marker for studying type-2 diabetes and obesity [29], ischemic stroke [30], cardiovascular functions [31], and Rett syndrome [28]. Its absence in chimps suggests that other primates would be better models for studying aspects of these diseases, specifically those involving GHRL.

4. Discussion

Here, we sketch a systematic approach to compare the proteome of an organism to that of others with respect to specific processes, pathways, and tissues. We focus on the human proteome as the reference. This allows us to rank various organisms in order of similarity to humans for each of the categories in the analysis, creating an approach and resource for human-centric comparative eukaryotic proteomics.

Previous work has used a similar approach as ours to identify “phenologs,” i.e., equivalence between a human disease and a phenotype in five other organisms, based on protein similarity between the two sets—disease-associated proteins in humans and phenotype-associated proteins in the other species [32, 33]. Our work has a different goal and a substantially greater scope with 54 eukaryotes and protein sets corresponding to biological processes, pathways, protein families, tissue specificity, etc.

One limitation of our study is that it is contingent on the quality and completeness of protein annotation in different species, which is variable across species. However, to minimize this concern, we carefully combined the annotations from NCBI and the Sanger center. Another data-related limitation is the functional annotations in various genomes. However, this is somewhat mitigated in our human-centric study as we refer only to human functional annotations, which are continuously improving.

Another limitation is the fact that we used extreme sequence similarity as a proxy for functional orthology and a lack of sequence similarity as a proxy for the absence of the protein in one of the species. Although this is not always true, using sequence similarity to infer functional orthology was shown to be more accurate than using structure and other protein features. Although large sets of experimentally characterized functional orthologs are not available to benchmark any of these methods, our specific pipeline, although based on arbitrary significance thresholds, should be considered stringent. Moreover, our pipeline is flexible and can be customized for specific needs.

The results above rely only on the protein annotation of genomes and the functional annotation of the proteins, and as such lack information about physiological contexts. Here we discuss how one can corroborate and refine such comparisons using experimental MS/MS data to assess the expression of proteins in specific physiological contexts. One can qualitatively and quantitatively identify the expression of the ortholog proteins independently from species-specific MS/MS data. One can identify the human and other eukaryotic proteins in each $O_c$ cluster and map species-specific MS/MS spectra on top of the sequence of the ortholog proteins (see method section). One can also map the species-specific MS/MS spectra on top of the protein sequences in the $h-A_c$ and $nh-A_c$ clusters, respectively, identified as unique proteins from humans and eukaryotes, in order to identify novel proteins in a species that might not be identified by the conventional method of mapping the species-specific MS/MS data directly to that species proteome.

5. Conclusion

In summary, we have reported the first comprehensive survey of gene set-centric conservation of the proteome between humans and 54 eukaryotes that have complete genome sequences and protein annotations. Our results should serve as an important resource for investigating the evolution of specific processes and for prioritizing model organisms to investigate specific aspects of human biology, including diseases.