Downloaded from genome.cshlp.org on September 16, 2015 - Published by Cold Spring Harbor Laboratory Press Research Comprehensive identification and analysis of human accelerated regulatory DNA Rachel M. Gittelman,1 Enna Hun,2 Ferhat Ay,1 Jennifer Madeoy,1 Len Pennacchio,3 William S. Noble,1 R. David Hawkins,1,2 and Joshua M. Akey1 1Department of Genome Sciences, University of Washington, Seattle, Washington 98195, USA; 2Division of Medical Genetics, University of Washington, Seattle, Washington 98195, USA; 3Lawrence Berkeley National Laboratory, Genomics Division, Berkeley, California 94701, USA It has long been hypothesized that changes in gene regulation have played an important role in human evolution, but reg- ulatory DNA has been much more difficult to study compared with protein-coding regions. Recent large-scale studies have created genome-scale catalogs of DNase I hypersensitive sites (DHSs), which demark potentially functional regulatory DNA. To better define regulatory DNA that has been subject to human-specific adaptive evolution, we performed comprehensive evolutionary and population genetics analyses on over 18 million DHSs discovered in 130 cell types. We identified 524 DHSs that are conserved in nonhuman primates but accelerated in the human lineage (haDHS), and estimate that 70% of substi- tutions in haDHSs are attributable to positive selection. Through extensive computational and experimental analyses, we demonstrate that haDHSs are often active in brain or neuronal cell types; play an important role in regulating the expression of developmentally important genes, including many transcription factors such as SOX6, POU3F2, and HOX genes; and iden- tify striking examples of adaptive regulatory evolution that may have contributed to human-specific phenotypes. More gen- erally, our results reveal new insights into conserved and adaptive regulatory DNA in humans and refine the set of genomic substrates that distinguish humans from their closest living primate relatives. [Supplemental material is available for this article.] A number of traits distinguish humans from our closest primate Nonetheless, interpreting patterns of interspecific divergence relatives, including bipedalism, increased cognition, and complex and intraspecific polymorphism in noncoding DNA has been language and social systems (for review, see O’Bleness et al. 2012). considerably more challenging compared with those of protein- To date, the genetic basis of human-specific phenotypes remains coding sequences. An elegant and powerful way to identify evolu- largely unknown, complicated by the difficulties in distinguishing tionary changes in noncoding DNA of potential significance, orig- between phenotypically significant and benign variation. Thus, inally described by Pollard et al. (2006b) and extensively used evolutionary changes in protein-coding sequences have received thereafter (Pollard et al. 2006a,b; Prabhakar et al. 2006; Kim considerable attention, as the phenotypic consequences of these and Pritchard 2007; Bush and Lahn 2008; McLean et al. 2010; mutations have historically been easier to interpret (Clark et al. Lindblad-Toh et al. 2011; Pertea et al. 2011), focuses on the discov- 2003; Stedman et al. 2004; Chimpanzee Sequencing and Analysis ery of sequences that are rapidly evolving or lost on the human lin- Consortium 2005; Nielsen et al. 2005; Arbiza et al. 2006; Dennis eage but that are otherwise phylogenetically conserved and thus et al. 2012; Sudmant et al. 2013). Although protein-coding likely functional. This approach has led to the discovery of several evolution has clearly played a role in human evolution, proteins regions with species-specific enhancer activity (Prabhakar et al. account for only ∼1.5% of the human genome, most of which ex- 2008; Capra et al. 2013; Kamm et al. 2013), as well as human-spe- hibit high sequence similarity between humans and chimpanzees cific deletion of regulatory DNA (McLean et al. 2011). (Chimpanzee Sequencing and Analysis Consortium 2005). However, phylogenetic conservation is an imperfect proxy for However, between ∼2.5% and 15% of the human genome function, particularly for noncoding regulatory sequences that can is estimated to be functionally constrained (Chinwalla et al. exhibit significantly high rates of turnover (Dermitzakis and Clark 2002; Lunter et al. 2006; Asthana et al. 2007; Meader et al. 2010; 2002; Wray et al. 2003; Villar et al. 2014). To more directly identify Ponting and Hardison 2011). Thus, the mutational target size of regulatory DNA, recent studies such as the ENCODE (The ENCODE noncoding DNA is considerably larger than protein-coding se- Project Consortium 2012) and Roadmap Epigenomics Projects quences, suggesting that regulatory DNA is also an important