Sexual dichromatism across primates

What a pigmentation-network approach reveals when pointed at evolution: a trait that maps to genes of major effect in birds is a coupled, polygenic, multi-module system in primates

Using the coupled pigmentation–sex-hormone gene network to test selection across the independent origins of primate hair sexual dichromatism — an evolutionary-genomics insight from the same network approach used for human association grading.

Terms used on this page, in plain language

So this page stands on its own — you should not need a genetics background to follow it.

Sexual dichromatism — the males and females of a species differ in color (here, in hair color).
Independent origin — a separate evolutionary appearance of the trait. If two species are both dichromatic but their shared ancestor was not, dichromatism arose twice, not once. Counting origins (rather than counting dichromatic species) is the central move on this page.
Polygenic — built by many genes acting together, rather than switched on or off by one gene.
Labile — evolutionarily unstable: readily gained and readily lost over time.
Ancestral-state reconstruction — working backward from the living species to infer whether each ancestor on the family tree had the trait, which is how we locate where it was gained or lost.
dN/dS (ω) — a number that measures the kind of natural selection acting on a gene’s protein. ω ≈ 1 means no net selection, ω < 1 means the protein is being conserved, ω > 1 means it is being actively changed (positive selection). “Episodic” selection means ω spikes above 1 on particular branches of the tree rather than everywhere.
RELAX, aBSREL, RERconverge — three standard statistical tests (from the HyPhy and RERconverge toolkits) that each ask, in a different way, whether specific genes evolved faster on the branches where dichromatism arose. Using three lets us cover origins with different amounts of data.

Point the network at evolution

The rest of this project builds a curated pigmentation gene network and uses it to grade human genotype→phenotype claims. This page asks a different question with the same object: what does a pigmentation-network approach reveal when you point it at evolutionary genomics instead of human association? The answer is the finding below — and it only becomes visible because we treat the genes as a coupled network rather than a flat candidate list.

Two things fall out of the network framing that a single-gene scan would miss:

The pigmentation and sex-hormone networks are not independent lists — they are physically coupled. The clearest bridge is POMC: the same pro-hormone is cleaved into α-MSH, which drives pigmentation through the melanocortin receptor, and ACTH, which drives the adrenal steroid axis. A gene sitting on both networks is exactly where a sexually dimorphic, hormone-linked color trait would be expected to act. And POMC is under selection: it is significantly intensified (HyPhy RELAX, K ≈ 3.4, p_BH < 0.001) at the Nomascus gibbon origin, with episodic diversifying selection (aBSREL) on five branches. But that signal is not confined to dichromatic lineages — only two of the five selected branches are dichromatic (the Nomascus gibbons); the other three are monochromatic macaque lineages. So the bridge gene overlaps one dichromatism origin without tracking the trait one-to-one — the same lesson as the whole-panel scan below. See §5 of the dichromatism-synthesis notebook.
Selection localizes to modules, and modules to branches. The signal that passes the significance and alignment-QC gate is not scattered noise: on the pigmentation side it concentrates in the melanoblast-development module (KITLG–EDN3–TFAP2A); on the hormone side in steroid-metabolism enzymes (HSD17B1/12/7, SRD5A1, CYP7B1) — not the receptors. The open question the per-origin design tests directly is whether the pigmentation module is remodeled in some origins while the hormone module is remodeled in others — i.e. different lineages reach the same phenotype through different parts of the coupled network.

That second point is the evolutionary-genomics insight the network approach yields: dichromatism looks less like “a gene” and more like a coupled pigmentation–hormone system with several accessible failure/remodeling points, which is why it can be gained and lost so readily.

