OBJECTIVES: The goal of this study was to investigate the association of perceived discrimination with health outcomes and disparities. MATERIALS AND METHODS: The study cohort consists of 60 180 participants from the 4 largest self-identified race and ethnicity (SIRE) groups in the All of Us Research Program participant body: Asian (1291), Black (4726), Hispanic (5336), and White (48 827). A perceived discrimination index (PDI) was derived from participant responses to the "Social Determinants of Health" survey, and the All of Us Researcher Workbench was used to analyze associations and mediation effects of PDI and SIRE with 1755 diseases. RESULTS: The Black SIRE group has the greatest median PDI, followed by the Asian, Hispanic, and White groups. The Black SIRE group shows the greatest number of diseases with elevated risk relative to the White reference group, followed by the Hispanic and Asian groups. Perceived discrimination index was found to be positively and significantly associated with 489 out of 1755 (27.86%) diseases. "Mental Disorders" is the disease category with the greatest proportion of diseases positively and significantly associated with PDI: 59 out of 72 (81.94%) diseases. Mediation analysis showed that PDI mediates 69 out of 351 (19.66%) Black-White disease disparities. DISCUSSION: Perceived discrimination is significantly associated with risk for numerous diseases and mediates Black-White disease disparities in the All of Us participant cohort. CONCLUSION: This work highlights the role of discrimination as an important social determinant of health and provides a means by which it can be quantified and modeled on the All of Us platform.

Transposable elements (TEs) are an important source of human genetic variation with demonstrable effects on phenotype. Recently, a number of computational methods for the detection of polymorphic TE (polyTE) insertion sites from next-generation sequence data have been developed. The use of such tools will become increasingly important as the pace of human genome sequencing accelerates. For this report, we performed a comparative benchmarking and validation analysis of polyTE detection tools in an effort to inform their selection and use by the TE research community. We analyzed a core set of seven tools with respect to ease of use and accessibility, polyTE detection performance and runtime parameters. An experimentally validated set of 893 human polyTE insertions was used for this purpose, along with a series of simulated data sets that allowed us to assess the impact of sequence coverage on tool performance. The recently developed tool MELT showed the best overall performance followed by Mobster and then RetroSeq. PolyTE detection tools can best detect Alu insertion events in the human genome with reduced reliability for L1 insertions and substantially lowered performance for SVA insertions. We also show evidence that different polyTE detection tools are complementary with respect to their ability to detect a complete set of insertion events. Accordingly, a combined approach, coupled with manual inspection of individual results, may yield the best overall performance. In addition to the benchmarking results, we also provide notes on tool installation and usage as well as suggestions for future polyTE detection algorithm development.

Rapid and accurate identification of the sequence type (ST) of bacterial pathogens is critical for epidemiological surveillance and outbreak control. Cheaper and faster next-generation sequencing (NGS) technologies have taken preference over the traditional method of amplicon sequencing for multilocus sequence typing (MLST). But data generated by NGS platforms necessitate quality control, genome assembly and sequence similarity searching before an isolate's ST can be determined. These are computationally intensive and time consuming steps, which are not ideally suited for real-time molecular epidemiology. Here, we present stringMLST, an assembly- and alignment-free, lightweight, platform-independent program capable of rapidly typing bacterial isolates directly from raw sequence reads. The program implements a simple hash table data structure to find exact matches between short sequence strings (k-mers) and an MLST allele library. We show that stringMLST is more accurate, and order of magnitude faster, than its contemporary genome-based ST detection tools. AVAILABILITY AND IMPLEMENTATION: The source code and documentations are available at http://jordan.biology.gatech.edu/page/software/stringMLST CONTACT: [email protected] information: Supplementary data are available at Bioinformatics online.

MOTIVATION: It has been suggested that presumably distinct classes of genomic regulatory elements may actually share common sets of features and mechanisms. However, there has been no genome-wide assessment of the prevalence of this phenomenon. RESULTS: To evaluate this possibility, we performed a bioinformatic screen for the existence of compound regulatory elements in the human genome. We identified numerous such colocated boundary and enhancer elements from human CD4(+) T cells. We report evidence that such compound regulatory elements possess unique chromatin features and facilitate cell type-specific functions related to inflammation and immune response in CD4(+) T cells.

SUMMARY: Although some histone modification chromatin immunoprecipitation followed by high-throughput sequencing (ChIP-seq) signals show abrupt peaks across narrow and specific genomic locations, others have diffuse distributions along chromosomes, and their large contiguous enrichment landscapes are better modeled as broad peaks. Here, we present BroadPeak, an algorithm for the identification of such broad peaks from diffuse ChIP-seq datasets. We show that BroadPeak is a linear time algorithm that requires only two parameters, and we validate its performance on real and simulated histone modification ChIP-seq datasets. BroadPeak calls peaks that are highly coincident with both the underlying ChIP-seq tag count distributions and relevant biological features, such as the gene bodies of actively transcribed genes, and it shows superior overall recall and precision of known broad peaks from simulated datasets. AVAILABILITY: The source code and documentations are available at http://jordan.biology.gatech.edu/page/software/broadpeak/.

MOTIVATION: New sequencing technologies have accelerated research on prokaryotic genomes and have made genome sequencing operations outside major genome sequencing centers routine. However, no off-the-shelf solution exists for the combined assembly, gene prediction, genome annotation and data presentation necessary to interpret sequencing data. The resulting requirement to invest significant resources into custom informatics support for genome sequencing projects remains a major impediment to the accessibility of high-throughput sequence data. RESULTS: We present a self-contained, automated high-throughput open source genome sequencing and computational genomics pipeline suitable for prokaryotic sequencing projects. The pipeline has been used at the Georgia Institute of Technology and the Centers for Disease Control and Prevention for the analysis of Neisseria meningitidis and Bordetella bronchiseptica genomes. The pipeline is capable of enhanced or manually assisted reference-based assembly using multiple assemblers and modes; gene predictor combining; and functional annotation of genes and gene products. Because every component of the pipeline is executed on a local machine with no need to access resources over the Internet, the pipeline is suitable for projects of a sensitive nature. Annotation of virulence-related features makes the pipeline particularly useful for projects working with pathogenic prokaryotes. AVAILABILITY AND IMPLEMENTATION: The pipeline is licensed under the open-source GNU General Public License and available at the Georgia Tech Neisseria Base (http://nbase.biology.gatech.edu/). The pipeline is implemented with a combination of Perl, Bourne Shell and MySQL and is compatible with Linux and other Unix systems.

MOTIVATION: Chromatin immunoprecipitation followed by high-throughput sequencing (ChIP-seq) is widely used in biological research. ChIP-seq experiments yield many ambiguous tags that can be mapped with equal probability to multiple genomic sites. Such ambiguous tags are typically eliminated from consideration resulting in a potential loss of important biological information. RESULTS: We have developed a Gibbs sampling-based algorithm for the genomic mapping of ambiguous sequence tags. Our algorithm relies on the local genomic tag context to guide the mapping of ambiguous tags. The Gibbs sampling procedure we use simultaneously maps ambiguous tags and updates the probabilities used to infer correct tag map positions. We show that our algorithm is able to correctly map more ambiguous tags than existing mapping methods. Our approach is also able to uncover mapped genomic sites from highly repetitive sequences that can not be detected based on unique tags alone, including transposable elements, segmental duplications and peri-centromeric regions. This mapping approach should prove to be useful for increasing biological knowledge on the too often neglected repetitive genomic regions. AVAILABILITY: http://esbg.gatech.edu/jordan/software/map CONTACT: [email protected] SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

MOTIVATION: Endogenous retrovirus (ERV) elements have been shown to contribute promoter sequences that can initiate transcription of adjacent human genes. However, the extent to which retroviral sequences initiate transcription within the human genome is currently unknown. We analyzed genome sequence and high-throughput expression data to systematically evaluate the presence of retroviral promoters in the human genome. RESULTS: We report the existence of 51,197 ERV-derived promoter sequences that initiate transcription within the human genome, including 1743 cases where transcription is initiated from ERV sequences that are located in gene proximal promoter or 5' untranslated regions (UTRs). A total of 114 of the ERV-derived transcription start sites can be demonstrated to drive transcription of 97 human genes, producing chimeric transcripts that are initiated within ERV long terminal repeat (LTR) sequences and read-through into known gene sequences. ERV promoters drive tissue-specific and lineage-specific patterns of gene expression and contribute to expression divergence between paralogs. These data illustrate the potential of retroviral sequences to regulate human transcription on a large scale consistent with a substantial effect of ERVs on the function and evolution of the human genome.

