Statistical Genomics Foundry
Paper
domain/molecular-evolutiontopic/gc-biased-gene-conversion

Ratnakumar et al. 2010 — Detecting positive selection within genomes: the problem of biased gene conversion

Quoting mode: own-words (source is © 2010 The Royal Society, free-to-read on PMC, no CC license). §7 is paraphrase, not verbatim prose. Functional strings kept verbatim: the W → S / S → W substitution-class labels, W = A/T (weak), S = G/C (strong), and all numeric values. Full text read via PMC; figures/tables read from caption + text, not from plotted points.

1. Citation

Ratnakumar A, Mousset S, Glémin S, Berglund J, Galtier N, Duret L, Webster MT. 2010. “Detecting positive selection within genomes: the problem of biased gene conversion.” Philosophical Transactions of the Royal Society B: Biological Sciences 365(1552):2571–2580. DOI 10.1098/rstb.2010.0007. PMID 20643747, PMCID PMC2935097. Open access: free-to-read at https://pmc.ncbi.nlm.nih.gov/articles/PMC2935097/ .

2. Access note

Full text read (Abstract, Introduction, Material and Methods, Results, Discussion) via PMC. No paywall boundary. Table/figure numeric values were read from caption + surrounding text (the source’s own reported numbers), cross-checked across multiple reads; per-branch Table 3 values (§6) are the source’s numbers as recovered from the table body and are flagged where confidence is lower.

3. Thesis (1 sentence)

GC-biased gene conversion (gBGC) can accelerate coding-sequence evolution and inflate dN/dS in a way that mimics positive selection, and the authors’ re-analysis of a primate positive-selection scan estimates that more than 20 per cent of significantly elevated dN/dS cases could actually be driven by gBGC rather than adaptation.

4. Problem & context

A common approach to detecting positive selection is genome-wide scanning of alignments to find regions evolving at accelerated rates on a particular branch of a phylogeny (elevated dN/dS = non-synonymous to synonymous substitution rate ratio). But positive selection is not the only process that accelerates evolution. gBGC is a recombination-associated process that biases fixation toward G and C nucleotides; it can generate bursts of substitutions within meiotic recombination hotspots. Such accelerated evolution of non-neutral sequence is typically read as evidence for positive selection in the literature — a very different interpretation. The paper asks how much of a primate positive-selection signal is actually gBGC.

5. Method / approach

  • Dataset: a scan for positive selection on genes across the primate phylogeny (branch tests of dN/dS); PSG = positive-selection-gene candidates identified on particular branches.
  • Substitution-class analysis: substitutions are classified as W → S (weak→strong, A/T→G/C), S → W (strong→weak, G/C→A/T), or GC-conservative. W = A and T (weak); S = G and C (strong).
  • Equilibrium GC content (GC*): inferred from observed W → S and S → W substitution rates; an elevated GC* is the genomic signature of gBGC.
  • Genomic correlates tested: association of accelerated genes with GC*, with distance to the nearest recombination hotspot, and with male vs female recombination rate.
  • Maximum-likelihood (ML) model: genes are modelled as belonging to two classes with different GC* — class 1 (not subject to gBGC) with GC* = g1, and class 2 (subject to gBGC) with GC* = g2 > g1. Model parameters are estimated, then individual genes are assigned to class 1 or class 2 via Bayes’ formula, giving the estimated fraction of PSG candidates that fall in the gBGC-affected class.
  • Case study: the gene ADCYAP1, a previously reported human positive-selection candidate, is examined directly against a gBGC-only explanation.