The one-sentence result

In birds, the best-studied cases of sexual dichromatism map to genes of major effect for a specific pigment system in a specific clade: MC1R for melanin-based dichromatism in galliforms (Nadeau et al. 2007 report a dN/dS–dichromatism correlation at that one locus), and BCO2 for carotenoid-based dichromatism in finches (Gazda et al. 2020, Science) — and these are hormonally regulated (there is an active estrogen/testosterone-control literature in birds, Griffith et al. 2026, Integr Comp Biol). In primates it is not. Hair dichromatism has arisen ~15 times independently across the primate radiation and is polygenic in every origin (and MC1R is not the hit). The trait is also evolutionarily labile, lost roughly 9× more readily than it is gained, consistent with dichromatism being reachable through several alternative genetic routes rather than one canonical switch. (We tested the coupling directly: the pigmentation and sex-hormone modules co-evolve with each other — concordant per-lineage selection, PGLS β ≈ 0.42, p ≈ 1e-4 — but that coupled selection does not track the trait: a phylogenetic correlated-evolution test (Pagel’s Discrete, via phytools::fitPagel) prefers the independent model for both modules — pigmentation ΔAIC = −3.69, p = 0.37; hormone ΔAIC = −5.71, p = 0.68 — over one in which module selection and dichromatism co-vary. The network is coupled to itself, not to the phenotype.) Whether the same or different parts of that network are recruited across origins — a shared versus heterogeneous architecture — is not yet resolvable with the current data: only 3 origins are powered for the selection test, and a homogeneity test across the 11 origins with any detectable selection does not reject a single shared architecture (χ² p = 0.42). We therefore report the polygenic, labile, non-MC1R result as the finding, and leave shared-vs-heterogeneous as an explicitly open question (see Limitations).

Why this is novel

This is the first clade-wide, multi-gene, selection-inference test of the genomic basis of sexual dichromatism in primates. Prior primate work is single-gene (MC1R) and framed as coat color, not sexual dichromatism; no published study or 2023–2026 preprint tests it as a network-level, per-origin question. The contribution is a shift in the unit of analysis — from “dichromatic species” to “independent origin of dichromatism” — and the finding that the trait is polygenic and non-MC1R in every origin tested. (Whether the architecture is shared or heterogeneous across origins is set up by this design but is not yet powered to resolve; see Limitations.)

Dichromatism is not one event — it is ~15

Sexual dichromatism is not a single event to be pooled across species: pooling every dichromatic species into one foreground and asking “does selection intensify in dichromatic lineages” presupposes a shared, convergent architecture — the very thing in question. Reconstructing the trait’s history from the raw 238-species trait table and tree, by ancestral-state reconstruction, shows why the independent origin, not the species, is the right unit:

Hair dichromatism arose ~15 times independently across the primate tree

The 24 genomically-sampled dichromatic species collapse to only 14 independent origins (≈15 across the full tree; one, Presbytis hosei, has no genome). Of the 14:

11 are single genomic tips — each a genuinely separate evolutionary gain.
3 are multi-tip radiations descended from one ancestral gain each: Trachypithecus (8 tips), Nomascus (3), Eulemur macaco/flavifrons (2).

So 10 of the 24 pooled tips (42%) are within-clade replicates of just 3 events, and Trachypithecus alone carried a third of the foreground weight. A pooled test reporting n = 24 was effectively an n = 14 analysis dominated by one clade.

Selection is scattered, not concentrated

With aBSREL fitting a dN/dS to every branch, the pattern on TFAP2A — one of the pooled “certified” pigmentation hits — is telling: episodic selection appears on 4 of 21 dichromatic tips and 8 of 67 monochromatic tips. It is scattered across the tree, not concentrated on dichromatic lineages. TFAP2A is a widely re-used gene under episodic selection in many primates; being dichromatic does not predict selection on it.

TFAP2A dN/dS across the primate tree — the bird figure’s grammar, inverted. Branches painted by dN/dS (ω = 1 neutral at the color center), dichromatic tips marked red, episodic selection starred. In galliforms (Nadeau 2007) one gene, MC1R, showed elevated dN/dS specifically in the dichromatic clades; here TFAP2A is conserved almost everywhere and the selected branches (★) are scattered across dichromatic and monochromatic lineages alike.

Compare Nadeau et al. (2007), Fig. 1 — the same tree-with-dN/dS-and-dichromatism layout for galliform birds, where the signal does track the dichromatic clades because it is essentially one gene.

The companion strip plot below separates the same result by dichromatic vs. monochromatic status, on a −log₁₀(corrected p) axis:

TFAP2A: episodic selection is not concentrated on dichromatic branches

Two branches showed an undefined ω from near-zero synonymous change (Macaca tonkeana, Mandrillus sphinx) — flagged as artifacts, not read as extreme positive selection. This is the standard aBSREL failure mode: infinite ω on a branch with dS ≈ 0 is undefined, not signal.

The three-layer test, matched to the origin structure

Because only 3 origins have ≥2 tips, no single method covers all of them. The design uses two executed selection layers — matched to what the data can support — plus a third, relative-rate layer that is set up but not yet run. The unifying question each layer asks is the coupled-network one: for a given origin, is it the pigmentation module or the hormone module that carries the selection signal — and does that differ from origin to origin?

