Insertion-deletion polymorphisms (indels) have to date received less attention in the study of sequence variation than have single nucleotide polymorphisms (SNPs), despite their frequency (estimated at approximately 16% to 25% of all sequence polymorphism events) and their potential functional importance 1. 5' Untranslated regions (UTRs) and gene coding regions have previously been observed to have lower indel rates compared with other regions, suggesting that the constraint may have arisen because of negative selection 2. In general, indels that give rise to frame shifts in coding sequence are more disruptive than non frame-shifts and single point mutations, because of third base degeneracy 3. As a result, coding sequence indels tend to have lengths that are multiples of three, whereas regulatory sequences tend to have more frequent indels that occur in distinct blocks 4. The majority of indels are di-allelic and small, with allele length differences of relatively few (one to four) nucleotides 256. Given their frequency, small indels could play an important role in contributing to phenotypic differences in humans, including susceptibility to diseases. It is therefore of interest to characterize indel distribution across the human genome, and to integrate indels into SNP marker maps in order to aid in the identification of natural genetic variation.

Recent theoretical work has considered the distribution of indels under neutrality and exploited the evolutionary imprint of sequence indels in order to pinpoint functional DNA regions that are subject to purifying selection 7. Snir and Pachter 8 used Encyclopedia of DNA Elements (ENCODE) data and multiple primate sequences to study indel events between species. This work suggests that indel rates genome wide are not uniform and that indel events are not neutral; in particular, the work has identified indel hotspots in the human genome. A minority of insertions and deletions may also have plausibly played a major role in speciation events, including human-chimpanzee phenotypic differences 910. An investigation of 2,000 human di-allelic indels found that the majority were monomorphic in chimpanzees and gorillas, indicating that most indels have arisen after the most recent common primate ancestor 6 and are lineage specific 5.

We used the small insertion and deletion ENCODE data 11 to address four questions. First, do the 14 manually selected regions have lower insertion and deletion rates compared with the 30 randomly selected regions? This might be expected to be the case if the selection process 12 for the manually selected ENCODE regions of interest were biased toward regions with greater density of genes or genes of evolutionary importance, with greater functional and evolutionary constraints. Second, do indel rates vary by genomic annotation feature (in turn reflecting varying levels of functional constraint)? Indels that arise in coding sequence are more likely to be deleterious and therefore subject to purifying selection. As a result, DNA sequences that encode proteins might be expected to have some of the lowest genomic indel rates, followed by a wide variety of functional features that are believed to regulate gene expression via an increasing number of previously unrecognized mechanisms 1314151617.

Third, are indel rates negatively correlated with measures of evolutionary constraint? We expect indel rates to be negatively associated with evolutionary constraint scores (see Materials and methods, below) where DNA sequences are subject to purifying selection. To address this question, we also correlated indel rates with ancestral repeat (AR) sequence. AR sequences are mobile elements that inserted before the common ancestor of most mammals and have subsequently become inactive 18. ARs are considered to be predominantly neutral sequences (not subject to purifying selection) and hence we would anticipate indels to accumulate in AR sequence regions with relatively little or no constraint. Based on the assumption that new indels have arisen in AR regions in the past at the same rate as elsewhere in the genome, observed indel rates might be expected to be positively correlated with AR sequence rates.

The fourth question we consider is how do ENCODE indel rates compare with SNP rates across genomic features and evolutionary constrained sequence?

Here we describe the distribution of small indels (ranging from 1 to 20 base pairs [bp]) in the manually and randomly selected ENCODE regions, their distribution in relation to genomic annotation features, and their relationship with measures of evolutionary constraint.

All identified small indels (n = 4486) in the ENCODE regions were mapped onto physical coordinates for ENCODE functional features. The average indel length of identified small indels is 2.8 bp, ranging from 1 to 20 bp. The overall density is on average 15 indels per 100 kilobases (kb; 99% confidence interval [CI] 13.4 to 16.7) or, in terms of total indel length, 43.4 bp per 100 kb (99% CI 38.3 to 49.1). All results in Tables 1 to 3 are presented in two ways: as numbers of indel events (indels per 100 kb) and total indel length (indel bp per 100 kb). In the interests of brevity, indel rates are referred to in the text to as indel bp per 100 kb unless stated otherwise. This also facilitates comparison with SNP rates.

There are no substantial differences in indel or gene density between manually and randomly selected regions (Table 1). The indel rates in manual regions are similarly variable (sd_{num/100 kb} = 5.0 number of indels per 100 kb; sd_{bp/100 kb} = 14.7 indel bp per 100 kb, where sd_{num/100 kb} and sd_{bp/100 kb} refer to the standard deviation for number of indels and indel bp per 100 kb, respectively) to those in random regions (sd_{num/100 kb} = 4.0; sd_{bp/100 kb} = 14.0), with no significant differences in the summary data (F_[13,29] = 1.52, P = 0.34).

