Human functional genetic studies are biased against the medically most relevant primate-specific genes
© Hao et al; licensee BioMed Central Ltd. 2010
Received: 2 July 2010
Accepted: 20 October 2010
Published: 20 October 2010
Many functional, structural and evolutionary features of human genes have been observed to correlate with expression breadth and/or gene age. Here, we systematically explore these correlations.
Gene age and expression breadth are strongly correlated, but contribute independently to the variation of functional, structural and evolutionary features, even when we take account of variation in mRNA expression level. Human genes without orthologs in distant species ('young' genes) tend to be tissue-specific in their expression. As computational inference of gene function often relies on the existence of homologs in other species, and experimental characterization is facilitated by broad and high expression, young, tissue-specific human genes are often the least characterized. At the same time, young genes are most likely to be medically relevant.
Our results indicate that functional characterization of human genes is biased against young, tissue-specific genes that are mostly medically relevant. The biases should not be taken lightly because they may pose serious obstacles to our understanding of the molecular basis of human diseases. Future studies should thus be designed to specifically explore the properties of primate-specific genes.
Proteins and their encoding genes can be characterized by functional attributes, such as which pathways they act in or what molecular functions they have; structural attributes, such as lengths of their coding regions or UTRs and GC contents; and evolutionary attributes, such as substitution rates between species and estimates of selection pressures. With the increasing availability of functional genomic data and systems biology tools, correlations between some of these attributes have been observed. The two factors with the strongest associations with other data types in humans are expression breadth (the number of tissues one protein is expressed in) and phyletic age (defined by the evolutionarily most distant species where homologs can be found) [1, 2]. For example, recent studies have shown that expression breadth correlated with promoter architecture, evolutionary rates (Ka and Ks) and gene length [2, 3], while human proteins of different phyletic age are enriched in distinct functional categories . Interestingly, many properties that correlate with expression breadth also correlate with phyletic age and vice versa; indeed, some studies have reported correlations between phyletic age and expression breadth [2, 4].
In evolutionary history, proteins involved in basic biological processes such as transcription and translation machineries, metabolism and cell cycle control were probably invented first. Accordingly, the corresponding genes are old, and conserved in multiple organisms, and are expressed in multiple tissues. Conversely, younger genes perform more specialized functions, and are thus typically used and expressed in specific tissues and/or environments. This line of reasoning predicts a significant correlation between phyletic age and expression breadth. Consequently, variables correlate with one of the two factors should also correlate with the other. For example, many old genes are expected to be broadly useful (and hence broadly expressed), to come from basic functional groups (such as transcription/translation or metabolism), and to require distinct promoter architectures and optimized gene structures; they may also evolve mostly under strong purifying selection.
This picture of increasing specialization with decreasing gene age is an oversimplification. For example, gene duplication may often be followed by subfunctionalization , where each of the two gene copies takes over the function of the ancestral gene in part of the tissues of the ancestral gene. A recent study suggested that subfunctionalization was indeed partially underlying the correlation between the rate of increase in gene tissue specificity and the rate of increase in the maximum number of cell types . The duplication thus creates two gene copies with the same phyletic age, but very different expression profiles [7, 8]; other changes of gene attributes (e.g., sequence changes to optimize tissue-specific function) might follow. Furthermore, genes expressed in different tissues may be under different constraints, resulting in variable evolutionary rates across proteins with similar expression breadth . Another important factor known to influence gene structure and evolution is expression abundance [10–12]. Thus, while we expect a strong overlap between the influence of phyletic age and expression breadth on human protein features, we expect age and breadth to differ in their influence on individual proteins; and we expect that other factors, such as expression abundance, also contribute to systematic variation in protein attributes.
Functional studies of proteins and their encoding genes are biased by gene properties. For example, highly and broadly expressed gene products are easy to detect and are thus studied preferentially. Older genes, which are most likely conserved across multiple species (especially model organisms used for disease models), are studied more intensively due to the availability of animal (or even yeast) models. Proteins with basic molecular functions were studied preferentially because of their importance in biology. Due to these biases, we expect that databases like Gene Ontology  and KEGG pathways [14, 15] contain more reliable functional annotations for old and broadly expressed genes. Conversely, younger genes, which may be important, e.g., in human-specific gene regulation, are not studied as much.