Briefings in Bioinformatics aims to provide working biologists with an awareness and understanding of the computational approaches available for research and discovery. The Abstracts section of the journal consists of summaries of bioinformatics manuscripts published in the preceding two months. Inclusion of an article in this section indicates that the editors consider it to be among the most interesting and/or useful contributions to the field for the quarter covered. The contents of these reports are briefly distilled for the readers with an emphasis placed on their biological context and potential utility. Publications from October and November 2006 are reviewed here. Alan M. Moses, Daniel A. Pollard, David A. Nix, Venky N. Iyer, Xiao-Yong Li, Mark D. Biggin and Michael B. Eisen PLoS Computational Biology (2008) Vol. 2, no. 10, p. e130 The importance of changes in gene regulation for organismic evolution—as opposed to changes in gene coding sequences per se—has long been recognized. However, sequences that control regulation tend to be less well characterized than coding sequences and hence not as intensively studied. The paper of Moses et al. presents a large-scale evaluation of the rate of gain and loss of Drosophila cis-regulatory transcription factor binding sites (TFBS). This work is notable for its combination of experimental and computational analyses along with the analytical rigor that is brought to bear on the question of TFBS evolution. The authors focused on the evolutionary turnover (gain and loss) of TFBS for a single transcription factor Zeste, but considered these dynamics on a whole genome scale. The genome scale comparative analysis was facilitated by the use of the high-throughput ChIP-chip experimental technique to identify genomic sites bound by Zeste as well as the availability of a number of complete genome sequences for closely related Drosophila species. The authors identified 294 Drosophila melanogaster genomic regions that were bound by Zeste and had corresponding orthologous sequences in three other closely Drosophila species. Evaluation of these regions with a Zeste position weight matrix led to the identification of 1,406 putative Zeste binding sites. In order to detect evolutionary turnover of these binding sites, the authors developed a probabilistic model that represents the position-specific evolutionary rates of Zeste sites based on the factors known binding specificity. Observation of the position-specific variation observed when the putative Zeste TFBS they identified were compared across genomes showed that Zeste sites evolve more rapidly than predicted by the binding site model but less rapidly than predicted by a neutral model. In other words, their population of putative Zeste sites includes conserved sites as well as nonconserved sites that have been lost or gained over time. A likelihood ratio test, based on the Zeste TFBS evolution model, was then applied to test whether each Zeste site was more likely to be conserved or nonconserved. This test revealed 215 (>15%) putative Zeste binding sites to show statistically significant nonconserved signals. A number of additional statistical tests were used to validate this result and in particular, to control for errors in sequence alignment that could confound the analysis. The rapid turnover of TFBS revealed by their analysis holds up well to this scrutiny. They come up with a fairly conservative estimate of 5% turnover in Zeste sites over the last 10 million years. The focus of most evolutionarily informed studies on regulatory sequence has been placed on the sequences that are conserved between genomes. This manuscript, along with other recent studies, shows how rapidly TFBS sequences evolve and underscores the importance of regulatory sequences that are diverged rather than conserved over time. After all, it is these nonconserved TFBS sequences that are most likely to contribute to the regulatory changes between species that are so important in evolution. Lars Juhl Jensen, Thomas Skøt Jensen, Ulrik de Lichtenberg, Søren Brunak and Peer Bork Nature (2006) Vol. 443, no. 7111, pp. 594–597 This large-scale study on the evolution of molecular complexes that function in the cell cycle has profound implications for the understanding of how biological systems evolve. Apparently, there are very different rules involved in governing evolutionary dynamics at different levels of biological organization. Jensen et al. performed a comparative analysis of the regulatory versus coding sequence evolution for genes that have periodic expression along the stages of the cell cycle. They took advantage of numerous large-scale gene expression data sets to identify hundreds of periodically expressed genes among four divergent eukaryotes: human, two yeasts Saccharomyces. cerevisiae and S. pombe and the plant Arabidopsis thaliana. At the same time, they also identified gene sets that contain orthologs of periodically expressed genes for the four organisms. The majority of orthologous gene sets that contain at least one periodically expressed member do not have orthologs in all four species. In addition, for those orthologous sets with at least one periodically expressed member and orthologs in all four species, expression periodicity was not conserved. Only five orthologous sets, out of 381 with at least one periodically expressed member, show periodic expression in all four species. In other words, even when sequences are demonstrably conserved between lineages, gene expression patterns are not. Less surprisingly, the sets with conserved expression and sequences correspond to essential regulators of the cell cycle or genes such as histones that encode the building blocks of eukaryotic chromatin. However, even for gene sets where the periodic expression is conserved, the actual timing of this expression, i.e. in which phase of the cell cycle the genes are expressed, can often differ. The authors also reveal that many of the changes in periodic expression of cell cycle genes are accompanied by changes in post-translational regulation, phosphorylation in particular, of the encoded proteins. This post-translational regulation adds another layer of control for cell cycle complexes and helps to ensure that the appropriate complexes are assembled ‘just in time’ to play their functional role. Often these functional complexes are assembled from a number of proteins encoded by constitutively expressed genes along with one or a few members that are periodically expressed and/or post-transcriptionally modified. The genes with periodic expression and post-translational modifications may act as triggers to ensure the proper timing and function of the complexes, and evolutionary conservation of the specific identity of the triggers is not as important as their existence. Furthermore, they show evidence that different levels of regulation co-evolved, since periodically expressed genes are more likely to be post-translationally regulated. The authors conclude that many different solutions to the problem of assembling the same molecular complexes at the appropriate point along the cell cycle have evolved over time. This indicates that while the protein-coding component of these molecular complexes is highly conserved, the regulatory part is far more dynamic. Thus, regulatory innovations may be a much more viable substrate for adaptive divergence than changes in gene coding sequence and protein structure. Yuri I. Wolf, Cecile Viboud, Edward C. Holmes, Eugene V. Koonin and David J. Lipman Biology Direct (2006) Vol. 1, p. 34 Yuri Wolf and colleagues have published the rare molecular evolution study that has both fundamental implications for the understanding of how natural selection is manifest at the level of gene sequences as well as urgent human health significance. Both of these accomplishments are in no small way related to their choice of study—the human influenza virus and its hemagglutinin (HA) encoding gene. Influenza epidemics lead to thousands of deaths every year, and the HA gene is thought to evolve rapidly via positive Darwinian selection that is predicated upon changing antigenicity of the HA epitope regions. However, it is the combination of an impressively large-scale analysis along with the judicious distinction of rapid versus static evolutionary regimes that allowed for the heterodox conclusions of this work. In short, the authors find that HA evolution is characterized by long periods of stasis, which show basically no evidence of positive selection, followed by short periods of rapid evolution under strong positive selection. This finding stands in stark contrast to the prevailing views on influenza virus evolution, which emphasize that the rapid evolution of HA is driven by continuous positive selection. It seems that the particular data set analyzed here may also have much to do with the novel conclusions reached. The authors took advantage of the existence of more than 1,000 publicly available influenza genomes that were sequenced as part of the Influenza Genome Sequencing Project (IGSP) between 1995–2005. Unlike many other influenza sequences that have been characterized because of their seriological novelty, the IGSP sequence data analyzed here are unbiased in the sense that they are not enriched for antigenically novel isolates. For the H3N2 influenza subtype, periods of stasis are marked by a higher rate of synonymous than non-synonymous substitutions, no demonstrable overrepresentation of HA amino acid changes in the epitope regions of the protein and slow extinction of co-circulating viruses from different lineages. On the other hand, short and rapid evolutionary phases are dominated by positive selection as demonstrated by an excess of amino acid changes in the HA epitope regions relative to the rest of the protein, and a relatively lower ratio of synonyomous to nonsynonymous substitution rate. In the rapid evolutionary phase, newly arisen dominant lineages quickly lead to the extinction of the previously coexisting lineages. There were also a number of cases of parallel amino acid substitutions in different lineages shown to have occurred during the rapid evolutionary phase. Independent parallel substitutions across lineages are also consistent with positive selection and may be important for vaccine development since these could be used as predictors of new dominant strains. It should be noted that this clear case of punctuated evolution is complicated by the H1N1 strains, which show no evidence of positive selection yet manage somehow to out compete H3N2 lineages in some seasons. In this case, the H3N2 lineages may be so successful as to limit the availability of susceptible hosts thus opening up a new niche for H1N1. Shyam Prabhakar, James P. Noonan, Svante Pääbo, and Edward M. Rubin Science (2006) Vol. 314, no. 5800, pp. 786 Only a tiny fraction of the human genome actually codes for protein sequences, and the functional relevance of the remaining noncoding sequence is an open question that has attracted quite a bit of attention as of late. Investigators have been taking advantage of comparative sequence data to hone in on the anomalously conserved regions of noncoding genomic sequence under the rationale that such regions are most likely to be functionally germane. However, from an evolutionary perspective such sequences may actually be the least interesting of all because they do not change (or change very little) between species. Thus, focusing on conserved sequences, such as when looking for transcription factor binding sites, may actually be positively misleading in terms of identifying evolutionarily important sites. Fortunately, after a few years of almost exclusive focus on conserved noncoding regions, several groups have begun to look for regions that show accelerated regions and this short paper by Prabhakar et al. exemplifies this approach. Interestingly, in looking for regions that change rapidly between species the authors have used regions previously identified as highly conserved as a starting point. Specifically, they evaluated more than 100,000 conserved noncoding sequences for evidence of human-specific substitutions using a statistical null model based on a pattern of constrained evolution. This approach yielded 992 distinct genomic sequences showing statistically significant evidence for an excess of human-specific substitutions. The authors went on to associate these regions with human genes and used their Gene Ontology (GO) annotations to check for enrichment of specific functional categories. The human gene set associated with accelerated regions was found to be enriched for the biological process GO term cell adhesion, and many of these genes were specifically involved in neuronal cell adhesion, e.g. cadherins and neuroligins. This finding has obvious implications for the evolution of uniquely human cognitive traits. As a control, the authors also looked for accelerated regions using the same approach in mouse and chimpanzee. Interestingly, while the accelerated sequences identified in chimpanzees were distinct from those identified in human, they too showed a significant enrichment near genes involved in neuronal adhesion. However, because the accelerated regions specific to human and chimpanzee neuronal adhesion genes are quite different, the authors argue that the phenotypic effects are likely to be quite different between the two lineages. In any case, it is intriguing that independent changes in the nervous system may have been particularly important in the evolutionary diversification of both species. Katherine S. Pollard, Sofie R. Salama, Bryan King, Andrew D. Kern, Tim Dreszer, Sol Katzman, Adam Siepel, Jakob S. Pedersen, Gill Bejerano, Robert Baertsch, Kate R. Rosenbloom, Jim Kent and David Haussler PLoS Genetics (2006) Vol. 2, no. 10, p. e168 Comparative genomic sequence analyses, in particular those involving the recently completed chimpanzee sequence, allow for the identification of the evolutionary sequence changes that may have led to uniquely human characteristics. This group led by David Haussler was among the first to identify so-called ‘ultra-conserved’ elements in the human genome. These are human genome sequences that were inferred by comparison with other genomes to have changed very little over evolutionary time. Such sequences are of interest because they are likely to have been conserved by natural selection because they are functionally important. Now this idea has been turned on its head and the emphasis is placed on genomic regions that are changing more rapidly than others. These accelerated regions can be inferred to be of outstanding evolutionary interest since they may lead to differences between lineages—in particular changes along the human lineage that may distinguish us from other closely related primate species. However, to get a handle on such regions in the human genome this team still relies on the presence of ultra-conserved regions. Specifically, what they do is identify genomic regions that are highly conserved in other species (chimp, mouse and rat) yet highly divergent in human. Using a statistical model to detect significant accelerations along the human evolutionary lineage, they were able to detect 202 genomic sequences that are highly conserved among other vertebrates yet divergent in humans. As with other highly conserved regions, these sequences are mostly noncoding. Analysis of the functional annotations of nearby genes revealed that many of these regions are close to genes involved in transcription and DNA binding. This underscores the importance of changes in the regulation of genes that are themselves in turn responsible for the regulation of other genes. In other words, a few pointed and specific sequence changes could lead to complex cascades of regulatory effects. The general assumption for these types of studies is that natural selection is a major player in determining rates of evolution. One of the most interesting aspects of this study was that the authors played close attention to sequence context and the actual kinds of substitutions that occur in accelerated regions. For instance, they show a preponderance of changes that result in the creation of guanine and cytosine as well as an overrepresentation of accelerated sequences in regions of high recombination. This suggests the possibility that other forces, such as biased gene conversion, may also play a role in accelerations along the human evolutionary lineage. Mark Pagel, Chris Venditti and Andrew Meade Science (2006) Vol. 314, no. 5796, pp. 119–121 The foundations of modern evolutionary theory—the so-called ‘New Synthesis’ or ‘Neo-Darwinian Synthesis’—were laid down in the middle of the 20th century when Mendelian genetics were united with Darwinian theory. This elaborate body of formal theory detailed the how and why allele frequencies could change within and between populations of a given species. It was assumed that an understanding such microevolutionary—i.e. within species—changes were both necessary and sufficient to explain longer term macroevolutionary changes between species. This model, which quickly became orthodoxy—and to some even dogma—strongly implied that evolution was a long process of steady and gradual change. The theory of punctuated equilibrium, which holds that macroevolution is better characterized by long periods of stasis followed by short bursts of rapid change that accompany speciation, is a fundamental challenge to Neo-Darwinian orthodoxy. Pagel et al. set out to assess the relative contributions of gradual versus punctuated change to genomic sequence changes by using a large-scale phylogenetic approach. The strength of their approach is the use of two simple and mutually exclusive phylogenetic models that discriminate between gradual and punctual modes of evolution. The model rests on an evaluation of path lengths, which are the sums of the branch lengths from the roots of phylogenetic trees to the tips of the trees. Since gene sequence alignments were used to build the phylogenetic trees under consideration here, the path lengths are expressed as the number of nucleotide substitutions per site. According to the particular branching pattern of the phylogeny under consideration, there will be more or fewer nodes along the different paths from the roots to the tips. Since the nodes represent speciation events, the punctuated model predicts that paths that cross more nodes should be longer than paths that traverse fewer nodes. The gradual model, on the other hand, predicts no such correlation between node number and path length.122 gene sequence alignments were analyzed according to a statistical model that evaluates the punctuated contribution to the observed path lengths. The authors found that 47% of these data sets were consistent with the punctuated model; after a correction for artifacts this figure fell to 35%. In addition, punctuated evolution appears to be much more common in plants and fungi than in animals. Given their finding that punctuated evolution makes a substantial contribution to sequence changes, the authors next tried to estimate the extent of this contribution to overall sequence diversity. Just over one-fifth (22%) of DNA sequence substitutions were attributed to punctuated evolution. The take home message of this work is that punctuated change has been far more important for sequence variation than previously thought and thus should be accounted for in models of molecular evolution. On a technical note, such punctuated change will lead to violations of clock-like molecular evolution and so should be accounted for when using molecular data to estimate divergence times between evolutionary lineages. It should be noted that while the authors observe a strong signal for punctuated molecular evolution, they do not see any evidence for long periods of stasis, the other part of the punctuated equilibrium picture.

