This Article Abstract Full Text (PDF) Supplemental Research Data Alert me when this article is cited Alert me if a correction is posted Citation Map Services Email this article to a friend Similar articles in this journal Similar articles in PubMed Alert me to new issues of the journal Download to citation manager Citing Articles Citing Articles via HighWire Citing Articles via Google Scholar Google Scholar Articles by Carlson, C. S. Articles by Nickerson, D. A. Search for Related Content PubMed PubMed Citation Articles by Carlson, C. S. Articles by Nickerson, D. A. Social Bookmarking                   What's this?

Letter

Genomic regions exhibiting positive selection identified from dense genotype data

¹ Department of Genome Sciences, University of Washington, Seattle, Washington 98195-7730, USA ² Center for Biomolecular Science and Engineering, University of California, Santa Cruz, California 95064-1099, USA

	ABSTRACT

Top ABSTRACT Results Discussion Methods REFERENCES Web site references

 
The allele frequency spectrum of polymorphisms in DNA sequences can be used to test for signatures of natural selection that depart from the expected frequency spectrum under the neutral theory. We observed a significant (P = 0.001) correlation between the Tajima's D test statistic in full resequencing data and Tajima's D in a dense, genome-wide data set of genotyped polymorphisms for a set of 179 genes. Based on this, we used a sliding window analysis of Tajima's D across the human genome to identify regions putatively subject to strong, recent, selective sweeps. This survey identified seven Contiguous Regions of Tajima's D Reduction (CRTRs) in an African-descent population (AD), 23 in a European-descent population (ED), and 29 in a Chinese-descent population (XD). Only four CRTRs overlapped between populations: three between ED and XD and one between AD and ED. Full resequencing of eight genes within six CRTRs demonstrated frequency spectra inconsistent with neutral expectations for at least one gene within each CRTR. Identification of the functional polymorphism (and/or haplotype) responsible for the selective sweeps within each CRTR may provide interesting insights into the strongest selective pressures experienced by the human genome over recent evolutionary history.

A number of statistical tests have been devised that compare an observed SFS against neutral theory predictions. One of the most frequently used tests is Tajima's D (Tajima 1989), a comparison of nucleotide diversity estimated from the number of polymorphic sites observed in a given set of chromosomes against nucleotide diversity estimated from the allele frequency of the polymorphic sites. Other tests of SFS against neutral expectations exist, including Fu and Li's D (Fu and Li 1993), which is based upon the number of singleton derived alleles observed, and Fay and Wu's H (Fay and Wu 2000), which is based upon the frequency spectrum of nonancestral alleles. By using these tests, a substantial number of genes have been identified where the observed SFS is inconsistent with neutrality, including ABO blood group (Seltsam et al. 2003), the major histocompatibility antigens (HLA) (Hughes and Yeager 1998), lactase (Bersaglieri et al. 2004), and TRPV6 (Akey et al. 2004; Stajich and Hahn 2005). Genes with an excess of high-frequency variation (observed as a positive Tajima's D, e.g., ABO, HLA) are consistent with balancing selection, while genes with an excess of low-frequency variation (significantly negative Tajima's D, e.g., lactase, TRPV6) are consistent with positive selective pressure, where an advantageous variant (haplotype) has recently replaced most of the variation in a region. Genes subjected to a recent selective sweep, where the advantageous allele has not yet become the major allele (e.g., Duffy [Hamblin and Di Rienzo 2000; Hamblin et al. 2002] and CCR5 [Libert et al. 1998]) do not have easily detectable departures using Tajima's D, but they may be detected by using Fay and Wu's test. Genes under recent balancing selection (e.g., G6PD [Verrelli et al. 2002], -globin [Straus and Taylor 1981]) are more difficult to detect and may not be detectable by any standard nucleotide diversity test. Other simple tests of neutrality exist such as the Ewens-Watterson haplotype diversity test (Ewens 1972), haplotype lineage diversity tests (Verrelli et al. 2002), and haplotype extent tests (Sabeti et al. 2002). In addition, tests for geographically restricted selective pressure measured by divergence between populations are also available (F_st) (Weir and Cockerham 1984). Several recent studies have evaluated the utility of combining multiple tests to identify genes showing signatures of natural selection and have introduced more sophisticated approaches to evaluate the robustness of these analyses to the neutral model. It is noteworthy that in these studies, all genes with significantly negative Tajima's D under the simplest neutral models were robust to a range of demographic parameters (Akey et al. 2004; Stajich and Hahn 2005), although questions about some of the tested models have been raised (Thornton 2005).

The identification of >10 million single nucleotide polymorphisms (SNPs) across the human genome, and the emergence of large-scale data sets of genotypes at a subset of these SNPs in multiple populations are leading to many insights into the patterns of sequence variation across the human genome (Sabeti et al. 2002; Clark et al. 2003a,b; Nielsen et al. 2005; Williamson et al. 2005). Here we describe the application of Tajima's D to a dense genotype data set (the Perlegen data set) (Hinds et al. 2005). We compared Tajima's D in completely resequenced human genes (SeattleSNPs, http://pga.gs.washington.edu) with the Perlegen data set and observed strong correlations in both African American and European American data for coding and non-coding regions. Based on these correlations, we used an empirically derived sliding window distribution of Tajima's D to examine the autosomal regions of the human genome. We identified large regions of the genome with an excess of rare variation in each of three populations, consistent with strong and recent selective sweeps that would lead to rapid increases in the frequency of an advantageous polymorphism and/or haplotype within each region (Smith and Haigh 1974). To confirm the validity of this approach, eight genes (defined as known genes or expressed sequence tags [ESTs]) from six such regions were selected for directed resequencing to obtain the full SFS and directly assess Tajima's D. Our results are consistent with a strong, recent, positive selective pressure in each of these resequenced regions, and they demonstrate the utility of dense genotype data in identifying such regions across the genome.