In this study, we focus on human protein-coding genes to explore systematic correlations of phyletic age and expression profiles with some intrinsic gene features, as well as the interdependence of age and expression breadth. Based on our observations, we discuss systematic biases in functional studies that affect the process of knowledge acquisition for human disease-related genes.
If it increases with expression breadth, it probably increases with age
Correlation of human protein-coding gene properties with phyletic age and expression breadth
Protein length (log)
Length 1st intron (log)
Length of 5' UTR (log)
GC content of CDS
Expression level (log)
1.3 × 10-11***
8.9 × 10-09***
Overall, we examined five structural gene properties: length of proteins and 5'-UTRs, number of exons, length of the first exon, and GC content of gene's coding regions. Each structural property correlates significantly with both phyletic age and expression breadth (P ≤ 0.0005 in each case, see Table 1, Additional file 1, Figure S1b-g and Figure S2a-f for details). The strongest correlations are those with protein length, gene length and exon numbers, while 5'-UTR length and GC content show weaker correlations. Of particular interest are the weak but statistically highly significant correlations of the length of first introns with phyletic age and expression breadth (R = 0.10 and 0.06, respectively). First introns are known to often harbor regulatory elements . Thus, the results in Table 1 suggest that tissue-specific gene expression is at least in part achieved through additional regulation in 5'-UTRs and first introns. This notion is consistent with the correlation between 5'-UTR length and first intron length (R = 0.095, P < 10-15).
We also examined gene-specific evolutionary rates, estimated from the fraction of synonymous (Ks) and non-synonymous (Ka) sites with nucleotide substitutions between human and mouse orthologs. Ka, Ks, and their ratio Ka/Ks all correlated negatively with both phyletic age and expression breadth, which is consistent with previous findings [1, 16] (see Table 1 and Additional file 1, Figure S1h and S2g for details).
Are more ancient human genes really involved in more basal cellular functions? While the relationship between functional categories and phyletic age has been investigated previously [1, 4], the trend among human proteins was relatively weak . Here, we used annotation data from two sources, GeneOntology (GO)  and pathway annotations from KEGG [14, 15]. Of the 22,165 human non-redundant protein-coding genes in this study, 7,452 had GO annotations supported by at least one of the experimental evidence codes (IDA, IPI, IMP, IGI, and IEP). Using a modified method from Freilich et al. , we grouped genes into four main categories according to their GO annotation: enzymatic activity, transporter, regulation (including transcription regulator activity, translation regulator activity, and enzyme regulator activity), and signal transduction. These functional annotations indeed correlated with expression breadth (Figure 1c): the fraction of enzymes increases with increasing expression breadth, while the fraction of genes involved in signal transduction decreases. In contrast, the fractions of transporters and genes involved in regulation remained roughly constant across different expression breadths.
While we also found global patterns in the distribution of the functional groups across different age groups, these were not a simple mirror image of the results for expression breadth. Similar to the trend with increasing expression breadth, we found that the combined fraction of enzymes and transporters decreases with phyletic age, while the combined relative number of genes involved in regulation and signal transduction increases (Figure 1d). For expression breadth these trends are almost entirely due to variation in enzyme and regulator fractions. For phyletic age, there is also a decrease in the fraction of transporters in primate-specific genes. Even more strikingly, regulatory genes show a U-shaped distribution, compensated by massively increased fractions of signaling genes in metazoa-, chordata-, and mammalian-specific genes (Figure 1d).
We found a comparable number of our genes (8,569) to be annotated in at least one of four pathway categories in KEGG: metabolism, genetic information processing, environmental information processing, and cellular processes. As shown in Figure 1e and 1f, global trends are very similar to the GO results. However, metabolism appears almost absent among mammalian-specific genes, while environmental information processing is the dominant category also in primate-specific genes. After examining a wide range of structural, evolutionary, and functional properties of human genes, we thus conclude that any trends with increasing expression breadth seem to be always mirrored by according trends with increasing phyletic age, and vice versa.