Briefings in Bioinformatics aims to provide working biologists with an awareness and understanding of the computational approaches available for research and discovery. The Abstracts section of the journal consists of summaries of bioinformatics manuscripts published in the preceding 2 months. Inclusion of an article in this section indicates that the editors consider it to be among the most interesting and/or useful contributions to the field for the quarter covered. The contents of these reports are briefly distilled for the readers with an emphasis placed on their biological context and potential utility. Publications from December 2006 and January 2007 are reviewed here. The (post)-genomics era in biology has often been marked by unexpected discoveries driven by new technologies. This phenomenon is exemplified by the cap analysis of gene expression (CAGE) technique. Recent results of CAGE surveys on the human and mouse genomes have the potential to fundamentally alter our understanding of eukaryotic gene regulation. Kawaji et al. have performed an illuminating computational follow-up on previous CAGE experimental studies to further explore the implications of these data. CAGE is a high-throughput sequence-tag based technique that allows for the simultaneous identification of transcriptional start sites (TSSs) and measuring of gene expression levels. Briefly, CAGE employs the characterization of 20–21 base pair sequence tags that correspond to the very 5′ ends of full length cDNAs. Analyzing thousands of CAGE tags reveals the expression profiles of numerous TSSs. The most surprising result from the large scale application of the CAGE technique to the human and mouse genomes was the finding that mammalian transcription is almost never initiated from a single precisely defined TSS. Rather, genes are transcribed from an array of different TSSs; in some cases these TSSs may be tightly clustered and in others they may be broadly distributed. Kawaji et al. sought to address the question of whether and how the selection of TSSs is regulated. To do this, they framed two hypotheses with mutually exclusive expectations regarding the use of alternative TSSs across different tissues. Their null hypothesis rested on the supposition that if TSS selection is governed mainly by genomic sequence context, then it would not change appreciably between different tissue or cell types. Alternatively, if TSS selection is differentially regulated, whether by different transcription factors or epigenetic modifications, then TSSs may differ systematically between tissues. To discriminate between these two hypotheses, they analyzed CAGE data from 22 different mouse tissues. This revealed evidence for tissue-specific TSS utilization for about half of all CAGE tag clusters analyzed. In other words, TSS selection is clearly regulated for numerous genes. Furthermore, the fraction of genes with regulated TSS usage is likely to be an underestimate since the number of different tissues they analyzed is in no way exhaustive. Genes that are prone to differential TSS usage include those with CpG island-containing promoters as well as genes with multi modal promoter structures characterized by two or more peaked CAGE tag clusters. For the most part, the specific regulatory mechanisms that give rise to tissue-specific TSS use lie beyond the reach of the computational approach employed here. However, one intriguing possibility arises from the authors’ demonstration that numerous imprinted genes (i.e. genes with different expression according to the parent of origin) use alternate TSSs. This finding points to a connection between TSS selection and epigenetic modifications such as methylation and/or chromatin modification. It has long been recognized that changes in gene regulation probably underlie many important aspects of phenotypic evolutionary divergence. However, it has been difficult to systematically analyze the genetic bases of changes in gene regulation and expression due to a lack of comparative data. The increasing availability of genomic sequence data from numerous species, combined with accumulating experimental data on gene regulation, is providing new opportunities to study the evolution of gene regulation. In this manuscript, Mariño-Ramírez and colleagues report surprisingly paradoxical results regarding the evolution of the core histone gene regulatory mechanisms. There are four families of core histone genes, which together encode the proteins that assemble as an octamer to form the nucleosome—the fundamental unit of eukaryotic chromatin. Core histone genes are deeply conserved across eukaryotes having changed relatively little in terms of both sequence and expression pattern from the yeast Saccharomyces cerevisiae to human. Core histone genes show periodic expression across the eukaryotic cell cycle with a pronounced peak during S-phase. This S-phase specific expression pattern allows for histone proteins to be produced at the same time DNA is being synthesized so that it can be readily bound by nucleosomes and compacted into chromatin. The authors of this report show that, despite the conservation of core histone gene expression patterns, the cis-trans regulatory machinery that controls core histone gene expression has changed greatly among eukaryotic evolutionary lineages. Specifically, the identity of the core histone gene cis-regulatory sequence motifs and the trans protein factors that bind them are distinct for the yeasts S. cerevisiae and Schizosaccharomyces pombe as well as for other fungi, plants, insects and mammals. In other words, while core histone gene expression has remained unchanged since these species last shared a common ancestor, the way these genes are regulated has been constantly reinvented along different evolutionary lineages. In addition to this rapid turnover of regulatory mechanisms, the authors also show that within evolutionary lineages different core histone gene families employ basically the same regulatory machinery. This is the exact opposite of the evolutionary pattern seen for core histone gene (protein) sequences where members of the same gene family are more similar to one another than they are to members of other families found in the same species. Thus, the regulatory machinery has changed between evolutionarily lineages and has been homogenized within lineages. Taken together these results underscore just how dynamic the evolution of gene regulation can be. Regulatory systems are somehow able to undergo wholesale changes even in the face of pressure to maintain the same expression patterns. The concerted evolution of the regulatory systems within lineages suggests that this mode of evolution may be due to epistatic pressures that are exerted on regulatory components that play multiple roles. Decades of research in molecular evolution have shown that rates of gene (protein) evolution are largely determined by purifying natural selection, which consists of the removal of deleterious variants. However, the biological source of this selective constraint was not systematically evaluated until recently. This advance was made possible by the combination of complete genome sequences of related organisms and high-throughput experimental data such as protein interaction data, gene expression profiles and gene dispensability measures. Using thousands of sequence comparisons and large scale functional data sets, researchers performed many correlations between genes’ evolutionary rates and their functional characteristics. Numerous, relatively weak, correlations were found. For instance, essential genes were found to evolve more slowly than nonessential genes, more highly expressed genes were shown to evolve more slowly than genes with lower expression levels, and it was demonstrated that proteins that interact with multiple partners evolve more slowly than those with fewer interactions. However, it quickly became apparent that many of these functional variables were also correlated with one another. Highly expressed genes tend to have more interaction partners and be more essential than lowly expressed genes. These cross-connections together with the general weakness observed for the correlations between evolutionary rates and functional measures have obscured the biological significance of these results. Shakhnovich and Koonin have uncovered a novel determinant of selective constraint related to gene family evolution that may prove to be one of the strongest determinants of evolutionary rates known to date. To do this, they partitioned families of related genes into two classes: those that contain at least one essential gene (E-families) and those that contain no essential genes (N-families). The members of N-families were found to be subject to weaker selective constraint, and as a result they are more likely to be fixed and/or to become pseudogenes with no function. The E-families, on the other hand, are subject to greater selective constraint. This means that the genes survive for longer periods of evolutionary time, and accordingly, the E-families accumulate greater evolutionary divergence among their members. Included in this divergence are changes in the upstream regulatory regions that control the expression patterns of E-family genes. Indeed, E-family genes are far less likely to share transcription factor binding sites than N-family genes. This finding is consistent with the evolution of novel functions by E-family genes. The authors conclude that E-family genes typically evolve through the acquisition of new or altered functions while N-family genes lose functionality. Biological function is carried out by complex networks of interacting players including, but not limited to, genes, proteins, metabolites and even individual organisms. In the last few years, the analytical tools of graph theory have been brought to bear on these biological networks resulting in some unexpected findings. Perhaps the most intensively studied networks have been those comprised of interacting proteins—nodes in such networks correspond to proteins and the nodes are connected by an edge if they can be shown to physically interact. Studies of protein interaction networks have yielded non-trivial revelations about cellular organization and the role of natural selection in evolution. In a recent issue of Science, Kim et al. add an important new wrinkle to this field of work. Previous studies of protein interaction networks have treated physical interactions as binary; either proteins interact or they do not. Kim et al. took a more nuanced approach to protein interaction by considering the three-dimensional structures of interacting proteins. By now it is well appreciated that the topological properties protein interaction networks, their connectivity in particular, are dominated by the so-called hubs, proteins that interact with numerous partners. The authors of this work evaluated the structural properties of such hubs with respect to the locations and identities of their binding interfaces. Apparently, some hub proteins interact with multiple partners using the same binding interface. Such interactions are by definition mutually exclusive; only one particular interaction can occur for any given time or condition. Other proteins interact with multiple partners using distinct non-overlapping interfaces, and these interactions may occur simultaneously. The distinction between singlish—meaning proteins with one or two binding interfaces, versus multi-interface hubs helped the authors to resolve a number of hotly debated issues regarding the relationship between protein interaction network topologies and evolution. For instance, while it had previously been noted that hub proteins evolve more slowly than less connected proteins, this matter was very contentious. Kim et al. show that this signal is largely confined to multi-interface hubs which evolve far more slowly than both non-hub proteins and singlish-interface hubs. In addition, the relatively transient nature for the interactions of singlish-interface hubs was confirmed by gene expression analysis, which showed that multi-interface hubs have more correlated expression patterns with their interaction partners. This work also calls into question a popular model for the evolution of complex networks. Hubs are thought to evolve via preferential attachment of new nodes to already highly connected nodes, and this model has been related to the process of gene duplication. However, this appears to only be the case for singlish-interface hubs consistent with the idea that duplicated proteins would bind to the same interface. Thus, while the added level of detail provided by the structural analysis of Kim et al. has helped to address some outstanding questions, it has also raised some new questions of its own. Transposable elements (TEs) are the single most abundant class of eukaryotic genomic sequence. For instance, 45% of the human genome is made up of TE sequences while only 1.5% corresponds to protein coding sequences. The genomic abundance of TEs can be explained by their selfish properties. Indeed, TEs are most often thought of as genomic parasites that serve only to replicate themselves at the expense of the host genomes in which they reside. However, in recent years numerous examples of TEs that have been ‘domesticated’ to play some functional role for their hosts have been uncovered. In this article, Michel Weber shows that two classes of structural RNA molecules—small nucleolar RNAs (snoRNAs) and small Cajal body-specific RNAs (scaRNAs)—represent novel families of TEs. snoRNAs are involved in the modification of ribosomal RNAs and scaRNAs help to modify spliceosomal RNAs. Using known human and mouse sno/scaRNA sequences as queries, Weber searched for orthologs and paralogs (i.e. within genome duplicates) among sequenced vertebrate genomes and identified many novel sno/scaRNA sequences. snoRNAs are known to be encoded in the introns of host genes and so it was not surprising to find some snoRNA paralogs that had been generated by duplication of their host genes. Far more unexpected was the finding that most snoRNA paralogs showed characteristics of sequences that are generated by retrotransposition—in other words they appeared to have been copied by the reverse transcription of a snoRNA transcript and inserted in a new location in the genome. Specifically, these snoRNA paralogs have polyA tails, target site duplications and insertion sites that can be clearly defined by comparisons with other genome sequences. Weber dubs these sequences snoRNA retroposons or snoRTs. Sequence alignment and phylogenetic analysis of related snoRTs allows for the determination of the original (parental) snoRNA sequence that gave rise to paralogs through retrotransposition. Examination of the genomic structure of these elements revealed three classes of snoRTs that differ with respect to the amount of sequence from the parental host gene that they contain. Weber proposes that the redundancy created by the retrotransposition of snoRNAs helps to both safeguard functional degeneracy of single copy snoRNAs through mutation and to provide increased diversity that could lead to novel functions. With respect to the latter prediction, it is most interesting to note that many of the mobile sno/scaRNA retroposons identified in this study are either lineage- or species-specific. Thus, sno/scaRNA evolution by retrotransposition may be related to between-species differences in RNA-modifications. The importance of non-coding RNA has recently been underscored by the realization that the majority of eukaryotic genome sequences are transcribed, and furthermore, most of these transcripts do not encode proteins. Hundreds of new non-coding RNA (ncRNA) genes are characterized every year, and it looks like the rate of discovery is still in the exponential growth phase. Having said that, the full complement, i.e. the number, size and extent, of ncRNA genes for any given eukaryotic genome is unknown, and knowledge of ncRNA genes lags far behind the current understanding of protein coding gene repertoires. Since most new genes are discovered and characterized via homology relationships with known genes, the application of homology search methods will be critical for ncRNA discovery. However, homology search methods were designed for protein coding sequences and their performance on ncRNAs is unknown. In light of this problem, Freyhult et al. have assessed how different sequence similarity search algorithms perform in the detection of ncRNAs. They evaluated 12 different search programs that fall into three broad classes: sequence-based methods, profile-based methods and structure-based methods. BLAST and FASTA are the most widely used of the sequence-based methods, while HMMer and Infernal exemplify the profile and structure-based methods respectively. The authors present a detailed accounting, which is beyond the scope of this abstract, of how all 12 of these methods perform with an emphasis on sensitivity, selectivity and relatively less focus on search speed. Perhaps most usefully however, they also provide a series of ‘practical recommendations’ to guide homology searches for ncRNAs. First of all, they point out that the most widely used methods often fare the worst in their comparisons. This has the unfortunate effect of rendering the results of many searches suspect. Sequence-based methods in particular need to use accurate scoring schemes, based on RNA-optimized substitution matrices, to ensure their reliability. The authors suggest that sequence-based searches be used as a starting point to build training sets, which can be employed to build covariance models for use with the program Infernal. Infernal uses the covariance matrix to do a structure-based search and was shown to perform very well in terms of both sensitivity and selectivity as did another structure-based method RSEARCH. However, both of these structure-based methods are substantially slower than straight profile-based methods such as HMMer. While the combined approach recommended by the authors may well represent the most rigorous way to search for ncRNAs, it also seems unlikely to be embraced by bench scientists who wish to do a quick and easy search on his or her favorite gene. Thus, it would seem that a good opportunity exists to implement a packaged software suite that could automatically implement this multipart strategy.