View larger version (23K): [in this window] [in a new window]   Figure 1. Comparison of Tajima's D between Perlegen and SeattleSNPs data sets. For each gene, Tajima's D was calculated from complete resequencing data in the SeattleSNPs data set, or from the region spanning 10 kb upstream of the transcript, the full transcript, and 10 kb downstream of the transcript in the Perlegen data. (A) Tajima's D from Perlegen vs. Tajima's D from SeattleSNPs for AD population. (B) Tajima's D from Perlegen vs. Tajima's D from SeattleSNPs for ED population. Genes previously resequenced by SeattleSNPs are shown in red, with a trend line representing a linear regression on the data. Genes resequenced as part of the present study are shown as purple dots, with filled circles indicating that the gene lay within a CRTR in the population being plotted. The seven SeattleSNPs genes with robust signatures of selection in SeattleSNPs data are shown in green (Akey et al. 2004).

	Results

Top ABSTRACT Results Discussion Methods REFERENCES Web site references

The correlation between these data sets suggested that the Perlegen data can be used to survey the genome for regions exhibiting extreme values of Tajima's D, so we applied a sliding window analysis to all three populations genotyped by Perlegen. The distribution of Tajima's D values for a 100-kbp sliding window is shown in Figure 2. The average windowed Tajima's D in the AD population was 1.20 (±0.73 SD) with a range of –2.41 to 4.03. In the ED population, the average Tajima's D was 1.40 (±1.01 SD) with a range of –2.80 to 4.34, while in the Chinese (XD) population the average was 1.45 (±1.13 SD) across a range of –3.11 to 4.42. These distributions were substantially skewed in all three populations, with a heavier tail to the distribution at low values.

View larger version (18K): [in this window] [in a new window]   Figure 2. A probability density plot of the distribution of Tajima's D in the sliding windows is shown for each population. All three distributions depart significantly from a normal distribution, most noticeably in the heavy tail at low values in each population.

View this table: [in this window] [in a new window]   Table 1. CRTRs

View larger version (57K): [in this window] [in a new window]   Figure 3. Tajima's D in 100-kbp sliding windows with 10-kbp steps is shown across the first 50 megabases of chromosome 1. Several CRTRs are visible, including a region near 35M in the ED population containing CLSPN (large blue arrowhead) and a region near 41M in the AD population spanning CTPS, FLJ23878, and SCMH1 (large green arrowhead). CRTRs at the less stringent 5% level are also indicated in the ED population as small blue arrowheads and in the XD population as small red arrowheads.

To further examine the possible effects of SNP class SFS bias in identification of CRTRs, we reanalyzed the genome by using only class A SNPs, which comprise 80% of the map overall. Fully 75% of the CRTRs detected by using all SNPs were also CRTRs using only class A. Even in the CRTRs that did not meet the CRTR definition using only class A, all but two (chr3, 89690000–90110000, in ED and chr2, 194650000–194990000, in XD, which were highly class B enriched) still showed a dramatic excess of rare variants, with between 50% and 75% of windows below the 1% empiric threshold. Thus, the majority of CRTRs are robust to potential bias introduced by SNP classes B and C, and nearly all show consistent trends toward unusual SFS using only class A SNPs.

It is also possible that low recombination rates biased our survey toward the identification of low diversity regions that coincide with low recombination rates. The average recombination rate for regions flanking the CRTRs was 0.67 cM/Mb (Table 1), which is significantly lower than the genome-wide average of 1.13 cM/Mb, but only four of the regions fell into recombination deserts with an estimated recombination rate of 0.0 cM/Mb. However, the observed correlation between CRTRs and low recombination regions is also expected if the CRTRs are the product of selective pressure, because the region that is swept along with an advantageous allele will be larger in regions of low recombination. Thus, there does appear to be a modest bias in the data toward identification of CRTRs in regions with low recombination rates.

To assess how well CRTRs in the Perlegen data predict Tajima's D in resequencing data, we selected eight targets from six population-specific CRTRs for resequencing (Table 2). In each CRTR, targets were chosen for one or more of the following reasons: dramatic allele frequency differences between populations in the Perlegen data (EDAR, CLSPN), central position within the CRTR (EDAR, CLSPN, SCMH1, FLJ23878), important gene function (GCG), or the target was a spliced EST in a CRTR without any known genes (AW183861 [GenBank] , BX115137 [GenBank] ). CTPS and FLJ23878 were included because these genes, in combination with SCMH1, accounted for all of the known genes within a single CRTR in the AD population. Values of Tajima's D below –2 for resequencing data are significant under the simplest neutral models (Tajima 1989) and have been asserted to be robustly inconsistent with neutrality under a variety of demographic models (Akey et al. 2004; Stajich and Hahn 2005), although questions regarding the bottleneck models have been raised (Thornton 2005). In four of six CRTRs resequenced, at least one gene was identified with a Tajima's D below –2 in the appropriate population, and the other two CRTRs also show a substantial excess of low-frequency variation in the SFS. In AW183861 [GenBank] the observed Tajima's D was –1.92 in the ED population, placing it in the same range as DCN, which has previously been suggested as a target of selective pressure (Akey et al. 2004; Stajich and Hahn 2005). The other gene with marginal Tajima's D was GCG, which was selected because of its importance in glucose metabolism but lay at the boundary of the CRTR. In this case, the promoter and transcript of the gene (bases 1–11349) lie within the CRTR and showed a significant excess of rare polymorphism in the SFS (Tajima's D = –2.48), while the region 3' of the transcript lies beyond the CRTR and showed a less extreme SFS (Tajima's D = –0.45). Thus, in all six resequenced regions we observed a strong trend toward a negative Tajima's D.