sub- (Bernstein et al. 2010) have created genome-scale maps of DNase strate of evolutionary change, as originally proposed four decades I hypersensitive sites (DHSs) in a large number of cell types. ago (Britten and Davidson 1969; King and Wilson 1975). In some DNase I preferentially cleaves regions of open and active DNA, cases, detailed studies of individual genes have revealed human- making it a powerful assay to identify regulatory elements, regard- specific regulatory evolution, such as in FOXP2, which is thought less of their specific function (Galas and Schmitz 1978; Dorschner to have influenced traits related to speech and language in humans et al. 2004). Although high-resolution maps of DHSs now exist, not (Enard et al. 2002). © 2015 Gittelman et al. This article is distributed exclusively by Cold Spring Harbor Laboratory Press for the first six months after the full-issue publication date (see http://genome.cshlp.org/site/misc/terms.xhtml). After six months, it Corresponding author: [email protected] is available under a Creative Commons License (Attribution-NonCommercial Article published online before print. Article, supplemental material, and publi- 4.0 International), as described at http://creativecommons.org/licenses/by- cation date are at http://www.genome.org/cgi/doi/10.1101/gr.192591.115. nc/4.0/. 25:1245–1255 Published by Cold Spring Harbor Laboratory Press; ISSN 1088-9051/15; www.genome.org Genome Research 1245 www.genome.org Downloaded from genome.cshlp.org on September 16, 2015 - Published by Cold Spring Harbor Laboratory Press Gittelman et al. all experimentally defined regulatory elements are expected to be identify 113,577 DHSs that exhibit significant evolutionary con- functionally or phenotypically significant (Eddy 2012; Doolittle straint across primates, which manifest as regions of low sequence 2013; Graur et al. 2013; Niu and Jiang 2013). divergence compared with carefully defined putatively neutral Thus, we hypothesized that the synergistic combination of flanking sequence (FDR = 0.01) (Fig. 1). Next, for DHSs that are comparative and functional genomics would facilitate the high- conserved in primates, we performed a second likelihood ratio resolution identification of conserved and human accelerated reg- test (Pollard et al. 2010) and identified 524 regulatory sequences ulatory sequences. Here we describe the genome-wide architecture that have experienced a significant acceleration of evolution in and characteristics of 113,577 DHSs that are conserved in primates the human lineage and therefore exhibit an excess of human-spe- and 524 DHSs that exhibit significantly accelerated rates of evolu- cific substitutions (FDR = 0.05) (Fig. 1; Supplemental Table 2). tion in the human lineage (haDHSs). We estimate that ∼70% of Importantly, to avoid biasing ourselves against identifying human substitutions within haDHSs are attributable to positive selection; acceleration, we excluded the human sequence in the first test for we experimentally validated a large number of elements; and we conservation. perform extensive bioinformatics analyses that integrate informa- tion across multiple functional genomics data sets to better under- Characteristics of primate conserved regulatory DNA stand the functional and biological characteristics of haDHSs. We first characterized the set of DHSs conserved across primates. Approximately 93% of conserved DHSs overlap a phastCons con- Results served element, but many also contain short segments of less con- served sequence, making them overall less conserved than those Framework for identifying conserved and human accelerated identified by phastCons (Fig. 2A). We hypothesize that these less regulatory DNA conserved sequences interspersed within DHSs may facilitate the To identify human accelerated regulatory DNA, we leveraged ex- rapid acquisition of novel transcription factor binding sites, as perimentally defined maps of DHSs from 130 cell types identified these regions are already actionable (i.e., accessible to proteins) in the ENCODE and Roadmap Epigenomics Projects (Supplemen- and poised to evolve new functions compared with nonconserved tal Table 1). After merging DHSs across cell types into 2,093,197 sequences outside of DHSs. distinct loci (median size = 290 bp, SD = 159 bp), we used a Patterns of conservation varied significantly across cell type − whole-genome alignment of six primates from the EPO pipeline category (Kruskal–Wallis test; P = 5.08 × 10 8; Methods) (Fig. 2B), (Paten et al. 2008) to obtain separate alignments for each DHS, us- ranging from 5.0% of DHSs in chronic lymphocyte leukemia cells ing strict filtering criteria for alignment quality. We performed two to 20.4% in fetal brain cells. DHSs active in fetal cell types showed likelihood