SNPGenie: estimating evolutionary parameters to detect natural selection using pooled next-generation sequencing data

Abstract

Summary: New applications of next-generation sequencing technologies use pools of DNA from multiple individuals to estimate population genetic parameters. However, no publicly available tools exist to analyse single-nucleotide polymorphism (SNP) calling results directly for evolutionary parameters important in detecting natural selection, including nucleotide diversity and gene diversity. We have developed SNPGenie to fill this gap. The user submits a FASTA reference sequence(s), a Gene Transfer Format (.GTF) file with CDS information and a SNP report(s) in an increasing selection of formats. The program estimates nucleotide diversity, distance from the reference and gene diversity. Sites are flagged for multiple overlapping reading frames, and are categorized by polymorphism type: nonsynonymous, synonymous, or ambiguous. The results allow single nucleotide, single codon, sliding window, whole gene and whole genome/population analyses that aid in the detection of positive and purifying natural selection in the source population.

Availability and implementation: SNPGenie version 1.2 is a Perl program with no additional dependencies. It is free, open-source, and available for download at https://github.com/hugheslab/snpgenie.

Contact: nelsoncw@email.sc.edu or austin@biol.sc.edu

Supplementary information: Supplementary data are available at Bioinformatics online.

1 Introduction

Next-generation sequencing (NGS) technologies allow the rapid sequencing of pooled DNA samples containing genetic material from multiple individuals. The resultant single-nucleotide polymorphism (SNP) data may be used to reliably estimate population genetic parameters with more accuracy and less expense than the separate sequencing of multiple individuals (Futschik and Schlötterer, 2010; Lynch et al., 2014), especially when samples are large and coverage is high. Unfortunately, high coverage data also suffer from a substantial false-positive error rate. SNP calling techniques can address this issue, but the only software currently available for evolutionary analysis of pooled NGS data, PoPoolation (Kofler et al., 2011), is inextricable from a problematic SNP caller that has an extremely high false-positive rate (Raineri et al., 2012). Further, PoPoolation relies on large pileup files and problematic simplifications, including use of the reference sequence alone to determine the number of nonsynonymous and synonymous sites. Ideally, software for evolutionary analyses of these data would allow users to first call SNPs using any preferred method, and then process the results using standard methods for determining the numbers of nonsynonymous and synonymous differences and sites.

We have developed SNPGenie to meet this need. Using SNP calling results, SNPGenie estimates: (i) nucleotide diversity (π), and its nonsynonymous and synonymous partitions (π_N and π_S) for coding regions; (ii) mean nonsynonymous and synonymous divergence (d_N and d_S) from a reference sequence; (iii) gene diversity (H_O; Nei, 1987); (iv) site type classification (nonsynonymous, synonymous or ambiguous) for polymorphic coding loci (Knapp et al. 2011); and (v) the constraint imposed by overlapping open reading frames. These parameters do not depend on linkage (Nelson and Hughes, 2015), circumventing a major limitation of pooled data for other applications (Cutler and Jensen, 2010). Indefinitely large genomes with multi-nucleotide variants may be analysed at speeds exceeding those of PoPoolation’s default settings. The results allow users to test evolutionary hypotheses on the roles of negative (purifying) selection, positive (Darwinian) selection and random genetic drift in the sampled population. In general, π_N = π_S indicates neutrality, π_N < π_S indicates purifying selection and π_N > π_S may indicate positive selection favoring multiple amino acid changes (Hughes, 1999). Comparing H_O at distinct polymorphic site categories may also address these hypotheses (Hughes et al. 2011). Parameter estimates are available at the nucleotide, codon, sliding window, whole gene and whole genome/population levels.

2 Methods

SNPGenie is a data processing Perl script, with no additional dependencies. The program accepts a reference sequence(s) (.FASTA), a Gene Transfer Format (.GTF) file with CDS annotations and an arbitrary number of SNP reports, currently including Geneious (Variations/SNPs Annotations Table) and CLC Genomics Workbench (Annotated Variant File) formats. The download package includes the Perl code, a detailed manual (README), various example files and scripts to aid in data preparation.

Nucleotide diversity (π) is the mean number of pairwise differences per site in a population of sequences. SNPGenie estimates this for all sites, and then separately for nonsynonymous and synonymous coding sites (π_N and π_S), using a new method (Nelson and Hughes, 2015) based on that of Nei and Gojobori (1986). Differences are calculated using all comparisons within every polymorphic codon and including all mutational pathways. To calculate the numbers of nonsynonymous and synonymous sites, SNPGenie weights by the sample allele frequencies. This becomes especially important when populations diverge from the reference sequence(s). Gene diversity is calculated as H₀ = 1 – ∑x_i², where x_i is the population frequency of nucleotide i. Polymorphic coding sites are classified following the methods of Knapp et al. (2011), with gene diversities given for each category.