View this table: [in this window] [in a new window]   Table 2. Regions selected for targeted resequencing

We further examined the possibility that positive selective pressure might account for the resequenced CRTRs by calculating Fay and Wu's H statistic for each region (Fay and Wu 2000). H compares the nucleotide diversity estimated from heterozygosity () against nucleotide diversity estimated from the allele frequency of the derived (nonancestral) allele at each position (_H). Significantly negative values of H indicate an excess of high-frequency–derived alleles, consistent with recent, positive selection. Evaluating the significance of a given H depends upon the number of samples and the number of polymorphisms analyzed, so for each of the resequenced regions and in each of the three populations, we simulated 10,000 sample data sets under a standard neutral model, with constant population size and no recombination. In five out of six regions (EDAR, GCG, CLSPN, BX115137 [GenBank] , and AW183861 [GenBank] ), significantly negative H statistics (P = 0.05) were observed in the appropriate populations, and in the AD CRTR (CTPS, FLJ23878, and SCMH1), the negative H statistic approached significance for both FLJ23878 and SCMH1 (Table 3). Taken altogether, the co-occurrence of (1) an unusually low Tajima's D, (2) an unusually high proportion of monomorphic sites in the Perlegen data (Table 2), (3) reduced absolute nucleotide diversity, and (4) the significantly negative Fay and Wu's H values suggests that selective pressure probably accounts for most of the CRTRs.

View this table: [in this window] [in a new window]   Table 3. Fay and Wu's H in resequenced regions

In contrast to the CLSPN CRTR, complete resequencing of the coding regions from CTP synthase (CTPS), FLJ23878 (a predicted gene with supporting spliced EST evidence), and Sex Comb on midleg homolog 1 (SCMH1) failed to identify a single coding variant that was significantly enriched in the AD population. Also in contrast to the CLSPN CRTR, very little recombination was observed across the 300 kilobases containing these three genes in any of the three populations (|D'| = 1 in >95% of pairwise comparisons), so this region is likely to be a recombination coldspot. However, the small amount of observed recombination in the AD population revealed a clear trend across the CRTR, with the lowest Tajima's D region spanning SCMH1. Thus, if a selectively advantageous allele exists in this CRTR, it may lie within the SCMH1 transcript region, but it does not appear to be a coding polymorphism.

EDAR was selected for resequencing because it showed reduced diversity in the XD population, as well as strikingly high levels of F_st between ED and XD populations at a few SNPs. This was confirmed by resequencing, where four SNPs were observed with an allele frequency difference of >85% between populations (SNPs 173, 1663, 2531, 93981; details available at the SeattleSNPs Web site), and three other SNPs showed a difference >50% in allele frequency (SNPs 429, 1158, and 62347). Among the four SNPs with the largest allele frequency differences, the 93891 SNP changes an amino acid, but the change is quite conservative (Val370Ala): It substitutes one small, nonpolar amino acid for another and is predicted to have "benign" effects by Polyphen (Sunyaev et al. 2001). The other three SNPs with the largest allele frequency difference fell within 2000 bp of the transcription start site, suggesting a possible regulatory function for one or more of these SNPs. Although no clear CRTR was detected in this region for the ED population, the nucleotide diversity at EDAR was also significantly reduced in ED. This suggests that a CRTR smaller than our definition (<300 kb across) may exist in the ED population, consistent with a weaker selective force or an older event in the ED population.

The majority of identified CRTRs spanned more than one known gene, although 20% of the CRTRs (11 of 55 CRTRs) did not contain a "Known Gene" in the UCSC Genome Browser track. For example, the CRTR on chromosome 11 in the XD population (chr11, 37820000–38290000) did not contain any known genes or RefSeq entries and had only one spliced EST with multiple exemplars (GenBank BX115137 [GenBank] at chr11, 37,916,727–37,932,789). Resequencing the region spanning BX115137 [GenBank] confirmed the significantly low diversity in this region (Tajima's D= –2.60 in the XD population) (Table 2), thereby confirming findings from the Perlegen data set. This CRTR might reflect direct selective pressure upon this EST, or it might reflect selection upon a long-range regulatory element affecting expression of genes outside of the CRTR in the XD population. It is worth noting that the two of the closest genes to this CRTR are RAG1 and RAG2 (chr11, 36,546,150–36,557,871, and chr11, 36,570,070–36,576,362, respectively), which are essential for adaptive immunity through the rearrangement of T Cell Receptor genes (Fugmann 2001). Therefore, it is possible that long-range regulatory elements affecting the function of these genes could be subject to strong selective pressure.

	Discussion

Top ABSTRACT Results Discussion Methods REFERENCES Web site references

 
Identifying regions of the human genome that have experienced substantial selective pressure can provide insights into the location of functionally important polymorphisms and may help prioritize targets for association mapping (Sabeti et al. 2002; Clark et al. 2003b). This is especially true in genomic regions where selective pressure has been experienced in a geographically restricted manner, where large allele frequency differences in functional variation can exist between geographic subpopulations (Akey et al. 2004; Stajich and Hahn 2005). We demonstrated a significant correlation between Tajima's D in full resequencing data from SeattleSNPs and dense genotyping data from Perlegen. This correlation was exploited to identify large regions of the genome where strong selective pressures have recently been experienced by at least one of the three populations surveyed: AD, ED, and XD. A 100-kbp sliding window analysis was used to identify CRTRs in each population, representing large regions with SFS enriched for low-frequency polymorphisms. We selected eight candidate genes or ESTs within a subset of these CRTRs for resequencing analysis, and although the parameters used to define a CRTR were not theoretically derived, they appear to be conservative in detecting positive selective pressure, given the resequencing results. Thus, we demonstrate that SFSs from dense genotyping data are a useful way to screen the genome for regions that have probably experienced recent and strong selective pressure.

