Abstract: We examined the synonymous vs. nonsynonymous substitution rate ratios (Ka/Ks, aka evolutionary rate) between human and chimpanzee for 166 successful drug target genes and compared them with a larger (10,298) set of genes representative of average human-chimpanzee evolutionary rates. We found that evolutionary rates differ significantly between successfully marketed drug targets and the broader set of genes (p<0.005 by ANOVA). Evolutionary rates were lower for successfully marketed drug targets versus non-target genes (0.311 versus 0.497). This rate discrepancy demonstrates that more conserved genes, even within protein families such as GPCRs (successful target GPCRs 0.391 versus non-target GPCRs 0.855) and protein kinases (0.131 versus 0.337), are better targets for traditional small molecule drug development than less strongly constrained genes. Evolutionary rate, therefore, is a factor that could be taken into account when selecting candidate target genes for drug discovery, in addition to the biochemical properties of the proteins these genes encode. We suggest therefore, that links be established between identified disease-causal or -associated genes and genes that are suitable targets for traditional small molecule pharmaceutical development.
INTRODUCTION
Currently, the pre-clinical phase of drug development-including target identification and validation, assay development, compound screening, transition from hit to lead, optimization of leads and finally, the development of those leads-takes an average of 4-6 years and costs, capitalized, over $30M per annum (Di Masi et al., 2003; Adams and Brantner, 2010). The identification of valid gene targets for therapeutic intervention early in this cycle, therefore, has long-term importance both to the public health and the economic strength of the pharmaceutical industry. With the rise of target-based drug discovery and in silico methods of screening, the process of target selection has become more tractable; for example, Bakheet and Doig (Bakheet and Doig, 2009). Identified sets of desirable properties including high hydrophobicity, long length and presence of SignalP motif and Zhu (Zhu et al., 2009) took into consideration sequence, structure, physicochemical characteristics and systems profiles. These studies are excellent demonstration of incorporation of protein biochemical property into drug discovery. The explosion of small molecule sources (Ramasamy et al., 2011; El-Said and Al-Barak, 2011; Kayode and Kayode, 2011; Meena et al., 2010; Gantait et al., 2010; Arkin and Wells, 2004; Dancik et al., 2010) and delivery systems (Akiyama et al., 2000; Harisa et al., 2011; Eleazu et al., 2011; Balamuralidhara et al., 2011) means that target selection will be highly important for most screening methodologies.
While the concept of examining the evolutionary rates of candidate genes is not new (Ma and Wang, 2009; Durand et al., 2008; Bradford et al., 2006), to date there has been no consensus on what the best features to examine are from a pharmaceutical perspective (Zhu et al., 2009). Is success more likely to come from targeting drugs at a gene that is evolving more rapidly, or evolving more conservatively? Alternatively, is there no correlation at all? Previous studies have found that disease-causal genes are likely to be more rapidly evolving (Longman-Jacobsen et al., 2003; Dawkins et al., 1999; Fay and Wittkopp, 2008; Ma and Wang, 2009; Durand et al., 2008; Blekhman et al., 2008; Holbrook and Sanseau, 2007) than the general rate of primate gene evolution. Taken with previous genomic studies (Winter et al., 2004; Bradford et al., 2006; Smith and Eyre-Walker, 2003; Vamathevan et al., 2008), our current study strongly suggests that disease genes and successful drug target genes are often different, incongruent sets. In this article, we also extend our analysis to include successful drug targets.
We examined the properties of drug target genes that have yielded successful drugs on the market. We asked if evolutionary rate is a viable trait to examine when attempting to narrow the analysis of drug targets from a broad list of potential drug target candidates.
MATERIALS AND METHODS
Orthologous gene pair acquisition and analysis: Data was acquired, in part, from the SPEED database (Vallender et al., 2006). When data from SPEED was not available or alignments were redone, the following procedure was followed. Datasets used were obtained from Ensembl. For purposes of this paper, we focused on human and chimp sequences (Fernandez-Banet et al., 2007).