Briefings in Bioinformatics aims to provide working biologists with an awareness and understanding of the computational approaches available for research and discovery. The Abstracts section of the journal consists of summaries of bioinformatics manuscripts published in the previous quarter. Inclusion of an article in this section indicates that the editors consider it to be among the most interesting and/or useful contributions to the field for the quarter covered. The contents of these reports are briefly distilled for the readers with an emphasis placed on their biological context and potential utility. Publications from the fourth quarter of 2005 (October–December) are reviewed here. One of the most important challenges facing bioinformatics is the characterization of regulatory sequences that control the timing and pattern of gene expression. Gene expression is controlled, to large extent, by the interaction of trans-regulatory proteins with their cis-regulatory DNA binding sites. Cis-regulatory sites tend to be short, often degenerate, sequences, and attempts to identify such sites in gene promoters based on sequence information alone are prone to numerous false positives. As such, the necessity of incorporating additional information sources into the process of cis-regulatory sequence identification, exemplified by the manuscript of Tharakaraman et al., is becoming increasingly obvious. The method of Tharakaraman et al. is unique in two important respects. First of all, it combines two broad approaches to cis-regulatory element identification—namely alignment and enumerative-based methods. Alignment methods rely on (local) similarity between regulatory sequences, while enumerative methods identify sequence motifs that are over-represented in regulatory regions. Secondly, their method relies on positional information, with respect to the start site of transcription, to help inform the identification of functionally relevant regulatory motifs. The approach starts with an ungapped alignment of proximal promoter sequences, from the human genome in the case reported here, that is anchored by the transcription start sites. After the application of a novel word-specific mask, the method identifies all octonucleotide words that form colocalized clusters along the promoter sequence alignment. The promoter sequences are then realigned using the positions of the significant words as anchors, and the new alignment is surveyed for potential cis-regulatory sequences using a Gibbs sampling algorithm. When applied to the human promoter dataset, this method identified 791 words with conserved promoter positions. Alignments based on these word positions led to the discovery of novel potential cis-regulatory motifs missed by other prediction methods. A potentially interesting bonus of the algorithm is the fact that it does not necessitate the masking of known repetitive regions prior to the identification of putative cis-regulatory elements. Thus it may be possible to identify an entirely new class of cis-regulatory sites, i.e. those donated by repetitive sequences, that is likely to be missed by almost all other methods. The authors make the program that runs their alignment freely available along with the results of the analysis of human promoter regions. The high levels of similarity between the protein coding sequences of humans and chimpanzees has been appreciated for the last thirty years and was confirmed by comparisons with the recently completed sequence of the chimpanzee genome. The lack of divergence at protein coding sequences has raised questions pertaining to the hereditary basis for the obvious phenotypic differences between the two species. One possible explanation lies in structural genomic differences that may exist between the two lineages, and indeed, cytogenetic studies have successfully identified a number of large-scale structural differences that result in distinct human and chimp karyotypes. However, many smaller scale structural differences that may exist between human and chimp have been obscured by the relatively low quality of the chimpanzee genome assembly. Such structural differences may have had profound evolutionary impacts due to their dramatic and irreversible nature. To deal with the assembly issue, Newman et al. have employed their own mapping procedure to compare human and chimp sequences directly, and in so doing, identified numerous, previously undetected, structural differences between the two genomes. A total of 651 sites of structural variation including insertions, deletions and inversions were discovered. Structural variants were found on all chromosomes and, not surprisingly, tend to be enriched in the areas where segmental duplications had occurred. The variant structural sites not only cover a substantial portion of the analyzed sequences, 24 megabases, but also overlap more than 200 genes. Importantly, a number of these computationally identified structural variants were confirmed experimentally using PCR and Southern blots. This work represents the first genome-scale analysis of structural differences between human and chimpanzees. In addition to increasing the number of known human-chimpanzee structural rearrangements by a factor of 50, this study may also help to point out chimpanzee genomic regions that should be priorities for future efforts at producing ‘finished’ sequences. Gene duplication is a major evolutionary force and the subject of numerous molecular evolutionary studies. There are a number of competing hypotheses that seek to explain the fate of duplicate genes—most of these address the basic question of why duplicate genes are retained in the genome. Kuepfer et al. approach this issue using the extremely well studied metabolism of the yeast Saccharomyces cerevisiae as a model system. The essence of this work is the functional characterization of 295 duplicate metabolic genes in yeast. The authors used an impressive and in depth combined functional annotation scheme that employed information from experimentally determined phenotypes (under five different environmental conditions), in vivo flux data, in silico flux balance analysis and topological analysis of the yeast metabolic network. Interestingly, they found that duplicates are no more frequently associated with essential metabolic reactions than singletons. These results suggest that duplicate genes are not maintained by any one particular dominant function, i.e. duplicates do not appear to be primarily maintained as back ups. Instead, it seems as if duplicates encode proteins involved in different, if somewhat overlapping, functions. This work has important implications for theories of gene duplication and genetic robustness. Studies that map single nucleotide polymorphisms (SNPs) onto genomes often use a two-step process: first genomic locations that harbor SNPs are identified using small samples and second the SNP allele frequencies of these locations are characterized by resequencing larger samples. The primary use of SNPs is for the detection of phenotype-genotype associations through linkage disequilibrium, and this two-step strategy works quite well irrespective of the mode of the initial SNP identification. However, SNP data gathered in this way will most likely not be suitable for population genetic studies. That's because the original sampling used to identify SNPs is small and thus biased towards high frequency alleles; rare SNPs are more likely to go undetected with this approach. This will inevitably alter statistical characteristics of the resulting SNP frequency distribution, such as nucleotide diversity and FST (between population variance). To demonstrate the effects of this ascertainment bias, Clark et al. compare two large-scale SNP datasets, each of which used a different strategy to uncover SNPs. The International HapMap project used an extremely heterogeneous SNP discovery step, while Perlegen's SNP map was produced using a more uniform approach where SNPs identified by hybridization were resequenced. A SNP data set produced by the NIEHS was taken as a reference for comparison since it was produced by complete resequencing and as such is not expected to have any ascertainment bias. The SNP site frequency spectra as well as measures of heterozygosity FST and were determined for genome windows of 500-kb on each of these datasets. There are substantial differences between the SNP datasets for all three of these measures, and these disparities underscore the relevance of the ascertainment biases inherent in the different SNP discovery methods. Clearly then, population genetic analyses on such datasets will be affected by these biases. The authors take an important step beyond issuing this caveat by performing an ascertainment correction. The result is a corrected SNP frequency spectrum with frequencies weighted by the probability of discovering each SNP in the original study. The corrected data are far more consistent in terms of the various measures based on the SNP site frequencies. Nevertheless, even after this correction some disparities between datasets still exist. On a slightly more optimistic note, while the effects of ascertainment bias on population genetic studies are profound, genotype-phenotype association studies are likely to be far less affected (and not prone to false positives). One of the enduring questions of human biology is the extent to which the evolution of our species has been shaped by natural selection. The recent completion of the closely related chimpanzee genome sequence as well as concerted efforts aimed at uncovering human genomic polymorphism in the form of single nucleotide polymorphisms (SNPs) provide the necessary raw data to address this issue in a systematic way and on genome-wide scale. Bustamante et al. present a comparative sequence analysis of human–chimp protein coding sequences that also employed abundant SNP data, based on a population of 20 European Americans and 19 African Americans, generated by Celera Genomics. The work focused on more than 11 000 well-annotated human genes that have demonstrable orthologs in the chimpanzee. Levels of coding sequence SNP polymorphism and human-chimp divergence were compared in order to assess the effects of selection on these genes. To do this, coding sequence sites were partitioned into synonymous sites, where a change does not result in an amino acid difference, and non-synonymous sites, where nucleotide changes result in a corresponding change in the encoded amino acid. Under this approach, known as the McDonald–Kreitman test, synonymous sites are taken as a standard of neutral evolution and an excess of non-synonymous divergence indicates adaptive evolution between species. On the other hand, a paucity of non-synonymous divergence indicates either weak negative selection or balancing selection acting to maintain polymorphism at a locus. The vast majority of genes analyzed in this study were either uninformative with respect to the McDonald–Kreitman test or did not show evidence for selection when the test was applied. However, when genes with at least four variable non-synonymous sites are considered, 9.0% showed evidence of rapid amino acid evolution due to adaptive divergence. It should be noted that this figure represents only ∼2.5% of all genes in the study since the majority of genes show very low human–chimp divergence and thus have not undergone adaptive protein evolution. As has been noted previously, genes involved in immunity and defense, gametogenesis and sensory perception were over-represented in the set of adaptively evolving genes. Transcription factors were also enriched in this set suggesting one mechanism by which regulatory differences may evolve between species. In addition, 13.5% of informative loci show a paucity of amino acid divergence suggesting weak negative selection. These loci were found to be enriched in genes involved in actin binding and cytoskeleton formation. Most studies of molecular evolution have focused on the coding sequences of genes and the protein products that they encode. However, the vast majority of genomic DNA in eukaryotes, e.g. >98% in the human genome, is non-coding. The tempo and mode of non-coding sequence evolution has been studied far less and, as a consequence, is not nearly as well understood. As more and more complete genome sequences accumulate, comparative genomics studies are beginning to remedy this situation. The recent report of Peter Andolfatto exemplifies the evolutionary insights that can be gained by the analysis of non-coding genome sequences. Andolfatto studied genome sequences within and between Drosophila species, with an emphasis on D. melanogaster, and demonstrated the extent to which natural selection has influenced the evolution of non-coding sequences. Two main aspects of the study were: (i) partitioning genomic sites into different categories that are expected to differ with respect to the intensity of selection and (ii) comparing levels of within species (polymorphism) to between species (divergence) sequence variation. Intergenic and gene regions were studied separately, and gene regions were partitioned into synonymous and non-synonymous sites of the coding sequence as well as introns and untranslated regions (UTRs). This resulted in one of the most surprising findings of the study, namely the discovery that the majority of non-coding sequences evolve more slowly than do synonymous sites of coding sequences. This includes not only non-coding genic sequences such as introns and UTRs but also intergenic regions quite distant from known genes; relative levels of both polymorphism and divergence were reduced for non-coding DNA. Apparently, the conservation of non-coding sequences cannot be attributed to reduced mutation rates. The estimate that 40-70% of Drosophila non-coding sequences are conserved by natural selection represents more than an order of magnitude increase over the fraction of conserved sites recently estimated for other eukaryotes such as mammals. Andolfatto also employed a version of the McDonald–Kreitman test that allows for inferences of adaptive selection. This test compares the ratio of polymorphism to divergence for sites that are thought to be subject to the effects of selection versus sites that are thought to evolve neutrally. Application of this test revealed a significant excess of divergence for UTRs indicating that these regions have evolved adaptively. Taken together, these results suggest that a large fraction, most likely a majority, of non-coding Drosophila DNA is functionally important to the organism, presumably due to the exertion of regulatory effects, and subject to the effects of both purifying and adaptive selection. In 1970, Susumu Ohno proposed the so-called 2R hypothesis, which explains the increase in vertebrate genome size and complexity by two rounds of genome duplication that were to have occurred early in vertebrate evolution. It was difficult to test this provocative hypothesis directly without the availability of complete (or nearly so) vertebrate and outgroup genome sequences, and so it was not addressed analytically (at the genome scale) for decades. In just the past few years or so, owing to the availability of numerous genome sequences, a plethora of tests of the 2R hypothesis have been published. These works have produced highly contradictory results—some studies confirm the 2R hypothesis, others find evidence for only one round of genome duplication, while still others find no support whatever for whole genome duplication in vertebrates. It is against this backdrop of controversy that Dehal and Boore contribute their own test of the 2R hypothesis. They report what they consider to be unequivocal evidence of two rounds of whole genome duplication in early vertebrate evolution. This conclusion is based on a comprehensive phylogenetic analysis of all multi-gene families from complete genome sequences of tunicate, fish, mouse and human. To date, most tests of 2R have similarly analyzed multi-gene families to look for evidence based on gene number or tree topology. Such works often find negligible remaining signal of two rounds of genome duplication, and the analysis reported here confirms this paucity of supporting evidence for 2R when gene number and tree topology alone are considered. It is only when the genomic positions of duplicate (paralogous) gene families are mapped to the genome that the evidence confirming 2R emerges. Specifically, genes that were duplicated prior to the divergence of fish and tetrapods tend to cluster into what the authors refer to as tetra-paralogons—that is four distinct genome regions, each of which represents an ancient, pre-duplicated, genomic segment. In contrast, more recently duplicated genes, which presumably result from tandem duplications of individual genes or chromosome segments, do not group into distinct tetra-paralogons. The authors estimate that 25–72% of human genes map to tetra-paralogons and thus may have been generated by whole genome duplications. A concerted and sustained effort is underway to characterize the regulatory genomic regions, cis-binding sequences in particular, that control gene expression. Bioinformatic studies are increasingly relying on the incorporation of evolutionary information to aid in the identification of functionally important cis-regulatory sequences. Phylogenetic footprinting in particular relies on the assumption that functionally important cis-regulatory sites will be anomalously conserved over evolutionary time—i.e. show low levels of sequence diversity. However, the mode of evolution of regulatory sequences remains poorly understood, and consequently the extent to which this assumption is justified not known. Bilu and Barkai report a timely and relevant attempt to understand the factors involved in the evolution, and by inference the design, of cis-regulatory binding sequences. They employ the yeast Saccharomyces cerevisiae for their study, taking advantage of the detailed map of cis-regulatory binding sites that exists for this model eukaryote. The specific aim of the study is to understand the relationship between the length and sequence specificity of cis-regulatory binding sequences. Prior to this work, it was recognized that cis-regulatory binding sequences that are short and lack sequence specificity may nevertheless be responsible for controlling gene expression. On the other hand, many apparent binding sites, including ones that are longer a closer to the sequence consensus for the site, are actually non-functional. This paradox has been attributed to combinatorial regulation, whereby the regulation of eukaryotic genes is controlled by the interaction of numerous regulatory factors and cis-binding sites. For the first time here, the authors actually demonstrate a connection between this type of combinatorial regulation and binding site specificity. They show that binding sites found in promoters that bind multiple transcription factors tend to be both shorter and ‘fuzzier’—i.e. less similar to the consensus sequence. In addition, novel binding sites were found to appear more often in promoters that already bind multiple transcription factors. The authors posit that this may be due to reduced selective pressure on such promoter sequences, which is also consistent with the fuzzier sequence motifs found therein. The observation that essential genes tend to possess fewer binding sequences seems to be consistent with this evolutionary model. The type of evolutionary design principles elucidated by the work of Bilu and Barkai would seem to have direct relevance to the use of comparative genomic-based cis-regulatory site prediction methods.