View larger version (89K): [in this window] [in a new window]   Figure 4. (A) A visual genotype for 1.5 Mbp spanning the CLSPN CRTR in the Perlegen data. Each row corresponds to an individual, and each column corresponds to a polymorphic site, with genotypes color coded as follows: Common allele homozygotes are shown in blue, heterozygotes are shown in red, rare allele homozygotes are shown in yellow, and missing data are shown as gray. The top 24 samples are ED, the middle 23 samples are AD, and the bottom 24 samples are XD. Although nucleotide diversity is depressed across a large region, there is no clear minimum within the CRTR. Nucleotide diversity was relatively constant across the region, so CLSPN (shown as a black box) was selected as a target for resequencing because of interesting patterns of Fst between ED and XD, in addition to low nucleotide diversity. (B) A visual genotype of the resequencing results for the CLSPN gene. The top 24 samples are ED; the middle 24 samples are AD, and the bottom 24 samples are XD. As expected, a number of polymorphisms nearly fixated between ED and XD were observed. One of these SNPs (10710, red arrowhead) changes an amino acid (Ser525Asn), whereas the other three are intronic (green arrowheads).

View larger version (79K): [in this window] [in a new window]   Figure 5. A close-up of the CLSPN CRTR from the UCSC genome browser is shown, with the Tajima's D tracks as well as a set of tracks showing the inferred relative recombination rate from LDhat for each population in grayscale (track label, LDhat log RR AD/ED/XD): Darker segments correspond to high inferred recombination rates. CLSPN is located at 35.9 Mbp. The left edge of the CLSPN CRTR (at 35 Mbp in the ED population) corresponds to a strong recombination hotspot observed in all three populations, but of greater interest are the hotspots spanned by the CRTR at 35.4 Mbp and 35.8 Mbp. Thus, although this CRTR does span a region with reduced recombination overall, there are several inferred hotspots within the CRTR that are shared between populations.

Data from these three major populations showed significant differences in the quantity of CRTRs between populations. Overall, relatively few CRTRs were observed in the AD population, and the observed CRTRs were generally smaller than were those in the ED and XD populations. This could be due to less dramatic selective pressure on African populations in recent evolutionary history, but we consider it unlikely that selective pressure from pathogens or diet is substantially weaker in this population. If anything, the pathogen load might be expected to be highest in regions where humans have lived the longest, although the degree of mortality from pathogens may have been attenuated. Alternatively, admixed European chromosomes might serve to obscure Africa-specific selective sweeps, but this possibility also seems unlikely because F_st is low and few if any polymorphisms have fixated between these populations. Thus, admixture between European and African populations tends to reduce Tajima's D by increasing the number of relatively rare SNPs, and admixture from the European population should actually have enhanced detection of AD-specific CRTRs. Demographic parameters such as a population bottleneck in the Eurasian populations (Marth et al. 2004) may also have enhanced detection of selective events in the non-African populations. Also, the higher average Tajima's D in ED populations may simply facilitate the identification of CRTRs in ED, relative to the AD population. A final possibility is that the relatively larger effective population size of the AD population allows greater opportunity for recombination between a selectively advantageous allele and neighboring regions during the course of a selective sweep. In a larger population, the ancestral haplotype bearing the advantageous allele is exposed to recombination in a larger number of individuals, and therefore, the segment of ancestral haplotype that eventually fixates will be shorter than in a smaller population. This hypothesis is supported by the observation that several genes with reportedly significant negative Tajima's D values in the AD population (FY [Hamblin and Di Rienzo 2000; Hamblin et al. 2002] and APOA2 [Fullerton et al. 2002]) were not observed within CRTRs in the AD population at a 1% empiric threshold, or at an even less stringent 5% threshold. Taken together with the relative infrequency of CRTRs in the AD population, this suggests that CRTRs in the AD population may cover shorter segments of the genome than in the ED and XD populations, and will therefore require smaller windows for detection. As the density of genotyping data sets increases (e.g., phase 2 of the HapMap will type millions of additional SNPs, to add to the more than one million SNPs typed in phase 1) (Gibbs et al. 2003), the sliding window size of this type of analysis can be reduced, and smaller CRTRs may become detectable in African populations.

Genes within the CRTR regions may provide important targets for genotype/phenotype studies. For example, CYP3A4 and CYP3A5 play a central role in the metabolism of some prescribed drugs, lie within a CRTR in the ED population, and have been shown to have significantly low Tajima's D in European samples (Thompson et al. 2004). Another example is VKORC1, which has been linked to human warfarin dosing and shown to have low haplotype diversity in Asian populations (Rieder et al. 2005), and is located in a CRTR on chromosome 16 in the XD population. Although we report CRTRs at a 1% empiric threshold, several previously reported genes with significantly negative Tajima's D in European populations could be identified in CRTRs at a less stringent 5% empiric threshold. For example, lactase (LCT) lies within a 1 Mbp CRTR in the ED population at the 5% threshold and has been suggested as a target of selective pressure in the European population (Bersaglieri et al. 2004). Furthermore, a cluster of genes previously suggested as subject to natural selection in Europeans (KEL, TRPV5, TRPV6, and EPHB6) (Akey et al. 2004; Stajich and Hahn 2005) exhibited low Tajima's D in the ED population, but not across a large enough segment to meet our definition of a CRTR. Based on this combined evidence, CRTRs observed at the more stringent 1% empiric threshold (Table 1) seem quite likely to represent selective sweeps.

Resequencing of candidate genes within a number of the CRTRs provides further support for this hypothesis: In every CRTR selected for resequencing, at least one gene with a dramatic departure from the expected distribution of Tajima's D under neutrality was observed (Table 2), and in most CRTRs, a significant departure from the expected distribution of Fay and Wu'sH was also observed (Table 3). In addition to theoretical evidence, the genes resequenced in the CRTRs also fell within the bottom 5% of the empirical distribution of Tajima's D in >170 genes resequenced by SeattleSNPs (Fig. 1). For example, the resequencing-based Tajima's D for SCMH1 was the lowest Tajima's D value that we have ever observed in the AD population, compared with data from 179 resequenced genes. Furthermore, the Tajima's D for genes selected from ED CRTRs was similar to the Tajima's D for genes previously demonstrated to be robustly incompatible with neutrality in several studies (Akey et al. 2004; Stajich and Hahn 2005).