6. Key claims / findings

  • Genes identified as positive-selection targets have a significant tendency to show the genomic signature of gBGC (elevated GC*).
  • Headline estimate: more than 20 per cent of cases of significantly elevated dN/dS — particularly on shorter branches — could be due to gBGC.
  • Human lineage: more than 20 per cent of PSG candidates identified on the human lineage may be due to gBGC.
  • Per-branch ML breakdown (Table 3, excess of genes assigned to the high-GC*/gBGC class g2 among PSG candidates) — the source’s numbers: Human 28% (1.9 excess); Chimpanzee 19% (1.9 excess); all hominids combined 22% (5.0 excess); all branches together 14% (12.3 excess). The gBGC excess is stronger among PSGs on short phylogenetic branches (22% in all hominids). [The per-branch integer/percentage pairings are recovered from the Table 3 body; the headline “>20 per cent” figures in the prose are the higher-confidence values.]
  • dN/dS magnitude: gBGC can, in some cases, lead to very high dN/dS (more than 2).
  • ADCYAP1 case: since divergence from chimpanzee, two exons accumulated 20 substitutions, all W → S, of which 17 non-synonymous and 3 synonymous; estimated dN/dS = 2.05 in the human lineage. gBGC alone is the most parsimonious hypothesis for this pattern; ADCYAP1’s evolution can be explained by scenarios invoking only gBGC, without adaptive evolution.
  • GC* signal (Table 1, human lineage): accelerated genes have GC* 0.48 vs 0.41 for the full set (elevated equilibrium GC in accelerated genes).
  • Recombination proximity (human lineage): accelerated genes are closer to the nearest recombination hotspot, 44 kb vs 54 kb overall.
  • Sex-specificity: elevated W → S substitution rates associate much more strongly with male than female recombination rate — a pattern selection would not predict (selection is not sex-specific).
  • Why dN/dS rises under gBGC: non-synonymous codon positions generally have lower GC content than synonymous positions (notably in GC-rich genes), so there are more opportunities for W → S substitutions at non-synonymous than at synonymous sites; gBGC therefore increases dN relative to dS.

7. Load-bearing statements (own-words paraphrase; functional strings verbatim)

  • The abstract’s core: using an ML framework the authors estimate that more than 20 per cent of cases of significantly elevated dN/dS, particularly on shorter branches, could be due to gBGC; gBGC can in some cases produce dN/dS more than 2; gBGC significantly affects coding-sequence evolution in primates and often produces patterns that can be mistaken for positive selection. (Abstract, paraphrased.)
  • Mechanism: gBGC is predicted to promote the fixation of slightly deleterious W → S substitutions that natural selection would otherwise discard, so episodes of gBGC can lead to bursts of neutral or deleterious substitutions at functional sites. (Discussion/Results, paraphrased; W → S kept verbatim.)
  • Working hypothesis for the biology: in primates gBGC is thought to arise from a bias in the repair of mismatches in heteroduplex DNA formed during meiotic recombination, favouring incorporation of S nucleotides. (Introduction, paraphrased.)
  • Three distinguishing features (see §9), paraphrased with labels verbatim: (i) gBGC produces W → S biased substitution whereas selection has a priori no general reason to favour W → S (functional regions are not generally GC-rich); (ii) selection acts only on functional sites but gBGC also affects flanking neutral sites; (iii) gBGC is associated with regions of high male recombination (e.g. subtelomeric regions).
  • Problem framing: episodes of accelerated evolution of non-neutral sequence are typically taken in the literature as evidence for positive selection — a very different interpretation from gBGC. (Paraphrased.)

8. Stated scope, assumptions, limitations

  • Scope is primates (the primate phylogeny; human/chimpanzee/hominid branches); the working mechanism (“bias in repair of mismatches in heteroduplex DNA during meiotic recombination”) is stated as the current working hypothesis for primates, drawing on yeast gene-conversion evidence and mismatch-repair assays in primate cell lines.
  • The 20-per-cent-class result is an ML estimate under a two-class GC* model with Bayesian per-gene assignment — a statistical attribution, not a direct demonstration that any individual gene is non-adaptive.
  • The gBGC signal is strongest on short branches; the effect on longer branches is weaker.
  • The paper argues gBGC is a confound for these candidates; it does not claim all positive selection is gBGC.

9. Failure modes / invalidity patterns

The failure mode the paper documents: a dN/dS-based positive-selection scan can misclassify a gBGC-driven acceleration as adaptation. Named symptoms / diagnostics that flag a gBGC-driven (vs selection-driven) acceleration:

  • Directionality: substitutions are biased W → S (not S → W, and not GC-conservative); selection has no general reason to favour W → S.
  • Elevated equilibrium GC (GC*) in the accelerated gene (e.g. 0.48 vs 0.41, human lineage).
  • Effect on flanking neutral sites: gBGC also elevates substitution at neighbouring neutral sites, whereas selection acts only on functional sites.
  • Recombination association: proximity to recombination hotspots (44 kb vs 54 kb) and association with high male recombination regions (e.g. subtelomeric).
  • Sex-specificity: association with male rather than female recombination rate — selection is not sex-specific.
  • Extreme dN/dS: very high dN/dS (more than 2) driven by an all-W → S substitution pattern (as in ADCYAP1: 20 substitutions, all W → S, 17 non-synonymous / 3 synonymous, dN/dS = 2.05).