When not already available, human-mouse orthologs were gathered from synteny maps and sequence homology (Eyre et al., 2007; Wheeler et al., 2006; Shah et al., 2005) and the sequences for each gene were recovered from RefSeq. Next, mouse-rat orthologs were identified by Blast (Altschul et al., 1997; Altschul et al., 2005). Rat orthologs to mouse genes were identified as those that were reciprocal best hits when compared with a mouse data file (from RefSeq) (Wheeler et al., 2006) using nucleotide Blast, with an e-value cut off of 1 x 10-50. Coding region sequences annotated from RefSeq were used to speed up the Blast search. In many cases, other supporting information was available to corroborate orthology (e.g., gene name annotation or orthology reported in other literature). From SPEED, a total of 13,506 groups containing orthologs in primate species, with 10,929 of these also containing mouse and rat orthologs. Alignments were re-checked utilizing CLUSTAL on translated sequences frame-aligned with nucleotide sequences, preserving gaps in the amino acid alignment (Thompson et al., 2002; Chenna et al., 2003). After considering only our confirmed ortholog alignments and removing cases where translation frame was uncertain, a total of 10,465 genes were available for analysis. Preliminary data analysis was performed using BioVassa, a statistical tool created by Vassa Informatics to assess information content (Moldover et al., 2009).
We determined the synonymous substitution rate (Ks) and the nonsynonymous substitution rate (Ka) using the (Li, 1993) method as employed in the Diverge program from Accelrys (San Diego) This ratio is commonly employed as a measure of constraint on gene evolution. Alternative methodologies for determining Ka/Ks were utilized and produced similar results throughout.
Target gene identification and functional sorting: Lists of targets were gathered by utilizing literature searches and lists of known drug targets gathered from public data sources (Bjarnadottir et al., 2006; Vassilatis et al., 2003; Kostich et al., 2002; Manning et al., 2002). Successful targets refer to genes utilized for drug development that lead to successfully marketed drugs and Research targets were identified as those candidate genes which are within current research pipelines but for which no drug has been marketed. Genes were identified as GPCRs (Bjarnadottir et al., 2006; Vassilatis et al., 2003; Takeda et al., 2002) or Protein Kinases (Kostich et al., 2002; Manning et al., 2002) based upon a review of their function from literature, Gene Ontology (GO) terms (Ashburner et al., 2000) (searched August, 2009) and Ingenuity Pathway Analysis (Leschly and Basset, 2009).
Statistical evaluation: Statistical analysis was performed utilizing Microsoft Excel (11.6560.6568) to examine variance and covariance of evolutionary rates within and between gene sets. Genes with missing data were excluded from these analyses.
RESULTS
An analysis of successful drug targets versus a representative set of genes showed a highly significant difference in the evolutionary rates of these gene sets (p-value, 0.003, Table 1). Raw data tables are available upon request.
This is mirrored when successful targets which are GPCRs or Kinases are examined (p-value, 0.046 and 0.047, respectively; Table 2).
| Table 1: | Comparison of primate ortholog in successful drug target
genes vs. non-target or all genes |
| Table 2: | Comparison of primate ortholog in successful GPCR and protein
kinase drug target vs. non-targets or all members of the family |
| Table 3: | Human primate orthologs of disease genes |
| Fig. 1: | Gene and publication dates, three categories of genes were
sampled, successful, research and unknown, p value among all pairs were
>0.05. This chart shows the number of genes in the successful and research
categories and the earliest publication date relating to that gene or protein |
To compare the human primate conservation of disease genes with drug target genes, we collected 71 confirmed disease-causal genes from Online Mendelian Inheritance in Man (OMIM, 2006) (OMIM) and 34 disease-associated genes from Genome Wide Association Studies (Altshuler et al., 2008) (GWAS) (with coding-nonsynonymous SNP and a p-value <= 1E-3). In contrast to the successful drug targets, the human primate Ka/Ks of disease genes does not show statistically significant difference when compared to all genes (Table 3).
We further considered whether or not there was ascertainment bias caused by publication date. The hypothesis here was that more conserved genes may have been under examination longer due to genetic manipulations of lower organisms that became available at an earlier date. These genes may be more likely to be successful drug targets, due solely to the amount of time they have been studied. Our analysis showed that there was no correlation between date of publication for a gene and the presence of a gene in the successful drug targets set (Fig. 1). When 168 genes were sampled randomly, both research (n = 47) and successful (n = 43) gene targets showed a similar pattern of publication dates; this is mirrored when a set of 78 non-targets was examined (p-value >0.05 for all comparison).