Within each CRTR, it is apparent that a single common haplotype has recently increased dramatically in frequency, at the expense of all other haplotypes within the CRTR. However, it is not yet clear whether the fitness advantage is attributable to a genotype at a single SNP or a haplotype of multiple SNPs. Haplotype effects are more plausible when a single transcript spans the majority of a CRTR (e.g., PHKB on chromosome 16 in XD) or across regions containing groups of functionally related genes (e.g., the Olfactory Receptor gene cluster contained within the AD CRTR on chromosome 11). However, the majority of CRTRs contain multiple genes without clearly related gene functions, so we have not pursued analyses of gene function within the CRTRs, because it is likely that many if not most of the genes within the CRTRs simply represent hitchhiking events where the advantageous allele within a single gene swept the neighboring genes along with it as it increased in frequency (Fay and Wu 2000). Identification of viable candidates for these selective effects would at a minimum require complete resequencing of all functional regions within a CRTR and follow-up of any potentially interesting variants. As more complete resequencing data become available within each CRTR, other SFS test statistics, such as the Fay and Wu's H test, can potentially be applied to each CRTR to narrow the candidate interval containing the advantageous variant.

The pattern of concordance for CRTRs between ED and XD populations was also interesting: An overlap of three CRTRs between these populations is quite striking, given that the CRTRs comprise <1% of the genome. Given that Tajima's D values from the resequencing data are in the range of genes previously reported to be inconsistent with neutrality under a range of demographic parameters, we believe that the shared CRTRs probably represent shared selective pressures between these populations. However, not all shared CRTRs necessarily represent a single selective sweep in multiple populations. For example, Tajima's D was significantly low at EDAR in both XD and ED populations, but the genomic extent of the CRTR is substantially greater in XD than ED populations. This could represent sweeps that occurred at different times historically, but the extreme F_st at a series of polymorphisms in EDAR is consistent with either a divergent sweep with one haplotype favored in ED and a different haplotype favored in XD, or parallel sweeps favoring an allele shared by both haplotypes but not by other African haplotypes (e.g., site 96563 in the 3' UTR, rs1478517). No CRTRs were observed to be shared between all three populations, but this would be consistent with the ascertainment bias toward high-frequency SNPs: If no high-frequency SNPs exist in any of the three populations, then no SNPs were available for use in the Perlegen data. Therefore, although global CRTRs were not observed, such regions may be present as large regions without genotype data in the Perlegen data set.

Considering each of the regions resequenced, identification of the specific SNP or SNPs conferring a selective advantage is not trivial. For example, although the patterns of SFS suggest that SCMH1 is likely to harbor the advantageous allele responsible for the selective sweep that created the CRTR in the AD population, no coding polymorphism was identified in the AD population with significantly enriched allele frequency in either SCMH1 or the other two genes. Given that we resequenced all of the known coding regions in this CRTR, if a polymorphism within SCMH1 drove the sweep, then the function of the polymorphism was probably regulatory rather than structural. In contrast, the CLSPN and EDAR resequencing data identified interesting candidate cSNPs, and selective pressure on these cSNPs could conceivably account for the EDAR and CLSPN CRTRs. More extensive resequencing within each CRTR is required to determine whether other candidate SNPs exist in neighboring genes.

In conclusion, the availability of adequately dense genotyping data sets clearly facilitates the identification of regions of the human genome with unusual SFS, which may have been subjected to strong positive selective pressure in the recent past. Current data appear to be adequate to identify such regions in ED and CD populations, but denser data will be necessary for analysis of AD populations, probably due to the larger effective population size of this population. Detection of regions subject to balancing selection (e.g., HLA) or with less complete selective sweeps (e.g., FY) will probably require a substantially denser data set than is currently available. Although most CRTRs span multiple genes, within each CRTR the selective sweep favored only one haplotype at the expense of all others, so a single selectively advantageous polymorphism in a single gene could conceivably account for each CRTR, with the reduced diversity in flanking regions representing a hitchhiking event. Dissection of the underlying functional variant (or variants) within each CRTR may require comprehensive resequencing within the CRTR to identify candidate functional variation, but where it is feasible, functional analysis of a priori functional variants (e.g., the CLSPN Ser525Asn cSNP) should substantially accelerate this process.

	Methods

Top ABSTRACT Results Discussion Methods REFERENCES Web site references

 
Samples
Twenty-four individuals from each of three populations were resequenced: 24 African American individuals from the Coriell HD100AA diversity panel (population AD), 24 CEPH individuals (population ED), and 24 Chinese Americans from the Coriell HD100A diversity panel (population XD). All ED individuals overlap with the Perlegen European panel (dbSNP population 1371); all but one of the AD individuals overlap with the Perlegen African American panel (dbSNP population 1372); and all XD individuals overlap with the Perlegen Chinese panel (dbSNP population 1373). Coriell accession numbers are as follows: AD population, NA17101–NA17116, NA17133–NA17140; ED population, NA06990, NA07019, NA07348, NA07349, NA10830, NA10831, NA10842, NA10843, NA10844, NA10845, NA10848, NA10850, NA10851, NA10852, NA10853, NA10854, NA10857, NA10858, NA10860, NA10861, NA12547, NA12548, NA12560, NA17201; and XD population, NA17733–NA17747, NA17749, NA17752–NA17757, NA17759, NA17761.

Perlegen data
All genotype data for populations 1371 (ED), 1372 (AD), and 1373 (XD) from build 124 of dbSNP were downloaded and parsed for analysis. Several quality controls were applied to the data set to verify that the data were completely and accurately retrieved. Comparison of the 46 samples overlapping between Perlegen and SeattleSNPs identified 95,354 concordant genotypes out of 96,170 genotypes compared (>99.2%; consistent with Hinds et al. 2005). We also observed good concordance between the total number of Perlegen SNPs retrieved from dbSNP (1.58 million) and the numbers reported by Hinds et al. The Perlegen SNP class data for each SNP were downloaded for each autosome from the Perlegen Web site (http://genome.perlegen.com/browser/download.html).