Briefings in Bioinformatics aims to provide working biologists with an awareness and understanding of the computational approaches available for research and discovery. The Abstracts section of the journal consists of summaries of bioinformatics manuscripts published in the previous quarter. Inclusion of an article in this section indicates that the editors consider it to be among the most interesting and/or useful contributions to the field for the quarter covered. The contents of these reports are briefly distilled for the readers with an emphasis placed on their biological context and potential utility. Publications from the first quarter of 2006 (January–March) are reviewed here. The vast majority of human genomic sequence is not protein coding, and putative functional roles for almost all of this sequence have yet to be discovered. In light of this knowledge gap, there is a sustained effort underway to develop computational approaches to aid in functional predictions for non-coding DNA. In the last several years, the importance of using comparative (between genome) sequence analysis in such efforts has become increasingly obvious. This is because functionally important sequences tend to evolve more slowly than non-functional sequences and thus can be identified by virtue of their low levels of sequence divergence. Lunter et al. propose a corollary to this approach by focusing on insertion and deletion (indel) events as opposed to nucleotide substitutions. The first part of their approach involves the derivation of the neutral background expectation for levels of indel when no selection is acting on the sequences. The model they devise towards this end shows a good fit to the indel pattern for human–mouse ancestral repeats, which are not thought to evolve under selection. Having defined the neutral indel rate in this way, the authors were then able to identify a number of long ungapped regions (significantly non-neutral) that they presume to be greatly enriched for functionally active non-coding DNA. Fully 2.5–3.2% of the human genome – about 2 times as much as coding regions – appears to have evolved under a regime of purifying selection against indel events. While many of the sequences in these selected regions are not annotated, the method performs quite well at identifying know functional elements such as microRNAs. One of the substantial benefits of their method is that it seems to require much lower genome coverage than comparative methods that evaluate sequence substitutions. Comparison of only 2–3 mammalian genomes is sufficient to identify 90 million bases of human DNA subject to purifying indel selection with predicted levels of only 10% false positives and 25% false negatives. Finally, the authors point out that application of their method can lead to the detection of genomic regions that have been subject to heterogeneous selection, like purifying selection against indels with diversifying selection on sequence substitutions. This raises the possibility of extending genome-scale surveys of positive selection from protein coding regions to non-coding DNA, an important development in molecular evolution. Network based analyses have helped to provide an integrated systems-level perspective on the functional and evolutionary relationships among thousands of genes. One of the first examples of this unified approach was the discovery that highly connected proteins, so-called hubs of protein interaction networks, tend to be more evolutionarily conserved than less prolific interactors. However, this seemingly straightforward claim has been controversial and subject to great scrutiny in the past few years. Much of the ensuing debate has focused on problems with the reliability of high-throughput protein interaction data sets and the accuracy of ortholog assignments based on sequence comparison. Wuchty and colleagues confirm that the evolutionary conservation of highly connected proteins is robust to both of these potential sources of noise in large-scale genomic data sets. Perhaps more importantly, their work entails an important qualitative shift in focus away from a node-centric perspective to a more link centered one. In other words, the emphasis here is placed on the conservation of interactions between gene products as opposed to the conservation of the genes (proteins) themselves. In so doing, the authors are able to discover that local network substructure, in terms of clustering around interactions, is correlated with evolutionary conservation of the proteins involved in the interactions. In addition, the genes found in these locally clustered sub-networks show elevated levels of coexpression consistent with bona fide functional relationships among them. One of the most intriguing things to come out the work is an expansion of the notion of the fundamental evolutionary unit from single proteins to ensembles proteins and the interactions among them. This has theoretical implications for the understanding of the action of natural selection as well as potential practical utility with respect to the prediction of interaction networks (i.e. functional information transfer) for non-model organisms. Indeed, the authors use a comparison of local interaction network structure and orthology between Saccharomyces cerevisiae and Plasmodium falciparum to demonstrate the potential of such predictive methods. Gene duplication is an important source of genetic novelty and has been particularly prevalent in plants such as the model system Arabidopsis thaliana. Comparative studies of the Arabidopsis genome sequence have revealed that it has experienced several rounds of large-scale, whole genome duplication in addition to many smaller, local duplication events. It has previously been shown that Arabidopsis genes often acquire distinct expression patterns after duplication but the roles of gene function and duplication age in this process have not been well appreciated. Casneuf et al. study the effects of different classes of duplication on patterns of gene expression and consider the relationship between expression divergence and the encoded functions of the duplicated genes. All-against-all sequence comparisons were used to detect duplicates, and the relative ages of gene duplications were defined by comparing the synonymous substitution divergence between duplicated genes. The authors then analyzed two publicly available Arabidopsis microarray data sets to assess the extent of expression pattern divergence between duplicate genes. They found that genes that duplicated via local small scale events had acquired relatively divergent patterns of expression compared to those duplicated in large scale events. Locally duplicated gene pairs also tended to diverge asymmetrically in the sense that one duplicate was expressed in a large number of tissues while the other was expressed in fewer tissues. It is not clear whether this asymmetric divergence pattern is due to a distinct sub-functionalization of the duplicate gene with the more narrow expression pattern or whether that same gene is in the process of losing much of its function (becoming a pseudogene). The comparison of pairs of genes also limits the interpretation of the results in the sense that the authors could not determine if the asymmetric expression patterns were due to gain or loss of tissue-specific expression. Nevertheless, the difference between the two modes of expression divergence for the two classes of duplicates is quite striking. In terms of function, the expression patterns of genes involved in signal transduction and response to external stimuli diverge relatively quickly. This could be an evolutionary mechanism by which Arabidopsis has evolved to meet the challenges of a changing environment. Mammalian genomes are made up primarily of transposable elements (TEs) – mobile sequences capable of replicating themselves independent of their host genomes. The human and mouse genomes, for instance, consist of more than 50% TE-derived sequence, made up of millions of individual elements that are separated by only 500bp on average. Despite the ubiquity and abundance of mammalian TEs, there are regions of the genome that are entirely devoid of these elements. Simon et al. focus on these so-called ‘transposon-free regions’ (TFRs), and their analysis of these anomalous genomic regions has important implications for understanding mechanisms of gene regulation. They show that the human and mouse genomes both have close to 1,000 TFRs more than 10kb in length, and most of the TFRs are present as corresponding orthologous regions in the human and mouse genomes. The latter fact is particularly interesting since most mammalian TEs are lineage-specific, in other words they have accumulated since the human and mouse genomes last shared a common ancestor. Thus, the presence of orthologous TFRs means that TE sequences have been selectively excluded – i.e. prevented from inserting – independently in human and mouse orthologous regions. This suggests similar selective constraint for orthologous human–mouse TFR regions, which must of course be based on some functional role for those genomic sequences. More than 90% of the TFR sequences are non-coding so the function of those selected regions is probably related to gene regulation. Consistent with this interpretation, most TFRs are found in the vicinity of genes involved in developmental regulation. Paradoxically, however, the orthologous TFRs are not highly conserved in terms of sequence. This is most surprising because it is expected that orthologous, functionally analogous regions will be anomalously conserved in terms of the level of sequence divergence. Indeed, this expectation is the basis of a widely employed class of approaches, known as phylogenetic footprinting, which use sequence conservation levels to guide the identification of non-coding functional genomic elements. This work, on the other hand, suggests a novel view of regulatory sequence evolution whereby genomic regions are somehow selected for regulatory function without conservation of their particular genome sequence. The importance of gene duplication for the evolution of novel biochemical functions has long been recognized. The basic model for this process holds that duplication provides the redundancy needed to free single copy genes from selective constraint, and most studies to date have focused on how this redundancy facilitates changes in sequences (i.e. point substitutions) between duplicate gene copies. Here, Su and colleagues investigate the role of alternative splicing (AS) in this process by evaluating AS differences between duplicate human genes. Previous studies found that gene duplication reduces AS suggesting that these two phenomena may represent complementary modes for generating diversity. However, there is much that remains to be learned about the relationship between gene duplication and AS. The current study addresses a number of outstanding questions including: i-how long does it take for AS to evolve after duplication, ii-is AS of duplicates asymmetric and iii-does AS evolve independently of gene duplication. Consistent with previous results, the authors did find that duplicate genes have fewer AS variants than single copy genes and members of larger gene families tend to have even fewer AS variants. It was also demonstrated that the loss of AS variants can occur shortly after gene duplication. Furthermore, there is pronounced asymmetry in the patterns of AS between duplicate genes whereby duplicates have pronounced differences in their splicing patterns. All of this points to a tightly coupled and complementary relationship between gene duplication and AS. Apparently, in the period immediately following gene duplication the duplicates take on distinct aspects of the function that was previously carried out by the single copy ancestral gene via AS. This type of reductive evolution is known as subfunctionalization. It is only later on that the evolution of AS between duplicate gene copies occurs more-or-less independently. The prevalence and importance of alternative splicing (AS) of eukaryotic genes is by now quite well appreciated. In this paper, Liang and Landweber analyze AS genes that encode multiple protein sequences from overlapping reading frames. This phenomenon is far less recognized among eukaryotes, and was previously considered to be largely confined to small prokaryotic and viral genomes. For such small genomes, overlapping reading frames represent an evolutionary strategy by which genomes can increase their coding content without concomitant increases in size. Since there is far less apparent selective pressure to streamline eukaryotic genomes, overlapping reading frames were presumed to be less prevalent. By closely examining a set of human mRNA transcript sequences, the authors were able to show that just under 7% of alternatively spliced genes contain multiple overlapping coding frames. It is important to note that the authors took measures to ensure the reliability of the analyzed transcript set, focusing only on relatively well-annotated sequences. The figure of 7% is slightly mitigated in that it represents only ∼1% of all human genes analyzed, nevertheless this still seems to be a surprisingly high figure for such a large eukaryotic genome and suggests the emergence of multiple overlapping reading frames is an important evolutionary force in eukaryotes. Interestingly, many of these secondary reading frames appear to have emerged recently, being mammalian and even primate-specific. The derived reading frames have significantly different codon usage and amino acid composition than ancestral frames and the rest of the genes in the genome. At first glance, these latter findings may seem to suggest that the derived secondary reading frames encode non-functional proteins, and this is a critical point regarding the significance of the work. However, the authors performed a number of comparative sequence analyses that suggest that most of these transcripts encode proteins that are at least potentially functional. Specifically, the authors tested whether proteins encoded in secondary reading frames were similar to known functional sequence from other organisms, whether they contained functional motifs, whether they could form know secondary structural elements and whether they could be found to be similar to distantly related structures via threading. Almost all proteins encoded by overlapping reading frames conform to at least one of these criteria and most meet multiple lines of evidence for function. Eukaryotic genome sequences are saddled with vast regions that do not encode any known protein; for instance, the human genome is 98–99% non-coding. One of the fundamental challenges of the post-genomic era is to characterize the functional significance, or lack thereof, for this non-coding DNA in eukaryotic genomes. One prominent class of non-coding DNA is introns, and several recent studies have show that intron lengths vary substantially between broadly expressed house-keeping genes, which possess relatively shorter introns, and genes with longer introns that are expressed in more tissue-specific and/or developmentally-specific patterns. There are two distinct classes of evolutionary explanations, one neutral and one adaptationist, for the differences in intron lengths between these two groups of genes. The neutral explanation holds that selection acts on the more highly expressed house-keeping genes to keep them small and streamlined while the less expressed tissue-specific genes are more free to accumulate intronic DNA. At this moment, the neutral, or ‘selection for economy’, model corresponds to more-or-less to received wisdom. Vinogradov, on the other hand, finds support for the less invoked adaptationist perspective in his comparative analysis of mammalian intron and expression data. He refers to this as the genome design hypothesis, which suggests that tissue-specific genes possess more intronic DNA due to its functional role in their complex regulation. What he did was employ some of the most reliable technology in local sequence alignment to compare the amount of intron length conserved between orthologous human and mouse genes and considered these data with respect to the relative breadth of gene expression. Use of these rigorous sequence alignment algorithms allowed for the discovery of the highest fractions of observed human–mouse conserved intronic DNA, excluding lineage-specific repeats, to date. More to the point, Vinogradov shows that the fraction of conserved intronic sequences is higher for tissue-specific genes than for house-keeping genes, and this observation holds up after a number of controls are implemented, e.g. for mutation rate and GC content. Intron length was also found to be correlated with the number of functional domains in a gene. Another interesting finding was that the length distributions for conserved intron sequences peak close to the nucleosomal (170–220nt) and dinucleosomal (320–510nt) intervals. Taken together, these results suggest a possible connection between intron sequences and gene regulation mediated by chromatin condensation. Ever since Darwin put forth the notion that all of life has descended with modification from a common ancestral origin, the reconstruction of a ‘tree-of-life’, which defines the evolutionary relationships among species, has been a fundamental goal for the biological sciences. The molecular revolution showed that all species, no matter how phenotypically distinct, were unified at the genomic level and made the realization of this goal a formal possibility. In recent years, as numerous whole genome sequences have become available, a number of investigators have attempted, with notable success, to reconstruct a comprehensive tree-of-life. Despite these recent advances however, there remain fundamental technical and conceptual challenges to this endeavor. For instance, the precise phylogenetic placement of numerous taxonomic groups remains hotly debated, and even the basic notion that evolution can be by a structure has been et al. this by an for a that of genome sequences is and They with close to genome sequences, and the first is to identify conserved gene families present in all genomes and orthology among members of these The set consists of genes all of which are involved in proteins for the most It be that the identification and of this conserved gene set a substantial amount of however, set up the in this way, work can be in a largely important part of the analysis and computational involves the identification and of genes that have evolved via – such genes will lead to in any to a of analysis to the and of families that were present in all but showed evidence of multiple of these genes were known to have evolved by expected was the finding that a number of the conserved genes showed evidence of for these the was a alignment of more than and this alignment was used to reconstruct a of The utility of their approach is by the fact that the on the of life have good This robust allowed the authors to propose to several outstanding taxonomic For instance, they find evidence for a of and for a last common ancestor. a small fraction of the human genome protein sequences, but there is evidence that if not the of the genome is discovery of functionally regions is an important goal of functional of genomic sequences has pronounced effects such as the of differences between the two of genomic DNA. Specifically, there are by coupled DNA as well as selection against The authors of this propose to take of these so-called to identify previously genes that do not encode protein sequences. This approach represents a and distinct method of gene prediction – the most widely used methods on detection of gene structural elements and sequence their method is complementary to gene prediction algorithms and it has the potential to many previously genes in mammalian genomes. The approach here involves the of of which some of and/or by the that these algorithms up are confined to sequences that have sustained evolutionary as opposed to noise by and of this integrated in the detection of thousands of genomic that do not to known genes. on the their method indicates that as much as of the human genome is While their method performs well when tested on known gene sequences, it appears to be towards the detection of long regions. This may their approach to in the detection of genes with anomalously long introns that are to discovery with the most used methods based on gene structure and sequence the emphasis in understanding the increases in that the emergence of the human evolutionary has been placed on the of novel genetic and biochemical function. However, a recent regarding evolutionary – to as – this by gene loss as an important force in evolutionary In support of this several studies have to gene loss may have to related to however, there has been no to characterize gene loss the human with the possible evolutionary implications of this et al. have just such a analysis by to characterize the of This was by comparative genome sequence analysis with an in study focused on than which are more but less functionally This in the discovery of human that have emerged since human and last shared a common ancestor. genes encoded proteins with a number of different but genes involved in and response are significantly in this how gene loss can lead to the authors focus in on one on a genetic the authors present evidence that of this was by positive selection related to from They were also able to that the process of just to human out of Consistent with this two other genes related to were non-functional the human This study the potential importance of gene loss in the evolution of