Age and expression breadth affect gene properties independently
Influence of expression breadth, phyletic age and expression abundance on protein properties using generalized linear modela
Protein length (log)
4.7 × 10-13***
5.50 × 10-13***
2.30 × 10-14***
1st intron length (log)
4.43 × 10-05***
Length 5' UTR (log)
4.31 × 10-05***
GC of CDS
8.34 × 10-11***
Molecular function (GO)
Pathway class (KEGG)
7.17 × 10-6***
7.84 × 10-5***
2.41 × 10-7***
Human functional studies are biased against disease-related genes
These biases are not surprising: regularly, the first step in functional annotation is comparison to characterized genes with similar amino acid sequence. The probability of a good match obviously increases with the phylogenetic distribution of homologs; in particular, this approach will seldom be successful for genes that are restricted to primates. Experimental characterization is usually done in model organisms, with a substantial part of our knowledge derived from species as distantly related to humans as insects, nematode worms, or even yeast. Finally, tissue-specific (and often lowly expressed) genes are harder to observe experimentally, adding a bias towards broadly expressed proteins to the phylogenetic bias.
Should we care about this bias? An analysis of disease-related genes in the Online Mendelian Inheritance in Man database (OMIM)  shows that such medically relevant genes are strongly enriched among primate-specific genes: about 20% of all genes restricted to humans and other primates are currently associated with diseases (Figure 2c), more than in any other age class; this percentage is very likely an underestimate. That this medically most relevant group of genes is the one least characterized appears problematic.
Grouping human protein-coding genes by both phyletic age and by expression profiles, we observed significant correlations of these two factors with a number of gene properties, including gene structure, GC content, promoter architecture, evolutionary rate, and functional classification. While several of these correlations had been observed previously [1, 2, 4], we show that in each case both factors, age and expression pattern, independently contribute to the observed variation. The trends are gradual, with substantial overlap between neighboring groups.
Another important gene property correlated with phyletic age as well as expression breadth is the probability of being annotated in public databases (Figure 2a, b and Additional file 1, Figure S1i): the youngest, primate-specific genes are the least characterized functionally, partly because of a lack of orthologs in model species, partly because of the typical gene properties analyzed above. Thus, there is a substantial bias against the functional characterization of certain genes: functional studies preferentially target genes that are conserved in other species, with broad expression breadth and basic functions, while younger genes with temporally and spatially restricted expression are less understood. These biases are not unexpected, as lowly expressed genes are difficult to capture even using high-throughput technologies, and as functional studies are facilitated by the availability of orthologs in model species.
Unfortunately, it is the primate-specific genes that contain the highest fraction of disease-related genes (19.4%, Figure 2c). The systematic biases against their study should thus not be taken lightly: the incomplete functional characterization of the youngest genes may pose serious obstacles to our understanding of the molecular basis of human diseases. Cell line culturing, tissue engineering and the emerging induced pluripotent stem cells technique  may help to provide solutions to this bias.
Our results differ markedly from those of Domazet-Loso and Tautz [21, 22], which found that the percentage of human disease-related sequences remained approximately constant up to the divergence of the mammalian lineages, and then decreased steeply with decreasing age. One major methodological difference between our study and [21, 22] is that these authors identified phylogenetic sequence age based on a very lenient Blast E-value cutoff (0.001). This approach places all gene families that share a particular protein domain into the age class where this domain emerged first, even though a particular gene may have evolved later, e.g., via gene duplication [21, 22]. The youngest sequences in [21, 22] are thus genes without any recognizable homology to genes in other phyla; it is conceivable that some of these may in fact be mis-annotations. In contrast, the method employed here  is designed to confidently assign phylogenetic ages to genes rather than domains. The youngest genes in our study often share protein domains with older sequences, but arose recently by gene duplication or exon shuffling. Thus, while the results of [21, 22] apply to the age distribution of disease-associated protein domains, our results give the age distribution of complete disease-associated genes.
Human protein coding genes and gene properties
22,165 non-redundant human protein-coding genes were downloaded from the HomoloGene database release 64 http://www.ncbi.nlm.nih.gov/homologene. Properties of the human genes, including number of exons, intron length, and 3' and 5' UTR length, were calculated based on BLAT mapping results of nucleotide gene sequences to the human genome available at the UCSC Genome Browser . Evolutionary rate of human and mouse orthologs, promoter type, as well as expression breadth of human genes were obtained from [2, 3]. Four types of promoter architectures were identified according to the presence and absence of CpG-island and TATA-box in the core promoter . mRNA expression levels across 12 diverse human tissues based on deep sequencing of cDNA fragments were downloaded from the NCBI GEO database . Gene expression (mRNA abundance) was pre-calculated as reads per kilobase of exon model per million uniquely mapped reads (RPKM), and deposited under GEO accession GSE12946 .