Sequencing analysis
Methods for sequence analysis of candidate genes obtained from SeattleSNPs (http://pga.gs.washington.edu) have previously been described in detail (Carlson et al. 2003, 2004; Livingston et al. 2004). Briefly, PCR amplicons were tiled across the full genomic sequence of a gene or selected genomic regions and sequenced by using standard BigDye Terminator v3.1 methods. The amplification primers and sequencing protocol are available at http://pga.gs.washington.edu. Sequence data from the ABI 3730XL instrument were base called with Phred (Ewing et al. 1998) and assembled onto a reference sequence with the add reads feature in Consed (Gordon et al. 1998). Polymorphisms were identified by PolyPhred v4.29 (Nickerson et al. 1997), and Consed was used to visualize the sequences and confirm polymorphisms.

Nucleotide diversity analysis
There are several statistics that can be used to describe nucleotide diversity, including _s (equation 1), (equation 2), and _H (equation 3). These statistics can be calculated for a given resequencing data set by using the following parameters: n is the number of chromosomes resequenced, Sn is the number of polymorphic sites observed, p_i is the derived (nonancestral) allele frequency of the ith SNP, and q_i is the ancestral allele frequency of the ith SNP.

(1)

(2)

(3)

There are many statistics that can evaluate departures from the expected patterns of neutral variation. One of these is Tajima's D (Tajima 1989), equation 4:

(4)

The theoretical distribution of Tajima's D (95% confidence interval between –2 and +2) assumes that polymorphism ascertainment is independent of allele frequency. High values of Tajima's D suggest an excess of common variation in a region, which can be consistent with balancing selection or population contraction. Negative values of Tajima's D, on the other hand, indicate an excess of rare variation, consistent with population growth, or positive selection. Population admixture can lead to either high or low Tajima's D values in theory. Demographic parameters would be expected to affect the genome more evenly than selective pressures, so previous analyses have suggested that using the empiric distribution of Tajima's D from a collection of regions across the genome provides advantages in assessing whether selection or demography might explain an observed deviation from expectation (Akey et al. 2004; Stajich and Hahn 2005). Because of the ascertainment bias toward common polymorphism in the Perlegen data set, extremely positive Tajima's D values are difficult to interpret, and modeling ascertainment is difficult. However, given that the ascertainment bias raises the mean of the distribution, extreme negative values in extended regions can be useful in qualitatively identifying interesting regions for full resequencing and more rigorous theoretical analysis of nucleotide diversity.

For genic comparisons between SeattleSNPs data and Perlegen data, Tajima's D in the SeattleSNPs data was calculated for all observed polymorphic sites. The median transcript size in the SeattleSNPs data set is 14,649 bp, and on average, an additional 3000 bp of flanking sequence was also resequenced, for a median analyzed region of 17.5 kbp. Given the density of the Perlegen map, the median number of polymorphic SNPs in the resequenced region was only six in the ED population and seven in the AD population, a rather small number of polymorphisms for Tajima's D estimation. One hundred nineteen out of 179 genes analyzed in the AD population had five or more polymorphic Perlegen SNPs within the resequenced region, and in the ED population, 107 genes had five or more polymorphic Perlegen SNPs. Significant linkage disequilibrium routinely extends 10 kb, even in the AD population, so patterns of nucleotide diversity should be conserved over similar distances. We extended the region analyzed from the Perlegen data to include 10 kb upstream and 10 kb downstream of the transcript, under the expectation that this would increase the number of sites per gene and therefore the accuracy of the Tajima's D estimate. Expanding the region raised the median number of polymorphic sites per gene to 16 in ED and 19 in AD. As expected, the larger number of sites per gene improved the correlation between the SeattleSNPs and Perlegen data substantially. In the AD population, R² = 0.29 in extended versus R² = 0.04 in the transcript for the 119 genes with five or more polymorphic SNPs in the transcript, and in the ED population, R² = 0.66 in extended versus R² = 0.15 in transcript for the 107 genes with five or more polymorphic SNPs in the transcript. Thus, for the genic comparison in Figure 1, Tajima's D in the Perlegen data was calculated on the basis of all observed polymorphic sites within 10 kb of the longest reported transcript in Entrez Gene. Within each population, only genes with five or more polymorphic SNPs in the Perlegen data were included in the comparison of data sets, which yielded 178 genes in the AD population and 173 genes in the ED population.

For the windowed analysis of the genome, Tajima's D was calculated independently in each population. Sliding windows of 100 kb were analyzed across all autosomal regions in the Perlegen data, stepping by 10 kb. Thus, the first window evaluated on chromosome 1 was genome coordinates chr1, 1–100,000; the second window was genome coordinates chr1, 10,001–110,000; and so forth. Because adjacent windows overlap, in the genome browser track, the Tajima's D is reported for each window using the coordinates of the central 10 kb. Thus, the observed Tajima's D for window chr1, 1–100,000, is reported at chr1, 45,001–55,000. These data have been made available as a track in the UCSC genome browser (http://genome.ucsc.edu/).

The empirically determined distribution of Tajima's D within the sliding windows was used to identify CRTRs, defined as a region of 20 contiguous windows where >75% of the windows were in the bottom 1% of the empirical Tajima's distribution. The empirically determined bottom 1% of Tajima's D values corresponded to Tajima's D <–0.70 in the AD population, Tajima's D<–1.49 in the ED population, and Tajima's D<–1.743 in the XD population. Window size was chosen to provide a reasonably large number of SNPs per window (average 54.3 SNPs per window in AD, 47.1 in ED, and 42.8 in XD). Under the restriction that 75% of all windows must be below threshold within a contiguous region, the distribution of contiguous region sizes for each population is shown in Supplemental Figure 1: In each population, <10% of all such contiguous regions exceeded 20 windows in length, so we used a threshold of 20 adjacent windows to define a CRTR. Given that each window overlaps its neighbor by 90,000 bp, a stretch of 20 contiguous windows corresponds to 300,000 bp.