Briefings in Bioinformatics aims to provide working biologists with an awareness and understanding of the computational approaches available for research and discovery. The Abstracts section of the journal consists of summaries of bioinformatics manuscripts published in the previous quarter. Inclusion of an article in this section indicates that the editors consider it to be among the most interesting and/or useful contributions to the field for the quarter covered. The contents of these reports are briefly distilled for the readers with an emphasis placed on their biological context and potential utility. Publications from the second quarter of 2006 (April–June), with an emphasis on the evolutionary divergence of the human and chimpanzee genomes, are reviewed here. Gavin C. Conant and Kenneth H. Wolfe PLoS Biology (2006) Vol. 4, no. 4, p. e109 Gene duplication is an important evolutionary force that often leads to the emergence of novel function. Many studies of this phenomenon have been conducted, and for the most part they have focused on the evolutionary rates and functions of individual paralogous gene pairs. Conant and Wolfe have added a new dimension to the evolutionary study of gene duplication by considering the effects of this process from a systems level perspective. They achieved this feat through the analysis of networks of interacting paralogous genes of the yeast Saccharomyces cerevisiae. A few years back, Wolfe and co-workers demonstrated that a whole genome duplication occurred along the evolutionary lineage leading to S. cerevisiae. Many of the paralogous gene copies generated by this duplication event have since been lost from the streamlined yeast genome, but many others, 551 duplicated pairs to be exact, remain and presumably contribute to functional diversification in the yeast genome. In order to better understand functional partitioning among yeast gene duplicates, the authors started with networks where the nodes were members of duplicated gene pairs and the edges consisted of coexpression relationships between genes. Given n pairs of paralogous genes, the authors devised and implemented a heuristic algorithm that partitions the two sets of n duplicated genes such that the number of interactions between the two sets (so-called crossing edges) is minimized. This approach has the effect of delineating maximally distinct subnetworks of paralogous gene sets. 19 paralogous sets were identified in this way, and their paralgous subnets were evaluated for asymmetry of interactions between subnets as well as functional coherence within subnets. Substantial asymmetry was found for paralogous subnetworks whereby one subnetwork contained far more interactions than its paralogous partner. However, there was also substantial redundancy between subnetworks, as indicated by the same within subnet interactions between paralogous genes. Taken together, these two observations suggest substantial, but incomplete, functional divergence between paralogous genes since the whole genome duplication event. In addition, individual subnets showed evidence of functional coherence in terms of tight clustering of interactions as well as similar subcellular localization and upstream regulatory sequence motifs. This work demonstrates the potential and utility of a systems level approach to evolutionary genomics. Jamie T. Bridgham, Sean M. Carroll and Joseph W. Thornton Science (2006) Vol. 312, no. 5770, pp. 97–101 Tightly integrated functional complexes, composed of multiple interacting parts, pose a conceptual challenge to Darwinian theory. This is because it is difficult to imagine how any single molecule's function can be selected unless its interacting partners are already present. Thus, the gradual stepwise emergence of such complexes via Darwinian selection would seem to be precluded. Darwin himself recognized this and even proposed a solution, namely the evaluation of ancient forms of the system, in order to better understand the transitions that have led to the emergence of the extant complex. Bridgham et al. analyze a specific kind of molecular complex, steroid hormones and their receptors, and their study exemplifies and illuminates how the evolutionary processes leading to an integrated functional complex actually occur. Using phylogenetic comparison of extant steroid hormone receptor sequences, they reconstructed an ancestral corticoid receptor (ACR). The ACR gave rise to two extant receptors through a gene duplication event – glucocorticoid (GC) and mineralocorticoid (MC) – each of which has a specific binding partner; GC binds cortisol and MC binds aldosterone. Paradoxically, aldosterone evolved much later than the MC receptor to which it binds, so it is unclear how MC could have evolved to bind the non-existent (i.e. as yet unevolved) aldosterone via Darwinian selection. The authors were able to demonstrate that the ACR already possessed a binding affinity for aldosterone, presumably as a by-product of its affinity to other substrates, thus resolving this conundrum. They went on to show the specific mutational steps that led to the retention of aldosterone binding by MC and its loss by GC. The authors refer to this process as ‘molecular exploitation’—meaning the capture of an ancient molecular function, encoded by a protein that evolved to perform a different function, and its recruitment into a novel functional complex. This study has important implications for the politically charged intelligent design debate. Although the authors do not explicitly raise this issue in their manuscript, Chris Adami frames their work in these terms in his commentary that can be found in the same issue of Science. Adami points out how a seemingly irreducibly complex system, i.e. one not capable of evolving in a stepwise fashion, actually evolved through a series of mutational intermediates. Leonardo Arbiza, Joaquín Dopazo and Hernán Dopazo PLoS Computational Biology (2006) Vol. 2, no. 4, p. e38 The recent completion of the chimpanzee genome sequence has allowed for genomic-scale comparative studies of human and chimp gene sequences. It is hoped that ultimately these types of studies will lead to an understanding of how natural selection has facilitated the emergence of biological features that are uniquely human. Arbiza et al. have conducted one such large-scale comparison of human and chimp gene sequences to search for the effects of natural selection. The idea is to identify gene sequences that have evolved under the influence of the kind of adaptive selection, also known as positive or diversifying selection, described by Darwin. The study of Arbiza et al. follows closely on the heels of several other attempts to identify positively selected human genes, but it is distinguished from these earlier works in its differentiation between cases of positive selection versus relaxation of selective constraint. Although positive selection and relaxation of selective constraint can both lead to increases in evolutionary rates and deviations from a strict molecular clock, they have very different biological implications. Relaxation of selective constraint suggests some degree of loss of function, while positive selection implies an active process of diversification leading to a new function. Starting with 13,198 human–chimp gene pairs, the authors identified statistically significant deviations from the molecular clock in 651 chimp and 469 human genes. Their test for positive selection revealed 577 cases in chimp and only 108 cases along the human lineage. The authors take these relatively small fractions of affected genes to suggest that the processes of accelerated evolution and positive selection have not been frequent events in the evolution of human and chimp genomes. They also suggest that this difference in the numbers of positively selected genes between the two species may be due to smaller population sizes for chimpanzees. Importantly, many accelerate genes did not show evidence of positive selection and numerous positively selected genes did not show evidence of evolutionary rate acceleration. This finding underscores the ability of their approach to distinguish these different modes of evolution. Most of the accelerated and positively selected genes identified by the authors fall into the same functional categories suggesting that it is the specific individual functions of positively selected genes that are important in generating differences between species. Weilong Hao and G. Brian Golding Genome Research (2006) Vol. 16, no. 5, pp. 636–643 With the complete genome sequences of literally hundreds of bacterial genomes now available, the prevalence of lateral gene transfer (LGT)—i.e. non-vertical transmission of genes across species boundaries—is becoming increasing apparent. Indeed, LGT is so widespread that it has been taken to threaten the very existence of any coherent bacterial species phylogeny. Hao and Golding have carefully examined both the pattern of LGT and the fate of laterally transferred genes by studying 13 closely related Bacillaceae genomes. An important aspect of this work is the methodology that the authors employ to measure whole gene insertion/deletion (indel) events among genomes. They combine (i) analysis of closely related species and (ii) whole genome comparison with (iii) a sensitive maximum likelihood method. Together these approaches allow for robust estimation of the extent and frequency of indels attributed to LGT events. Closely related genome sequences facilitate the reliable estimation of the number of transferred genes. Comparison of whole genomes is critical because it rules out the mis-inference of LGT based on either hidden paralogs or genome re-arrangements that obscure homologs. The maximum-likelihood method allows for a robust and sensitive phylogeny based approach to estimating LGT, with indel parameters free to vary across the tree. Given what was already known about bacterial genome evolution, it was not so surprising that they found abundant evidence for LGT. However, the extent and speed by which this process was found to occur is impressive. The rate of LGT among the Bacilliacea is even greater than the rate of nucleotide substitution. Furthermore, considering how relatively large-scale, i.e. the transfer of entire genes or even operons, LGT events are, their frequency really emphasizes the profound evolutionary impact that they must have had. This study shows how the power of the comparative genomics approach is enhanced by the ever-increasing genome sequence space and the continued refinement of the analytical methods brought to bear on these data. Nalini Polavarapu, Nathan J. Bowen and John F. McDonald Genome Biology (2006) Vol. 7, no. 6, p. R51 Transposable elements (TEs) are DNA sequences capable of moving around the genome, and mammalian genomes are made up primarily of the remnants of TE insertions. Retrotransposons—TEs that transpose via the reverse transcription of an RNA intermediate—are particularly abundant in the human genome, making up >40% of the genome sequence. Most studies of human–chimp genome divergence have focused on the evolution of typical protein encoding genes. However, the most divergent parts of these genomes correspond to TEs, retrotransposons in particular. In other words, if one wants to fully comprehend the genomic differences between humans and their closest living relatives, the differences among their TEs must be laid out in full. Nalini Polavarapu and colleagues have done just this for one particular class of retrotransposons—the endogenous retroviruses (ERVs). ERVs are genomic elements that are very similar to the more familiar infectious retroviruses, such as HIV, but they are unable to move from cell-to-cell. These elements are thought to be remnants of ancient germline infections, and they are capable of transposing within the genome. These transposition events lead to the accumulation of ERV sequences in the genome and, along with the elimination of element sequences, can lead to pronounced differences between evolutionary lineages. The authors report 42 families of ERVs in the chimp genome including the discovery of 9 previously unknown families. The vast majority of these families were found to have orthologs, i.e. elements in corresponding genomic positions, in the human genome. The presence of orthologous families indicates that these elements were around prior to the diversification of the two species. Nevertheless, nine families of chimp ERVs have been transpositionally active since the human–chimp divergence, while only one family has been active along the human lineage. It has previously been reported that the biggest differences between the human and chimp genomes result from insertion and deletion (indel) events rather than point substitutions. The authors of this study were able to demonstrate that 7% of the indel variation between the human and chimp genomes can be localized to ERV sequences. This underscores the role that ERVs have played in driving changes between the two species, and suggests the possibility that they may possess some as yet undiscovered particular evolutionary significance. Nick Patterson, Daniel J. Richter, Sante Gnerre, Eric S. Lander and David Reich Nature (2006) Vol. 441, no. 7097, pp. 1103–1108 Patterson and colleagues have published a startling paper with results that have the potential to substantially alter prevailing notions about human evolution. The issue at hand is the timing of the divergence event that led to the two separate human and chimpanzee evolutionary lineages. For years, the only way to date this speciation event was through the comparison of fossil data. Paleontologists typically placed the human–chimp divergence event at ∼10–20 million years ago. Starting in the late sixties, comparison of genetic data led to substantially more recent estimates for the time of the human–chimp divergence. The current consensus among molecular evolutionists is that the human and chimp lineages diverged ∼5–6 million years ago. Patterson et al. compared genome sequences from five primate genomes—human, chimp, gorilla, orangutan and macaque—and concurred with this date, estimating that the human–chimp divergence occurred just over 6 million years ago. This estimate is more recent than fossils suggest, and so the issue remains contentious, for paleontologists in particular. However, it should be noted that the estimate error levels associated with genetic dating could bring the two camps substantially closer together. The really provocative finding of this report is based on the demonstration that different regions of the genome show different levels of average divergence between human and chimpanzee. Regional human–chimp sequence divergence ranges from 87 to 147% of the average genome levels, which corresponds to a difference in age of 4 million years. The X-chromosome, in particular, shows exceedingly low levels of human–chimp sequence divergence along its entire length. From this difference, it can be estimated that the X-chromosome became species-specific 1.2 million years after the rest of the genome. This could have occurred if the human and chimpanzee genomic lineages continued to exchange genes before separating completely. In other words, our early hominid ancestors appear to have been interbreeding with chimpanzees and the resulting hybrids were, to some extent, fertile. X-chromosome compatibility would have been important in order to maintain hybrid fertility, and this could have been the source of strong purifying selection that kept the X-sequences more similar between lineages. This issue is, not surprisingly, highly controversial and more data will be needed to resolve it. In any case, this work highlights the power of genomic comparisons to reveal unexpected aspects of our evolutionary history and tell us things that fossil gazing cannot.