Layer	Method	Covers	Status	Question
Per-origin	HyPhy RELAX	the 3 multi-tip origins	run	for Trachypithecus, Nomascus, Eulemur — is the intensified module the pigmentation axis, the hormone axis, or both, and is it the same module across origins (convergence) or different (heterogeneous, branch-specific)?
Per-branch	HyPhy aBSREL	all 14 origins (each single-tip origin = one branch)	run	which individual lineages carry episodic selection, at which genes?
Relative-rate	RERconverge	all origins at once	planned, not yet run	across the tree, do the origins accelerate the same genes — no per-origin foreground needed?

The two executed layers (RELAX, aBSREL) are what the results on this page rest on. The codon alignments for RERconverge are staged (comparative-genomics/results/full_panel_117/), but the relative-rate scan has not been run and is not reported as a result here.

The cross-taxon contrast

	birds	primates
genetic basis	genes of major effect, clade- and pigment-specific — MC1R/melanin in galliforms (Nadeau 2007), BCO2/carotenoid in finches (Gazda 2020)	polygenic; no single gene of major effect (MC1R is not the hit)
pigmentation signal sits at	a named large-effect locus per clade	melanoblast development / signaling (KITLG–EDN3–TFAP2A) — no single dominant node
architecture across origins	not systematically tested clade-wide	polygenic; shared-vs-heterogeneous underpowered to resolve (χ² p = 0.42)
hormone coupling	present — dichromatism genes hormonally regulated (Griffith et al. 2026)	the two modules co-evolve with each other (concordant selection, tested) but that coupling does not track the trait (correlated-evolution test: independent model preferred)
trait stability	—	labile; loss ≈ 9× faster than gain

How the study was built

This is the shape of the analysis in one place; the detailed pipeline and its limits follow below.

The question. Does the genetic basis of primate sexual dichromatism rest on a gene of major effect, as the best-studied bird cases do — MC1R/melanin in galliforms (Nadeau et al. 2007) and BCO2/carotenoid in finches (Gazda et al. 2020)?

The data and tools. A 238-species primate hair-trait table and a dated primate tree; a curated selection panel of 80 genes — 27 pigmentation and 53 sex-hormone (expanded to ~105 for the per-origin and cross-species-GWAS layers after the pigmentation module was widened); and ~117 primate genome assemblies. Trait history comes from ancestral-state reconstruction (ape/phytools, ARD/ER Mk models, stochastic mapping); coding sequences from homology-based extraction (miniprot), codon alignment from MAFFT, and selection inference from the HyPhy suite (RELAX per origin, aBSREL per branch) — run on an HPC cluster from a committed, parameterized protocol. A relative-rate layer (RERconverge) is designed and its alignments are staged, but it has not yet been run and is not among the results reported here.

The finding. Primate sexual dichromatism is polyphyletic (~15 origins), labile (loss ≈ 9× gain), and polygenic with no single shared signature — unlike the avian genes-of-major-effect model. A phylogenetic correlated-evolution test finds selection on the pigment and hormone modules concordant with each other but not with the trait, and a phylogeny-controlled association scan flags one suggestive candidate (AKR1C4, hormone module — see Limitations) rather than an established locus.

The design choice that makes it work. Taking the independent origin of dichromatism, not the dichromatic species, as the unit of selection testing — the only unit at which a test can ask whether the architecture is shared across origins or heterogeneous.