SNPGenie (version 1.2) and PoPoolation (version 1.2.2) were used to analyse pooled H5N1 data from ferret #3501-DPI5, obtained from Wilker et al. (2013) (Jorge Dinis, personal correspondence). SNPGenie used SNP calling results from Geneious Version 5.6.3, while PoPoolation necessarily performed its own SNP calling. For SNPGenie, all default values were used. For PoPoolation, the Syn-nonsyn-sliding.pl script was used with default settings, except max-coverage = 100 000, dissable-corrections = on, min-count = 1, window-size = 3, step-size = 3 (single codon analysis). Statistical analyses were performed using RStudio version 0.98.1049.

3 Results

To validate SNPGenie’s execution of the Nei-Gojobori (1986) method, we constructed sequences with all 61 non-STOP codons and known numbers of differences. MEGA Version 6 (Tamura et al., 2013) was used to calculate π_N and π_S. SNP reports and GTF files were then constructed to reflect the known variant frequencies and reference sequence, and SNPGenie was used to estimate the same parameters. All results agreed to the last decimal.

Next, both SNPGenie and PoPoolation were used to analyse a pooled H5N1 sample. The nonsynonymous and synonymous mean numbers of pairwise differences per site and numbers of sites (the numerator and denominator of π_N and π_S) were then estimated for the hemagglutinin (HA; Supplementary Tables S1A and S1B) and neuraminidase (NA; Supplementary Tables S2A and S2B) genes.

When PoPoolation estimates were regressed on those from SNPGenie, all R² values were significant (P < 0.001; F-test), but smaller for differences in NA. PoPoolation overestimated differences for HA and underestimated them for NA, while overestimating the number of nonsynonymous sites (Table 1). π was significantly lower in HA (P < 0.01 for π_N; P < 0.001 for π_S; two-sample t-tests), consistent with previous evidence for a population bottleneck upon viral transmission that is driven by selection for specific HA residues (Wilker et al., 2013). Because PoPoolation overestimated differences in HA, this suggests that its false discovery rate may be exacerbated in low-diversity (e.g. bottlenecked) contexts.

Table 1.

Mean nonsynonymous (N) and synonymous (S) differences and sites in hemagglutinin (HA) and neuraminidase (NA) genes of an H5N1 influenza population, estimated by PoPoolation and SNPGenie

Gene	Param.	R2	PoPoolation	SNPGenie	P
HA	N diffs	0.991	0.0039 ± 0.0015	0.0033 ± 0.0016	0.001
	S diffs	0.995	0.0011 ± 0.0006	0.0008 ± 0.0006	0.002
	N sites	0.830	2.3844 ± 0.0144	2.3483 ± 0.0144	<0.001
	S sites	0.830	0.6156 ± 0.0144	0.6517 ± 0.0144	<0.001
NA	N diffs	0.437	0.0015 ± 0.0001	0.0089 ± 0.0007	<0.001
	S diffs	0.231	0.0007 ± 0.0001	0.0029 ± 0.0002	<0.001
	N sites	0.882	2.3667 ± 0.0159	2.3347 ± 0.0158	<0.001
	S sites	0.884	0.6333 ± 0.0159	0.6647 ± 0.0158	<0.001

Values shown are means ± standard errors. P-values refer to a paired t-test comparing PoPoolation and SNPGenie, with the codon as the unit. For all R², P < 0.001 (F test).

Most differences between SNPGenie and PoPoolation can be attributed to: (i) differences in SNP calling; (ii) PoPoolation’s treatment of STOP codon variants as nonsynonymous; and (iii) SNPGenie’s use of allele frequency data in determining the number of sites, contrasted to PoPoolation’s use of the reference sequence alone. PoPoolation also reports π_S = 0 for codons with no synonymous sites, where π_S should be undefined. This could highly inflate the π_N/π_S ratio, overestimating the prevalence of positive natural selection. If the false positive calls are random, ∼75% will be nonsynonymous (Graur and Li, 2000), exacerbating this problem.

Planned future improvements in SNPGenie include additional SNP report formats (e.g. VCF) and weighted mutational pathways.

Funding

This research was supported by NIH grant Al077376> to David H. O’Connor (D.H.O.) and A.L.H.; by NIH grant Al096882> to Christopher Walker and A.L.H; by NIH R01 grant Al084787> to D.H.O. and Thomas C. Friedrich (L.H.M.); by UW-Madison Molecular Biosciences Training Grant T32 GM07215> to L.H.M.; and by an NSF GRF (DGE-0929297>), a University of South Carolina (USC) Presidential Fellowship, and a USC Dept. Biol. Sci. Kathryn Hinnant-Johnson, M.D. Memorial Fellowship to C.W.N.

Conflict of Interest: none declared.