DISCUSSION
Disease genes are significantly more rapidly evolving than non-disease genes, particularly when tissue specificity is considered (Smith and Eyre-Walker, 2003; Blekhman et al., 2008; Holbrook and Sanseau, 2007; Vamathevan et al., 2008). Our analysis shows that genes which have successfully been developed into drug targets are, however, generally more conserved. Discovery of genes whose proteins are targets for successful drug development is one of the lynchpins of pharmaceutical research. Mirroring this, the goal of many human disease-mapping studies is the identification and analysis of genes where variant alleles cause protein products causal or contributory to human disease conditions. There is, in many cases, an inherent assumption that discovering a human disease gene will directly lead to a candidate gene for pharmaceutical research. Our results, taken in the context of previous literature, suggest otherwise. Our analysis strongly suggests that the process of treating a disease with a drug is very distinct from the process of identifying causal genes. That is, while variations in rapidly-evolving genes may tend to cause disease, it is more clock-like, slowly evolving genes that are better candidates as drug targets. There are a number of reasons why this may be so.
Tissue specificity is a likely culprit. Several studies found that genes expressed in specific tissues are more likely to be rapidly evolving than those expressed in multiple tissues (Duret et al., 2002; Duret and Mouchiroud, 2000; Duret and Mouchiroud, 1999; Duret et al., 1995; Duret et al., 1994). However, housekeeping genes, which are not often found to be involved in human diseases, are typically highly conservative; likely, diseases caused by these conserved genes are so, deleterious that they are incompatible with development and viability (Santibanez Koref et al., 2003). However, conservative genes appear to be better targets for drug development, given our results. This brings up two related issues: first, that conservative genes are likely to have broader tissue expression and secondly that they may be involved with basal functionality (Duret and Mouchiroud, 2000). More conserved genes may represent suitable drug development targets as behavior between species may be more predictable. Agents may bind species homologs with similar affinity and in similar orientations, making transition from pre-clinical primate model data to clinical human data more reliable.
Our analysis specifically explored the possibility that specific functions may be contributing to the observed bias in evolutionary rate in drug target genes. Our analysis examines, specifically, GPCR and Kinase proteins. In this case, the observed difference between successful targets and all proteins remained in these subclasses. This suggests that specific functionality is not the likely explanation for the observed evolutionary rate bias.
We also specifically were concerned with ascertainment bias due to length of study of specific proteins. Our analysis ruled out this bias, suggesting that “first in, first out” drug development is not necessarily the norm; that is, that those proteins that have been studied the longest are not likely to yield drugs more rapidly than novel targets. Fresh drug targets, chosen rationally based on conservation, may be as likely to eventually generate a drug as longer studied proteins. This approach may be extensible to human disease gene datasets (Bakheet and Doig, 2009) as well as databases of other species, such as human pathogens (Shakyawar et al., 2011) and even plant pathogens (Kizil et al., 2005).
This finding explains, in part, why the large number of human disease genes identified has not led to a similar explosion in successful drug targets. It also suggests a rational approach to determining new targets for drug development following the identification of a gene involved with human disease. Instead of studying the human disease gene and suspecting it will lead to a drug, these results suggest that analyzing for example the pathway that the gene is in and picking a conservative protein within that pathway may be a better approach to drug design. For example, polymorphism of endothelin 1, a potent vasoconstrictor, has been associated with hypertension (Jin et al., 2003). When the human-chimpanzee Ka/Ks value was overlayed on endothelin 1 pathway in Ingenuity Pathway Analysis (http://www.ingenuity.com) (Leschly and Basset, 2009) (Fig. 2 and Table 4), it was noted that its receptor, a GPCR named endothelin receptor type B (EDNRB or ETB), has a much smaller human-chimpanzee Ka/Ks value (0.307) than endothelin 1 (1.657), making it a potentially more successful drug target than endothelin 1.
| Fig. 2: | The endothelin-1 signaling pathway, The Ka/Ks values (in
small numbers) are high for ET1 (Endothelin-1), but low for EDNRB (labelled
as ETE). Bosentan, a recently licensed drug targeted against hypertension,
targets EDNRB |
| Table 4: | Human primate ortholog in endothelin 1 pathway |
Indeed, consistent with our hypothesis, a known inhibitor of EDNRB, Bosentan (Tracleer) has been approved for pulmonary hypertension (Krum et al., 1998).