Briefings in Bioinformatics aims to provide working biologists with an awareness and understanding of the computational approaches available for research and discovery. The Abstracts section of the journal consists of summaries of bioinformatics manuscripts published in the previous quarter. Inclusion of an article in this section indicates that the editors consider it to be among the most interesting and/or useful contributions to the field for the quarter covered. The contents of these reports are briefly distilled for the readers with an emphasis placed on their biological context and potential utility. Publications from the third quarter of 2006 (July–September), with an emphasis on computational resources for the detection and analysis of orthologous groups of genes (proteins), are reviewed here. Todd F. DeLuca, I-Hsien Wu, Jian Pu, Thomas Monaghan, Leonid Peshkin, Saurav Singh and Dennis P. Wall Bioinformatics (2006) Vol. 22, no. 16, pp. 2044–2046 Orthologs are genes (proteins) that share a common ancestor and diverged via a speciation event. In this sense, orthologs can be thought of as directly corresponding genes from different genomes, and orthologous relationships are the fundamental unit of comparative genomics. Orthologous relationships are most often used to guide functional inferences, since orthologs are very likely to encode proteins with the same function. The importance of defining orthologs is underscored by the increasing number of databases dedicated to defining orthologous sets of genes. Orthologs are operationally defined as sequences with the highest similarity between two genomes, and as such, the accurate identification of orthologs relies on comparisons between complete (or nearly so) genome sequences. Accurate ortholog identification can be a computationally daunting task since it rests on the sequence comparison of many thousands of gene pairs, and one of the biggest challenges facing the field is the increasing number of complete genome sequences. The developers of Roundup, a database of orthologous gene pairs, have met this challenge head on by computing all possible pairwise sequence relationships between 250 complete genome sequences, including 32 eukaryotes. This represents a more than two-fold increase over any other ortholog database. Roundup uses a previously reported method for ortholog detection—reciprocal smallest distance—which is a refinement of the classically employed reciprocal best BLAST hit method. The reciprocal smallest distance method is distinguished by the use of global, rather than local, sequence alignments as well as the calculation of evolutionary distances instead of the reliance on BLAST scores. The combination of these two approaches yields improvements in both the sensitivity and selectivity of ortholog detection. The evolutionary distance metric used is a maximum likelihood estimate of the number of amino acid substitutions between orthologs, and these values are reported along with the pairs of orthologs. There are three distinct ways to access the ortholog data in Roundup. Users may retrieve phylogenetic profiles, which are binary vectors indicating the presence and absence of orthologs across a set of genomes. Users can also use individual genes to browse orthologous relationships among specified sets of genomes. This option is less inclusive then the phylogenetic profile approach, but is faster and more direct. To their credit, the database developers also make the raw data available for download. This data consists of pairs of orthologous genes for any two genomes along with their evolutionary distances. One of the nice features of Roundup is that it is conceived and implemented as a ‘community evolvable resource’—meaning that users may actually request that particular genomes be added to the database. Tim Hulsen, Jacob de Vlieg and Peter M.A. Groenen BMC Bioinformatics (2006) Vol. 7, p. 398 Functional inferences on genomic sequence data are increasingly relying on evolutionary information. An important source of evolutionary information is phylogenetic patterns, which reflect the presence or absence of genes across a set of species. PhyloPat is a utility that allows users to query the Ensembl gene database using phylogenetic patterns. Phylogenetic pattern queries allow for the identification of evolutionarily and/or functionally related groups of genes. In order to define phylogenetic patterns, it is necessary to first establish orthologous relationships between gene (protein) sequences. PhyloPat uses orthologous relationships defined by the Ensmart database. A number of other orthology databases exist and several also allow for queries using phylogenetic patterns. PhyloPat is distinguished from these earlier tools in that it uses a strictly gene-centric approach as opposed to one that relies on relationships among proteins. This is important because orthology is most accurately defined at the level of the genome rather than the proteome. In particular, individual genomic loci can encode multiple proteins via alternative transcripts, and the presence of multiple protein variants from single genes can confound attempts to accurately define orthologous relationships as well as lineage-specific gene family expansions and deletions. There are two basic ways to search PhyloPat: (i) users can start with a given phylogenetic pattern, using check boxes or regular expressions, and extract the set of genes that match the pattern or (ii) users can provide a list of genes to evaluate the phylogenetic pattern of each. This flexibility, combined with the easy to use interface, allows users to readily construct complex phylogenetic queries. PhyloPat output can be stored in HTML, MS Excel or plain text format. In addition to the phylogenetic pattern information, functional information is provided via the integration of the FatiGO web interface that reports Gene Ontology annotations. The integration with Ensemble and Ensmart helps to ensure the reliability of the orthologous relationships while providing additional utility to the user. Nandini Krishnamurthy, Duncan P. Brown, Dan Kirshner and Kimmen Sjölander Genome Biology (2006) Vol. 7, no. 9, p. R83 The fundamental method for functionally annotating newly sequenced genes is information transfer. Information transfer is a computational technique whereby experimentally characterized information is transferred to the query sequence of interest by virtue of sequence similarity to a known gene. While this approach has been largely successful, it is not without substantial problems. For instance, the most similar sequence in a database search may have a different function from the query sequence, and many query sequences do not retrieve any statistically significant hit when searched against a database. In light of these challenges, the Berkeley Phylogenomics Group has developed the database PhyloFacts. The essence of PhyloFacts is data integration from multiple sources, and the use of consensus bioinformatics approaches that help to avoid systematic errors inherent to individual methods. PhyloFacts is an online encyclopedia with close to 10 000 entries, or ‘books’ as they are known, each of which reports the results of pre-calculated structural, functional and evolutionary analyses for a protein family or domain. To assemble such an integrated resource PhyloFacts combines a number of publicly available bioinformatic tools with their own phylogenomic methods of analysis. Standard database homology search tools such as BLAST and FASTA are employed along with more recently developed iterative search methods like PSI-BLAST. PhyloFacts also uses domain-based annotation and structure prediction as implemented in databases like PFAM and SMART as well as prediction of protein localization. The in-house phylogenomic methods used allow for more precise classification based on the discrimination between orthologous and paralogous evolutionary relationships. PhyloFacts books include a number of distinct data types that characterize a protein family such as homologous protein clusters, multiple sequence alignments, phylogenetic trees, predicted structures, subfamily definitions, taxonomic distributions, domain characterizations and Gene Ontology functional annotations as well as links to the literature and other online databases. Users can browse this repository of data by viewing the individual web pages for each PhyloFact book, or they can query their sequence of interest against the database in order to classify it. PhyloFact books are assembled according to several different classes of evidentiary standards which correspond roughly to different levels of confidence. More than half of the book entries correspond to experimentally characterized structural domains; presumably, these represent the gold standard of the resource. The remaining entries are successively less confident, being derived from groups of proteins that can be globally aligned, conserved regions, motifs and ‘pending’ groups that remain to be manually validated. Hidenori Nishihara, Arian F.A. Smit and Norihiro Okada Genome Research (2006) Vol. 16, no. 7, pp. 864–874 The single most abundant class of eukaryotic genomic DNA is made up of the inserted remnants of formerly mobile transposable elements (TEs). For years, it was assumed that these sequences were merely ‘junk’ or parasitic DNA that served no function other than their own propagation. However, numerous molecular biology studies have uncovered different ways in which TEs have become domesticated to play some functional role for the host genome in which they reside. Most of these studies have been anecdotal in nature and their functional revelations about TEs were accidental. More recently, investigators have begun to explore the functional contributions of TEs to their host genomes in a more direct and systematic way. This report from the Okada group demonstrating the domestication of SINEs (short interspersed nuclear elements) in mammalian genomes represents one such study. In this case, the investigators were aware that (i) large numbers of mammalian non-coding sequences evolve under selective constraint that points to their functional relevance and (ii) non-coding DNA is vastly enriched with TE sequences. This led them to search for cases of conserved sequences derived from TEs. They were able to demonstrate that numerous ancient SINEs, from several families that they discovered, were evolving under strong selective constraint. The authors started with their discovery of a number of ancient SINE families from mammalian and chicken genomes as well as coelacanth, shark, hagfish and amphioxus. All of the novel SINEs they reported were discovered by virtue of a shared central domain that is highly conserved within and between element families. Together these elements comprise an ancient superfamily of TEs named DeuSINE for their distribution among deuterostome genomes. Perhaps the most interesting family discovered is named Amniota SINE1 (AmnSINE1) to reflect its distribution among mammalian and bird genomes. AmnSINE1 is more than 300 million years old and has a chimaeric structure consisting of a 5′ promoter region derived from 5S rRNA followed by a tRNA derived sequence and the conserved core domain characteristic of all the DeuSINEs. The 3′ tail is also shared by other DeuSINEs and shows similarity to the 3′ end of a zebrafish LINE (long interspersed nuclear element). There are ∼1000 copies of AmnSINE1 in the human genome, and remarkably, more than 10% of these copies correspond to loci that are highly conserved across mammalian genomes. Furthermore, the conservation is highest in the core domain of the element shared among all DeuSINEs. Thus, AmnSINE1 represents and ancient TE with a substantial fraction of copies that have evolved some function that benefits their host genome. While the authors speculate as to what this particular function may be, they were not able to point to any specific role based on computational analysis alone. Elfar Torarinsson, Milena Sawera, Jakob H. Havgaard, Merete Fredholm and Jan Gorodkin Genome Research (2006) Vol. 16, no. 7, pp. 885–889 This is an exciting and provocative article that points to an entirely new area of post-genomic sequence analysis. Comparative analyses of eukaryotic genomes have revealed that numerous non-coding sequences have been selectively conserved for tens of millions of years. Since this conservation must be based on function, efforts to functionally characterize non-coding portion of mammalian genomes are increasingly focused on these slowly evolving regions. Torarinsson et al. departed from this standard line of reasoning and focused their efforts instead on the non-repetitive part of the human genome that is cannot be aligned with mouse genomic sequence. They point out that while these genomic regions may not be conserved at the level of sequence, they may be structurally conserved. To evaluate this possibility, the authors carefully selected corresponding regions of the human and mouse genome that were syntenic and contiguous but nevertheless could not be aligned. In other words, these pairs of human–mouse sequences represent orthologous regions that share a common ancestor but have evolved beyond sequence similarity recognition. They then ran the program FOLDALIGN to compare the structures of the corresponding human–mouse sequences. FOLDALIGN conducts a local structural alignment, simultaneously aligning and folding regions between the two sequences, which facilitates the identification of conserved local structures. This allowed them to identify ∼1800 structurally conserved regions with no sequence similarity. Simulation with randomly shuffled sequences was used to indicate that this conservation was statistically significant. When these structurally conserved regions were compared to a recently published transcriptional map of the human genome, they were found to be twice as likely to map to transcriptionally active regions of the genome. This strongly suggests that these regions are structurally conserved by virtue of some role that their transcripts play. However, the set analysed in this study does not correspond to any of the known non-coding RNAs stored in the Rfam database. In other words, these represent completely novel discoveries. Studies of this kind emphasize the functional relevance of vast uncharacterized regions of genomic DNA and reveal the nuances of natural selection on the genome. As an aside, the computational analysis needed to complete this study was impressive and daunting. Alignment of the RNA structures took 5 months to run on a Linux cluster consisting of 70 nodes with 2GB of RAM each. This suggests an outstanding need for the development of heuristic algorithms that can accomplish similar structural comparisons, albeit with some loss of accuracy, but at far greater speeds. Ryan E. Mills, Christopher T. Luttig, Christine E. Larkins, Adam Beauchamp, Circe Tsui, W. Stephen Pittard and Scott E. Devine Genome Research (2006) Vol. 16, no. 9, pp. 1182–1190 A sustained and vigorous effort has been made to survey variation among human genome sequences in the form of single nucleotide polymorphisms (SNPs). There are more than 10 million distinct SNPs and a large, well-funded consortium is dedicated to assembling these data into a haplotype map that can be used for genetic linkage association studies. Far less work has been done on other sources of genetic variation; insertion and deletion polymorphism in particular has not been nearly as well characterized. This is surprising because indels are known to be common among genomes and can be expected to have even more profound mutagenic effects than SNPs. The lab of Scott Devine has been working to remedy this inequality, and here they present their own map of human genome indel polymorphism. This study is particularly notable because the investigators simply took advantage of sequencing studies that had been previously conducted to identify SNPs. Apparently the investigators behind these original re-sequencing studies were so focused on SNPs that they neglected indels. The authors of this report were able to identify more than 400 000 unique indel polymorphisms by computationally reanalysing the previously reported sequence traces. When the gaps in the sequence traces were polarized using chimpanzee genome sequence as an outgroup, they were found to correspond to roughly equal proportions of insertions and deletions. The indels ranged in size from 1bp to almost 10 00 bp and could be classified into several groups. These groups include 1bp indels, indels that result from local base pair repeat expansions, transposable element derived indels and all other indels of random genomic sequence. Over the entire human genome there is, on average, one indel for every 7.2kb of sequence. However, indels are far from randomly distributed; there are hotspots that have an order of magnitude more indel variation than the chromosome average. A number of indels that map to gene regions including promoters and coding sequences were identified in this study. Importantly, the authors have deposited their human indel map into the NCBI's SNP database (dbSNP), and in so doing substantially enhanced the publicly available information regarding human genome variation. Irma Lozada-Chávez, Sarath Chandra Janga and Julio Collado-Vides Nucleic Acids Research (2006) Vol. 34, no. 12, pp. 3434–3445 The combination of accumulating complete genome sequences and decades of experimental effort is leading to a more systematic understanding of higher-order functional and evolutionary relationships among numerous genes and their encoded protein products. This is the basis of the renaissance of systems biology, which can now serve as a powerful explanatory paradigm at the molecular level where reductionism one held sway exclusively. The leitmotif of molecular systems biology is the network—a graphical representation of the interactions among numerous genes, proteins, metabolites or other biological players. The articulation of transcriptional regulatory networks, whereby transcription factors are directionally connected to their target genes, allows for the comprehensive understanding of an regulatory A that has with the of transcriptional is the of evolutionary conservation of regulatory interactions and between is this Julio Collado-Vides and their to in this far report on regulatory They were able to demonstrate that regulatory have over evolutionary and that this the of to to vastly different The authors defined regulatory networks, based on interactions between transcription factors and target genes, in the two and These two have been and so it was possible to on experimentally regulatory from each. They the of these regulatory at three different (i) the of individual proteins in the (ii) the of the interactions between proteins and the of which of groups of genes by a single transcription All three levels of revealed of and substantial of regulatory factors were found to evolve more than target genes and which the of numerous target genes, were found to be particularly conserved. This result is the of what be expected based on biological where of to be evolutionarily and it the of regulatory to in one or a evolutionary interactions were also found to be conserved and there was of between in the regulatory were also to be in all of which points to the of gene and the importance of this in W. and Nucleic Acids Research (2006) Vol. 34, no. pp. A of research efforts is to characterize the functional of the vast of non-coding DNA that among eukaryotic genomes. this end there are many efforts to map the DNA motifs that with transcription factors to the of genes. The article of et al. a analysis of one particular genetic regulatory sequence known as element is a transcription that is by the transcription a number of genes. The authors and their target genes in the human in mouse genome by a with which to query the genome sequences. The is a representation of the sequence among experimentally characterized sequences. The authors were able to both sensitivity and selectivity in the computational detection of by the with known sequences as well as a known set that does not They were able to identify ∼1000 in both the human and mouse genome and more than of these represent novel discoveries. In addition to the computational the functional relevance of multiple was using an experimental transcription the most interesting of the study was that many of the are related and out to be derived from transposable elements (TEs). the that the TE derived are similar in sequence, they from a number of different TE families including and interspersed nuclear elements as well as sequences. This suggests a specific by which multiple novel can be genes the genome and to in their