Limitations

Per-origin RELAX is statistically powered only for the 3 multi-tip origins; the 11 single-tip origins are covered by aBSREL (per branch). A relative-rate layer (RERconverge) is designed and its alignments are staged but it has not yet been run, so single-tip origins currently rest on aBSREL alone rather than being treated as missing data.
Two phenotypic origins (Presbytis hosei, Hoolock hoolock) have no genome in the 117-species set, so they contribute to the origin count from the phenotype tree but not to the selection scans. A targeted genome-expansion round (a Presbytis proxy plus a Hoolock genome) was scoped for exactly this gap but has not been run; the selection results here are on the 117-genome set as it stands.
The pigmentation-vs-hormone set-level contrast is not significant (permutation p ≈ 0.87) and is not claimed. The apparent hormone-tilt in the pooled selected set (≈59% hormone) is consistent with the panel’s own 2:1 hormone:pigmentation gene ratio (binomial p ≈ 0.17), not with preferential selection on either module.
A phylogeny-controlled cross-species association scan finds one suggestive gene. Treating the origins as a cross-species GWAS — dichromatic/monochromatic as case/control, per-lineage dN/dS as the per-gene marker, and phylogenetic covariance (per-gene Pagel’s λ) as the stratification control — the strongest association is AKR1C4, a sex-hormone-module steroid reductase, with elevated dN/dS on dichromatic lineages. It is consistent in direction across three tests (PGLS, canonical binary phyloglm, and a tree-structured permutation, perm p ≈ 5e-4), but its survival of multiple-testing correction is model-dependent: it clears BH under PGLS (p_BH ≈ 0.046) but not under the more appropriate binary model, phyloglm (p_BH = 1.0). Since PGLS is an anticonservative linear-probability model, this is a single suggestive candidate, not an established locus: one hit among ~100 genes with only ~24 cases, and the binding constraint is the number of independent dichromatic lineages, which no method can inflate. See the cross-species GWAS notebook.
Shared vs. heterogeneous architecture is currently underpowered to resolve. A χ² homogeneity test across the 11 origins with any detectable selection does not reject a single shared architecture (p ≈ 0.42; Monte-Carlo p ≈ 0.45), and the two “pure-module” poles (Eulemur, Pithecia) are each defined by only 1–2 selected genes — a small-sample funnel effect, not a measured contrast. Gene count per origin is ≈92% explained by foreground-tip count. The established finding is that the trait is polygenic and non-MC1R; “different genes in different lineages” is a hypothesis this data cannot yet confirm or reject. A rank-based cross-lineage test (e.g. PicMin) on the 14 origins, and a network-subgraph test (e.g. signet), are the powered next steps — see internal/lit_review/phylo-grn/.

Reproducibility

The two results notebooks are the executable version of this page, each self-contained and rendered from stored outputs:

Sexual dichromatism synthesis recomputes the lability and origin-count analyses from the tree, and reports the per-origin module-balance, circular-tree, and sister-pair contrasts.
Cross-species GWAS runs the phylogeny-controlled association scan (PGLS + phyloglm) and produces the AKR1C4 result.
Correlated-evolution test computes the Pagel’s-λ, transition-rate, and fitPagel results above from frozen inputs in its own data/ directory (no network, no cluster). quarto render coevolution_test.qmd reproduces every number and the figure; the notebook ends with an assertion block that fails if any value drifts from the committed ones (verified: λ = 0.6547, loss ≈ 9.1× gain, fitPagel independent model preferred for both modules).

The full pipeline (miniprot → MAFFT → RELAX / aBSREL; RERconverge staged but not yet run), the 14-origin foreground map, the scoped genome-expansion roster, and the exact cluster run protocol are in comparative-genomics/ and the handoff notes under internal/handoffs/notes/. The selection run executes on the University of Michigan Great Lakes cluster; only small summary tables are committed and pasted back — genomes and alignments stay on scratch.