10. What the source does NOT address (confident silences)

  • No numeric decision threshold is stated for calling a case gBGC — see §11. In particular there is no stated W → S : S → W ratio cutoff and no stated recombination-rate cutoff; the criteria are directional/associational and ML-model-based, not a fixed ratio rule.
  • Scope is primates; the paper does not extend the 20% estimate to non-primate lineages.
  • It does not provide a turnkey software tool/pipeline recommendation for routinely screening scans (it describes the ML two-class GC* model and substitution-class analysis it used).

11. Open questions / ambiguities

  • The headline attribution is “more than 20 per cent” — a lower bound, not a point estimate; the true fraction on any given branch is left as an estimate under the two-class model.
  • Individual-gene assignment is probabilistic (Bayes), so any single PSG’s status (gBGC vs selection) is not resolved with certainty except via case study (ADCYAP1).

12. Guidance answers

  • Mechanism (W → S fixation during meiotic recombination; elevated rate at functional sites without selection): Yes, stated. The working hypothesis is that primate gBGC results from a bias in the repair of mismatches in heteroduplex DNA formed during meiotic recombination, with mismatch repair biased toward incorporating S (G/C) nucleotides; this preferentially fixes W → S alleles. gBGC is predicted to promote the fixation of slightly deleterious W → S substitutions that selection would otherwise discard, so episodes of gBGC lead to bursts of neutral or deleterious substitutions at functional sites. dN/dS specifically rises because non-synonymous positions have lower GC content than synonymous positions, giving more W → S opportunities at non-synonymous sites, raising dN relative to dS. (Paraphrased; own-words mode.)
  • Operational criteria to distinguish gBGC from selection: Yes — three main features, enumerated (labels verbatim): (i) gBGC generates W → S biased substitution patterns whereas selection has a priori no general reason to favour W → S (functional regions are not generally GC-rich); (ii) selection operates only on functional sites, but gBGC also affects flanking neutral sites; (iii) gBGC is associated with regions of high male recombination (such as subtelomeric regions). Supporting signals used: elevated equilibrium GC (GC*), proximity to recombination hotspots, and the male-vs-female recombination-rate asymmetry (selection is not sex-specific).
  • Numeric threshold (W → S : S → W ratio or recombination cutoff): No numeric threshold stated. The paper does not give a W → S : S → W ratio cutoff or a recombination-rate cutoff; it distinguishes gBGC by directionality + associations + the ML two-class GC* model, not a fixed ratio. (Confirms the guidance expectation that the “W → S ratio > 1.5” figure is a convention, not this paper’s number.)
  • Magnitude attributed to gBGC (paper’s own re-analysis): The paper’s own re-analysis of the primate PSG scan estimates that more than 20 per cent of significantly elevated dN/dS cases could be due to gBGC, particularly on shorter branches; likewise more than 20 per cent of PSG candidates on the human lineage may be due to gBGC. Table 3 per-branch excess in the high-GC*/gBGC class: Human 28%, Chimpanzee 19%, all hominids 22%, all branches 14%. gBGC can drive dN/dS more than 2 (ADCYAP1: dN/dS = 2.05, human lineage, 20 substitutions all W → S). These are the paper’s own re-analysis numbers, not a general field statistic.
  • gBGC as an extended NULL to exclude before inferring adaptation: The paper does not use the explicit words “null hypothesis to exclude.” Its framing is that accelerated evolution of non-neutral sequence is typically read as evidence for positive selection, which is a very different interpretation from gBGC, and that for ADCYAP1 “gBGC alone remains the most parsimonious hypothesis” — i.e. gBGC can fully explain a signal without adaptive evolution. So the paper strongly implies gBGC must be ruled out before inferring adaptation, but does not state it in explicit null-hypothesis language. [summarizer-inferred: the “extended null” phrasing is not the paper’s; the paper’s own framing is parsimony + mistaken-interpretation.]