Fay and Wu's H analysis
The chimpanzee allele was used to determine the ancestral human allele within the eight resequenced regions. Only SNPs where the chimpanzee alignment was unambiguous and matched one of the existing human alleles were used in this analysis. In each population, 24 samples were resequenced, representing 48 chromosomes, so Fay and Wu's H (H= – _H) was calculated within each population and then compared against simulated data. Simulations were run for each region in each of the three populations, under the conservative assumption of no recombination. Simulations were performed by using mksamples (Hudson 2002), assuming 48 chromosomes and the appropriate number of SNPs (Sn) with ancestral allele information (Table 3). For each analysis, 10,000 simulations were performed, and the proportion of simulations with more negative H than the observed data (equivalent to a P-value) is shown in Table 3.

Recombination rate analysis
Recombination was addressed in two ways: Large-scale recombination rates for the region spanning each CRTR were estimated from the deCODE sex averaged recombination rate (Kong et al. 2002) for the flanking megabase, when the CRTR fell within a single deCODE interval, and by weighted averages of the flanking megabase when the CRTR spanned a boundary between deCODE intervals. At a much finer scale, the relative recombination rate per nucleotide was estimated across 3 Mb flanking the CLSPN CRTR by using the pairwise option within the LDhat program (McVean et al. 2004). This program first estimates the coalescent likelihood of observing each pair of segregating sites, treating each pair as independent, and then estimates the recombination rate for the entire region over a grid. Polymorphism data from chr1, 34,500,000–37,500,000, was analyzed independently for each population. The estimated fine-scale recombination rates across this region for each population are shown in Figure 5.

	Acknowledgements

 
This work was supported by a Program for Genomic Applications grant from the National Heart, Lung, and Blood Institute (HL66682 and HL66642 to D.N. and M.R.). D.T. was supported by grants from the National Human Genome Research Institute (IP41HG02371 and HG02238 to David Haussler). We thank Dana Crawford, Alex Reiner, and Eric Torskey for comments on the manuscript, as well as the entire SeattleSNPs resequencing team for their extraordinary efforts on this project.

	Footnotes

 
[Supplemental material is available online at www.genome.org.]

Article and publication are at http://www.genome.org/cgi/doi/10.1101/gr.4326505. Freely available online through the Genome Research Immediate Open Access option.

³ Corresponding author.
E-mail csc47{at}u.washington.edu; fax (206) 221-6498. E-mail debnick{at}u.washington.edu; fax (206) 221-6498.

	REFERENCES

Top ABSTRACT Results Discussion Methods REFERENCES Web site references

 

Akey, J.M., Eberle, M.A., Rieder, M.J., Carlson, C.S., Shriver, M.D., Nickerson, D.A., and Kruglyak, L. 2004. Population history and natural selection shape patterns of genetic variation in 132 genes. PLoS Biol. 2: e286.[CrossRef][Medline]

Bersaglieri, T., Sabeti, P.C., Patterson, N., Vanderploeg, T., Schaffner, S.F., Drake, J.A., Rhodes, M., Reich, D.E., and Hirschhorn, J.N. 2004. Genetic signatures of strong recent positive selection at the lactase gene. Am. J. Hum. Genet. 74: 1111–1120.[CrossRef][Medline]

Carlson, C.S., Eberle, M.A., Rieder, M.J., Smith, J.D., Kruglyak, L., and Nickerson, D.A. 2003. Additional SNPs and linkage-disequilibrium analyses are necessary for whole-genome association studies in humans. Nat. Genet. 33: 518–521.[CrossRef][Medline]

Carlson, C.S., Eberle, M.A., Rieder, M.J., Yi, Q., Kruglyak, L., and Nickerson, D.A. 2004. Selecting a maximally informative set of single-nucleotide polymorphisms for association analyses using linkage disequilibrium. Am. J. Hum. Genet. 74: 106–120.[CrossRef][Medline]

Clark, A.G., Glanowski, S., Nielsen, R., Thomas, P., Kejariwal, A., Todd, M.J., Tanenbaum, D.M., Civello, D., Lu, F., Murphy, B., et al. 2003a. Positive selection in the human genome inferred from human–chimp–mouse orthologous gene alignments. Cold Spring Harb. Symp. Quant. Biol. 68: 471–477.[Medline]

Clark, A.G., Glanowski, S., Nielsen, R., Thomas, P., Kejariwal, A., Todd, M.J., Tanenbaum, D.M., Civello, D., Lu, F., Murphy, B., et al. 2003b. Inferring nonneutral evolution from human–chimp–-mouse orthologous gene trios. Science 302: 1960–1963.[Abstract/Free Full Text]

Ewens, W.J. 1972. The sampling theory of selectively neutral alleles. Theor. Popul. Biol. 3: 87–112.[CrossRef][Medline]

Ewens, W.J. 1979. Mathematical population genetics. Springer-Verlag, New York.

Ewing, B., Hillier, L., Wendl, M.C., and Green, P. 1998. Base-calling of automated sequencer traces using phred, I: Accuracy assessment. Genome Res. 8: 175–185.[Abstract/Free Full Text]

Fay, J.C. and Wu, C.I. 2000. Hitchhiking under positive Darwinian selection. Genetics 155: 1405–1413.[Abstract/Free Full Text]

Fu, Y.X. and Li, W.H. 1993. Statistical tests of neutrality of mutations. Genetics 133: 693–709.[Abstract]

Fugmann, S.D. 2001. RAG1 and RAG2 in V(D)J recombination and transposition. Immunol. Res. 23: 23–39.[CrossRef][Medline]

Fullerton, S.M., Clark, A.G., Weiss, K.M., Taylor, S.L., Stengard, J.H., Salomaa, V., Boerwinkle, E., and Nickerson, D.A. 2002. Sequence polymorphism at the human apolipoprotein AII gene (APOA2): Unexpected deficit of variation in an African-American sample. Hum. Genet. 111: 75–87.[CrossRef][Medline]