Briefings in Bioinformatics aims to provide working biologists with an awareness and understanding of the computational approaches available for research and discovery.The Abstracts section of the journal consists of summaries of bioinformatics manuscripts published in the previous quarter.Inclusion of an article in this section indicates that the editors consider it to be among the most interesting and/or useful contributions to the field for the quarter covered.The contents of these reports are briefly distilled for the readers with an emphasis placed on their biological context and potential utility.Publications in the areas of comparative genomics, networks and the chimpanzee genome from the third quarter of 2005 (July-September) are reviewed here.

Briefings in Bioinformatics aims to provide working biologists with an awareness and understanding of the computational approaches available for research and discovery.The Abstracts section of the journal consists of summaries of bioinformatics manuscripts published in the previous quarter.Inclusion of an article in this section indicates that the editors consider it to be among the most interesting and/or useful contributions to the field for the quarter covered.The contents of these reports are briefly distilled for the readers with an emphasis placed on their potential utility.Publications in the areas of genome evolution and biological networks from the fourth quarter of 2003 (October-December) are reviewed here.

Briefings in Bioinformatics aims to provide working biologists with an awareness and understanding of the computational approaches available for research and discovery.The Abstracts section of the journal consists of summaries of bioinformatics manuscripts published in the previous quarter.Inclusion of an article in this section indicates that the editors consider it to be among the most interesting and/or useful contributions to the field for the quarter covered.The contents of these reports are briefly distilled for the readers with an emphasis placed on their potential utility.Publications in the areas of gene regulation, molecular evolution and single nucleotide polymorphisms (SNPs), from the fourth quarter of 2002 (October-December), are reviewed here.The Abstract section of this issue also includes a special subsection that contains reviews of the bioinformatics analysis of the recently published mouse genome sequence.

Briefings in Bioinformatics aims to provide working biologists with an awareness and understanding of the computational approaches available for research and discovery.The Abstracts section of the journal consists of summaries of bioinformatics manuscripts published in the previous quarter.Inclusion of an article in this section indicates that the editors consider it to be among the most interesting and/or useful contributions to the field for the quarter covered.The contents of these reports are briefly distilled for the readers with an emphasis placed on their potential utility.Publications in the areas of transposable elements, molecular evolution and bioinformatics methods from the first quarter of 2003 (January-March) are reviewed here.TRANSPOSABLE ELEMENTS Origin of a substantial fraction of human regulatory sequences from transposable elements I. King Jordan, Igor B. Rogozin, Galina V. Glazko Glazko and

Briefings in Bioinformatics aims to provide working biologists with an awareness and understanding of the computational approaches available for research and discovery.The Abstracts section of the journal consists of summaries of bioinformatics manuscripts published in the previous quarter.Inclusion of an article in this section indicates that the editors consider it to be among the most interesting and/or useful contributions to the field for the quarter covered.The contents of these reports are briefly distilled for the readers, with an emphasis placed on their potential utility.Publications in the areas of molecular evolution and genome annotation from the third quarter of 2003 (July-September) are reviewed here.

Briefings in Bioinformatics aims to provide working biologists with an awareness and understanding of the computational approaches available for research and discovery.The Abstracts section of the journal consists of summaries of bioinformatics manuscripts published in the previous quarter.Inclusion of an article in this section indicates that the editors consider it to be among the most interesting and/or useful contributions to the field for the quarter covered.The contents of these reports are briefly distilled for the readers with an emphasis placed on their potential utility.Publications in the areas of molecular evolution, gene regulation and functional genomics, from the second quarter of 2002 (April-June), are reviewed in this issue.

MOTIVATION: Class I alpha-mannosidases comprise a homologous and functionally diverse family of glycoside hydrolases. Phylogenetic analysis based on an amino acid sequence alignment of the catalytic domain of class I alpha-mannosidases reveals four well-supported phylogenetic groups within this family. These groups include a number of paralogous members generated by gene duplications that occurred as far back as the initial divergence of the crown-group of eukaryotes. Three of the four phylogenetic groups consist of enzymes that have group-specific biochemical specificity and/or sites of activity. An attempt has been made to uncover the role that natural selection played in the sequence and structural divergence between the phylogenetically and functionally distinct Endoplasmic Reticulum (ER) and Golgi apparatus groups. RESULTS: Comparison of site-specific amino acid variability profiles for the ER and Golgi groups revealed statistically significant evidence for functional diversification at the sequence level and indicated a number of residues that are most likely to have played a role in the functional divergence between the two groups. The majority of these sites appear to contain residues that have been fixed within one organelle-specific group by positive selection. Somewhat surprisingly these selected residues map to the periphery of the alpha-mannosidase catalytic domain tertiary structure. Changes in these peripherally located residues would not seem to have a gross effect on protein function. Thus diversifying selection between the two groups may have acted in a gradual manner consistent with the Darwinian model of natural selection. CONTACT: [email protected].