Phyletic ages of all non-redundant protein-coding genes were obtained from .
GO and KEGG pathway annotations and functional classification
Human GO annotations were obtained from the Gene Ontology  website http://www.geneontology.org; Only GO annotations with any of the experimental evidence codes IDA, IPI, IMP, IGI or IEP (GO-EXP) were used in this study. Based on GO, human protein-coding genes were divided into four groups: enzyme, transporter, regulation and signal transduction, using a method similar to .
Pathway annotations of human genes were obtained from KEGG [14, 15]. According to these annotations, human genes were classified into one of five categories: metabolism, genetic information processing, environmental information processing, cellular processes, and human disease. We did not consider the category "human disease", as only a few genes were classified into this category by KEGG.
Diseases related genes
The 'genemap' file containing a list of the best-curated disease genes was downloaded from the Online Mendelian Inheritance in Man (OMIM) database  on April 27, 2010. Out of 12,489 entries, we selected 2,428 entries with the "(3)" tag, for which there is strong evidence that at least one mutation in the particular gene is causative for the disease. In total, 2,349 genes in the gene set analysed in this study were marked as disease-related genes.
We are grateful to Fuhong He and Jiang Zhu for providing the expression breadth data, and to Yuri I. Wolf for providing the phyletic age data. The study is supported by grants from the National Basic Research Program (2006CB910401, 2006CB910403 and 2006CB910404), the Special Foundation Work Program (2009FY120100) and the National Key Technology R&D Program (2008BA164B02) from the Ministry of Science and Technology of the People's Republic of China.
- Wolf YI, Novichkov PS, Karev GP, Koonin EV, Lipman DJ: Inaugural Article: The universal distribution of evolutionary rates of genes and distinct characteristics of eukaryotic genes of different apparent ages. Proc Natl Acad Sci USA. 2009, 106 (18): 7273-7280. 10.1073/pnas.0901808106.PubMed CentralView ArticlePubMedGoogle Scholar
- Zhu J, He F, Hu S, Yu J: On the nature of human housekeeping genes. Trends Genet. 2008, 24 (10): 481-484. 10.1016/j.tig.2008.08.004.View ArticlePubMedGoogle Scholar
- Zhu J, He F, Song S, Wang J, Yu J: How many human genes can be defined as housekeeping with current expression data?. BMC Genomics. 2008, 9: 172-10.1186/1471-2164-9-172.PubMed CentralView ArticlePubMedGoogle Scholar
- Freilich S, Massingham T, Bhattacharyya S, Ponsting H, Lyons PA, Freeman TC, Thornton JM: Relationship between the tissue-specificity of mouse gene expression and the evolutionary origin and function of the proteins. Genome biology. 2005, 6 (7): R56-10.1186/gb-2005-6-7-r56.PubMed CentralView ArticlePubMedGoogle Scholar
- Lynch M, Force A: The Probability of Duplicate Gene Preservation by Subfunctionalization. Genetics. 2000, 154 (1): 459-473.PubMed CentralPubMedGoogle Scholar
- Milinkovitch MC, Helaers R, Tzika AC: Historical constraints on vertebrate genome evolution. Genome Biol Evol. 2010, 2: 13-18. 10.1093/gbe/evp052.PubMed CentralView ArticleGoogle Scholar
- Gu Z, Nicolae D, Lu HH, Li WH: Rapid divergence in expression between duplicate genes inferred from microarray data. Trends Genet. 2002, 18 (12): 609-613. 10.1016/S0168-9525(02)02837-8.View ArticlePubMedGoogle Scholar
- Adams KL, Cronn R, Percifield R, Wendel JF: Genes duplicated by polyploidy show unequal contributions to the transcriptome and organ-specific reciprocal silencing. Proceedings of the National Academy of Sciences of the United States of America. 2003, 100 (8): 4649-4654. 10.1073/pnas.0630618100.PubMed CentralView ArticlePubMedGoogle Scholar