Gibbs, R.A., Belmont, J.W., Hardenbol, P., Willis, T.D., Yu, F., Yang, H., Ch'ang, L.Y., Huang, W., Liu, B., Shen, Y., et al. 2003. The International HapMap Project. Nature 426: 789–796.[CrossRef][Medline]

Gordon, D., Abajian, C., and Green, P. 1998. Consed: A graphical tool for sequence finishing. Genome Res. 8: 195–202.[Abstract/Free Full Text]

Hamblin, M.T. and Di Rienzo, A. 2000. Detection of the signature of natural selection in humans: Evidence from the Duffy blood group locus. Am. J. Hum. Genet. 66: 1669–1679.[CrossRef][Medline]

Hamblin, M.T., Thompson, E.E., and Di Rienzo, A. 2002. Complex signatures of natural selection at the Duffy blood group locus. Am. J. Hum. Genet. 70: 369–383.[CrossRef][Medline]

Harpending, H. and Rogers, A. 2000. Genetic perspectives on human origins and differentiation. Annu. Rev. Genomics Hum. Genet. 1: 361–385.[CrossRef][Medline]

Hinds, D.A., Stuve, L.L., Nilsen, G.B., Halperin, E., Eskin, E., Ballinger, D.G., Frazer, K.A., and Cox, D.R. 2005. Whole-genome patterns of common DNA variation in three human populations. Science 307: 1072–1079.[Abstract/Free Full Text]

Hudson, R.R. 2002. Generating samples under a Wright-Fisher neutral model of genetic variation. Bioinformatics 18: 337–338.[Abstract/Free Full Text]

Hughes, A.L. and Yeager, M. 1998. Natural selection and the evolutionary history of major histocompatibility complex loci. Front. Biosci. 3: d509–d516.[Medline]

Kent, W.J., Sugnet, C.W., Furey, T.S., Roskin, K.M., Pringle, T.H., Zahler, A.M., and Haussler, D. 2002. The human genome browser at UCSC. Genome Res. 12: 996–1006.[Abstract/Free Full Text]

Kimura, M. 1983. The neutral theory of molecular evolution. Cambridge University Press, Cambridge, UK.

Kong, A., Gudbjartsson, D.F., Sainz, J., Jonsdottir, G.M., Gudjonsson, S.A., Richardsson, B., Sigurdardottir, S., Barnard, J., Hallbeck, B., Masson, G., et al. 2002. A high-resolution recombination map of the human genome. Nat. Genet. 31: 241–247.[CrossRef][Medline]

Libert, F., Cochaux, P., Beckman, G., Samson, M., Aksenova, M., Cao, A., Czeizel, A., Claustres, M., de la Rua, C., Ferrari, M., et al. 1998. The deltaccr5 mutation conferring protection against HIV-1 in Caucasian populations has a single and recent origin in Northeastern Europe. Hum. Mol. Genet. 7: 399–406.[Abstract/Free Full Text]

Livingston, R.J., von Niederhausern, A., Jegga, A.G., Crawford, D.C., Carlson, C.S., Rieder, M.J., Gowrisankar, S., Aronow, B.J., Weiss, R.B., and Nickerson, D.A. 2004. Pattern of sequence variation across 213 environmental response genes. Genome Res. 14: 1821–1831.[Abstract/Free Full Text]

Marth, G.T., Czabarka, E., Murvai, J., and Sherry, S.T. 2004. The allele frequency spectrum in genome-wide human variation data reveals signals of differential demographic history in three large world populations. Genetics 166: 351–372.[Abstract/Free Full Text]

McVean, G.A., Myers, S.R., Hunt, S., Deloukas, P., Bentley, D.R., and Donnelly, P. 2004. The fine-scale structure of recombination rate variation in the human genome. Science 304: 581–584.[Abstract/Free Full Text]

Nickerson, D.A., Tobe, V.O., and Taylor, S.L. 1997. PolyPhred: Automating the detection and genotyping of single nucleotide substitutions using fluorescence-based resequencing. Nucleic Acids Res. 25: 2745–2751.[Abstract/Free Full Text]

Nielsen, R., Hubisz, M.J., and Clark, A.G. 2004. Reconstituting the frequency spectrum of ascertained single-nucleotide polymorphism data. Genetics 168: 2373–2382.[Abstract/Free Full Text]

Nielsen, R., Bustamante, C., Clark, A.G., Glanowski, S., Sackton, T.B., Hubisz, M.J., Fledel-Alon, A., Tanenbaum, D.M., Civello, D., White, T.J., et al. 2005. A Scan for positively selected genes in the genomes of humans and chimpanzees. PLoS. Biol. 3: e170.[CrossRef][Medline]

Rieder, M.J., Reiner, A.P., Gage, B.F., Nickerson, D.A., Eby, C.S., McLeod, H.L., Blough, D.K., Thummel, K.E., Veenstra, D.L., and Rettie, A.E. 2005. Effect of VKORC1 haplotypes on transcriptional regulation and warfarin dose. N. Engl. J. Med. 352: 2285–2293.[Abstract/Free Full Text]

Sabeti, P.C., Reich, D.E., Higgins, J.M., Levine, H.Z., Richter, D.J., Schaffner, S.F., Gabriel, S.B., Platko, J.V., Patterson, N.J., McDonald, G.J., et al. 2002. Detecting recent positive selection in the human genome from haplotype structure. Nature 419: 832–837.[CrossRef][Medline]

Seltsam, A., Hallensleben, M., Kollmann, A., and Blasczyk, R. 2003. The nature of diversity and diversification at the ABO locus. Blood 102: 3035–3042.[Abstract/Free Full Text]

Smith, J.M. and Haigh, J. 1974. The hitch-hiking effect of a favourable gene. Genet. Res. 23: 23–35.[Medline]

Stajich, J.E. and Hahn, M.W. 2005. Disentangling the effects of demography and selection in human history. Mol. Biol. Evol. 22: 63–73.[Abstract/Free Full Text]

Straus, D.S. and Taylor, C.E. 1981. Hitchhiking and linkage disequilibrium between hemoglobin S and nearby restr