We observed a reduction in indel rates for coding sequence and annotation features that are believed to play a regulatory role in gene expression (Table 2). Compared with the overall mean (43.4 bp per 100 kb), ENCODE coding sequences all exhibit a significant reduction in indel rates, as assessed by identifying open reading frames (coding sequence [CDS] mean indel rate: 0.7 bp per 100 kb), transcription start sites (TSSs; 3.3 bp per 100 kb), rapid amplification of cDNA ends fragments (RACEfrags; 6.6 bp per 100 kb), and transcribed fragments (12.3 bp per 100 kb). Pseudo-exons (19.1 bp per 100 kb), 3' UTRs (23.6 bp per 100 kb), 5' UTRs (27.4 bp per 100 kb), and transcripts of unknown function (36.9 bp per 100 kb) all exhibit a reduction in indel rates compared with the overall mean for all ENCODE sequence, but these findings are not statistically significant.

Potential regulatory elements, assessed by measuring open chromatin sites, also reveal sequences with constrained indel rates (Table 2). Formaldehyde assisted isolation of regulatory elements (FAIRE) sites (23.8 bp per 100 kb) and DNAse hypersensitive sites (DHS; [NHGRI group] 19.7 bp per 100 kb and [Regulome group] 27.0 bp per 100 kb) both exhibit reduced indel rates. DHS are short regions of DNA that are relatively easily cleaved by DNAse I.

Acetylated histones are usually associated with transcriptionally active chromatin and deacetylated histones with inactive chromatin. Hence, histone modified regions often signify regulatory sites. Selected histone modifications and binding sites for RNA polymerase II and the general transcription factor TAF250 were assayed for the ENCODE regions (see ENCODE Project Consortium 19 and Table 4 for details). These sites show modestly reduced indel rates (HisPolTAF: 32.4 bp per 100 kb), along with sites occupied by sequence specific binding proteins (all motifs: 35.8 bp per 100 kb), but neither finding is statistically significant.

Experimental feature definitions

Feature

Term

Definition

---

RNA transcription (coding and noncoding)

CDS

Coding sequence: well characterized transcribed regions with an annotated protein-coding open reading frame (ORF)

RACEfrags

5' and 3' rapid Amplification of cDNA ends (RACE), using polyA or total RNA to construct full-length cDNA. This technique has revealed previously unrecognized UTRs

TARs/transfrags

Transcriptionally active regions/transcribed fragments as determined by analyses of cellular RNA (polyA or total) hybridizations to multiple microarray platforms. For the analyses reported here, portions of TARs/transfrags overlapping any CDS, 5' or 3' UTR annotations were removed from the dataset

Pseudo-exons

A pre-mRNA sequence that resembles an exon but is not recognized as such by the splicing machinery

TSS

Transcription start site

5' UTR

Untranslated region: portions of CDS-containing transcripts before the start codon. For the analyses reported here, 5' UTRs overlapping alternatively transcribed CDS annotations were removed from the dataset

TUF

Transcripts of unknown function for noncoding transcripts

3' UTR

Untranslated region: portions of CDS-containing transcripts after the stop codon

Transcript regulation: open chromatin/DNA-protein interaction

DHS

DNAse I hypersensitive sites are short regions of DNA that are relatively easily cleaved by deoxyribonuclease. Regions of open chromatin detected by quantitative chromatin profiling and novel microarray-based methods. For the analyses reported here, regions that overlap repetitive sequence were removed. Measures of DHS are reported using two sources: the ENCODE Regulome group and the NHGRI

FAIRE-sites

Formaldehyde assisted isolation of regulatory elements: a procedure used to isolate chromatin that is resistant to the formation of protein-DNA crosslinks. Data suggest that depletion of nucleosomes (the most basic organizational unit of chromatin) at active regulatory regions, such as promotors, is the primary underlying basis for FAIRE [38]

HisPolTAF

Histone modifications, RNA polymerase II (PolII), and transcription regulator TAF250

Sequence specific factors

Regions of DNA determined to be bound by sequence-specific transcription factors through chromatin immunoprecipitation followed by microarray chip hybridization (so-called 'ChIP-Chip') analyses

Sequence specific (all motifs)

Computationally identified short sequence motifs found to be over-represented in the sequence specific factors dataset

Ancestral repeats

Mobile elements with well defined consensus sequences that inserted into the ancestral genome prior to mammalian radiation. These sequences are considered to be predominantly non-functional and are often used as models of neutrally evolving DNA

Cell cycle

EarlyRepSeg

Early replicating segments

MidRepSeg

Mid replicating segments

LateRepSeg

Late replicating segments

Evolutionary constraint

MCS strict

Multi-species conserved sequences: strict criteria

MCS moderate

Multi-species conserved sequences: modest criteria