- Su Z, Huang Y, Gu X: Tissue-driven hypothesis with Gene Ontology (GO) analysis. Annals of biomedical engineering. 2007, 35 (6): 1088-1094. 10.1007/s10439-007-9269-y.View ArticlePubMedGoogle Scholar
- Carmel L, Koonin EV: A Universal Nonmonotonic Relationship between Gene Compactness and Expression Levels in Multicellular Eukaryotes. Genome Biol Evol. 2009, 2009 (0): 382-390.Google Scholar
- Pal C, Papp B, Hurst LD: Highly expressed genes in yeast evolve slowly. Genetics. 2001, 158 (2): 927-931.PubMed CentralPubMedGoogle Scholar
- Pal C, Papp B, Lercher MJ: An integrated view of protein evolution. Nat Rev Genet. 2006, 7 (5): 337-348. 10.1038/nrg1838.View ArticlePubMedGoogle Scholar
- Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, Davis AP, Dolinski K, Dwight SS, Eppig JT, et al: Gene ontology: tool for the unification of biology. The Gene Ontology Consortium. Nat Genet. 2000, 25 (1): 25-29. 10.1038/75556.PubMed CentralView ArticlePubMedGoogle Scholar
- Kanehisa M, Goto S, Furumichi M, Tanabe M, Hirakawa M: KEGG for representation and analysis of molecular networks involving diseases and drugs. Nucleic Acids Res. 2009, D355-360. 38 Database
- Kanehisa M, Goto S: KEGG: kyoto encyclopedia of genes and genomes. Nucleic Acids Res. 2000, 28 (1): 27-30. 10.1093/nar/28.1.27.PubMed CentralView ArticlePubMedGoogle Scholar
- Schug J, Schuller WP, Kappen C, Salbaum JM, Bucan M, Stoeckert CJ: Promoter features related to tissue specificity as measured by Shannon entropy. Genome biology. 2005, 6 (4): R33-10.1186/gb-2005-6-4-r33.PubMed CentralView ArticlePubMedGoogle Scholar
- Yoshiro N, Tsukasa N, Satoshi S, Takeshi M, Shinji K, Takeo K: Transcriptional regulatory elements in the 5' upstream and first intron regions of the human smooth muscle (aortic type) [alpha]-actin-encoding gene. Gene. 1991, 99 (2): 285-289. 10.1016/0378-1119(91)90140-7.View ArticleGoogle Scholar
- Pal C, Papp B, Lercher MJ: Adaptive evolution of bacterial metabolic networks by horizontal gene transfer. Nature genetics. 2005, 37 (12): 1372-1375. 10.1038/ng1686.View ArticlePubMedGoogle Scholar
- McKusick VA: Mendelian Inheritance in Man and its online version, OMIM. Am J Hum Genet. 2007, 80 (4): 588-604. 10.1086/514346.PubMed CentralView ArticlePubMedGoogle Scholar
- Takahashi K, Tanabe K, Ohnuki M, Narita M, Ichisaka T, Tomoda K, Yamanaka S: Induction of pluripotent stem cells from adult human fibroblasts by defined factors. Cell. 2007, 131 (5): 861-872. 10.1016/j.cell.2007.11.019.View ArticlePubMedGoogle Scholar
- Domazet-Loso T, Tautz D: An ancient evolutionary origin of genes associated with human genetic diseases. Molecular biology and evolution. 2008, 25 (12): 2699-2707. 10.1093/molbev/msn214.PubMed CentralView ArticlePubMedGoogle Scholar
- Domazet-Loso T, Tautz D: Phylostratigraphic tracking of cancer genes suggests a link to the emergence of multicellularity in metazoa. BMC Biol. 2010, 8: 66-10.1186/1741-7007-8-66.PubMed CentralView ArticlePubMedGoogle Scholar
- Rhead B, Karolchik D, Kuhn RM, Hinrichs AS, Zweig AS, Fujita PA, Diekhans M, Smith KE, Rosenbloom KR, Raney BJ, et al: The UCSC Genome Browser database: update 2010. Nucleic Acids Res. 2010, D613-619. 10.1093/nar/gkp939. 38 Database
- Barrett T, Troup DB, Wilhite SE, Ledoux P, Rudnev D, Evangelista C, Kim IF, Soboleva A, Tomashevsky M, Marshall KA, et al: NCBI GEO: archive for high-throughput functional genomic data. Nucleic Acids Res. 2009, D885-890. 10.1093/nar/gkn764. 37 Database
- Wang ET, Sandberg R, Luo S, Khrebtukova I, Zhang L, Mayr C, Kingsmore SF, Schroth GP, Burge CB: Alternative isoform regulation in human tissue transcriptomes. Nature. 2008, 456 (7221): 470-476. 10.1038/nature07509.PubMed CentralView ArticlePubMedGoogle Scholar
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.