--- title: "Sexual dichromatism across primates" subtitle: "What a pigmentation-network approach reveals when pointed at evolution: a trait that maps to genes of major effect in birds is a coupled, polygenic, multi-module system in primates" description: "Using the coupled pigmentation–sex-hormone gene network to test selection across the independent origins of primate hair sexual dichromatism — an evolutionary-genomics insight from the same network approach used for human association grading." toc: true execute: enabled: false --- ::: {.callout-note} ## Terms used on this page, in plain language So this page stands on its own — you should not need a genetics background to follow it. - **Sexual dichromatism** — the males and females of a species differ in color (here, in hair color). - **Independent origin** — a separate evolutionary appearance of the trait. If two species are both dichromatic but their shared ancestor was not, dichromatism arose *twice*, not once. Counting origins (rather than counting dichromatic species) is the central move on this page. - **Polygenic** — built by many genes acting together, rather than switched on or off by one gene. - **Labile** — evolutionarily unstable: readily gained *and* readily lost over time. - **Ancestral-state reconstruction** — working backward from the living species to infer whether each ancestor on the family tree had the trait, which is how we locate where it was gained or lost. - **dN/dS (ω)** — a number that measures the kind of natural selection acting on a gene's protein. ω ≈ 1 means no net selection, ω < 1 means the protein is being conserved, ω > 1 means it is being actively changed (positive selection). "Episodic" selection means ω spikes above 1 on particular branches of the tree rather than everywhere. - **RELAX, aBSREL, RERconverge** — three standard statistical tests (from the HyPhy and RERconverge toolkits) that each ask, in a different way, whether specific genes evolved faster on the branches where dichromatism arose. Using three lets us cover origins with different amounts of data. ::: ## Point the network at evolution The rest of this project builds a curated pigmentation gene network and uses it to grade *human* genotype→phenotype claims. This page asks a different question with the **same object**: *what does a pigmentation-network approach reveal when you point it at evolutionary genomics instead of human association?* The answer is the finding below — and it only becomes visible **because** we treat the genes as a coupled network rather than a flat candidate list. Two things fall out of the network framing that a single-gene scan would miss: - **The pigmentation and sex-hormone networks are not independent lists — they are physically coupled.** The clearest bridge is **POMC**: the same pro-hormone is cleaved into α-MSH, which drives pigmentation through the melanocortin receptor, **and** ACTH, which drives the adrenal steroid axis. A gene sitting on both networks is exactly where a sexually dimorphic, hormone-linked color trait would be expected to act. And POMC *is* under selection: it is significantly intensified (HyPhy RELAX, *K* ≈ 3.4, *p*~BH~ < 0.001) at the *Nomascus* gibbon origin, with episodic diversifying selection (aBSREL) on five branches. But that signal is **not confined to dichromatic lineages** — only two of the five selected branches are dichromatic (the *Nomascus* gibbons); the other three are monochromatic macaque lineages. So the bridge gene overlaps one dichromatism origin without tracking the trait one-to-one — the same lesson as the whole-panel scan below. See §5 of the [dichromatism-synthesis notebook](notebooks/15_dichromatism_synthesis.ipynb). - **Selection localizes to modules, and modules to branches.** The signal that passes the significance and alignment-QC gate is not scattered noise: on the pigmentation side it concentrates in the melanoblast-development module (KITLG–EDN3–TFAP2A); on the hormone side in steroid-metabolism enzymes (HSD17B1/12/7, SRD5A1, CYP7B1) — not the receptors. The open question the per-origin design tests directly is whether **the pigmentation module is remodeled in some origins while the hormone module is remodeled in others** — i.e. different lineages reach the same phenotype through different parts of the coupled network. That second point is the evolutionary-genomics insight the network approach yields: dichromatism looks less like "a gene" and more like *a coupled pigmentation–hormone system with several accessible failure/remodeling points*, which is why it can be gained and lost so readily. ## The one-sentence result In birds, the best-studied cases of sexual dichromatism map to **genes of major effect** for a specific pigment system in a specific clade: *MC1R* for melanin-based dichromatism in galliforms ([Nadeau et al. 2007](https://doi.org/10.1098/rspb.2007.0174) report a dN/dS–dichromatism correlation at that one locus), and *BCO2* for carotenoid-based dichromatism in finches ([Gazda et al. 2020, *Science*](https://doi.org/10.1126/science.aba0803)) — and these are hormonally regulated (there is an active estrogen/testosterone-control literature in birds, [Griffith et al. 2026, *Integr Comp Biol*](https://doi.org/10.1093/icb/icag079)). **In primates it is not.** Hair dichromatism has arisen **~15 times independently** across the primate radiation and is **polygenic in every origin** (and *MC1R* is not the hit). The trait is also evolutionarily labile, lost roughly **9× more readily than it is gained**, consistent with dichromatism being reachable through several alternative genetic routes rather than one canonical switch. (We tested the coupling directly: the **pigmentation and sex-hormone modules co-evolve with each other** — concordant per-lineage selection, PGLS β ≈ 0.42, *p* ≈ 1e-4 — but that coupled selection **does not track the trait**: a phylogenetic correlated-evolution test ([Pagel's *Discrete*, via `phytools::fitPagel`](https://github.com/tinalasisi/pigmentation-gene-network/blob/main/comparative-genomics/analysis/coevolution_test/coevolution_test.qmd)) prefers the independent model for both modules — pigmentation ΔAIC = −3.69, *p* = 0.37; hormone ΔAIC = −5.71, *p* = 0.68 — over one in which module selection and dichromatism co-vary. The network is coupled to itself, not to the phenotype.) Whether the *same* or *different* parts of that network are recruited across origins — a shared versus heterogeneous architecture — is **not yet resolvable with the current data**: only 3 origins are powered for the selection test, and a homogeneity test across the 11 origins with any detectable selection does not reject a single shared architecture (χ² p = 0.42). We therefore report the polygenic, labile, non-*MC1R* result as the finding, and leave shared-vs-heterogeneous as an explicitly open question (see *Limitations*). ::: {.callout-note} ## Why this is novel This is the first clade-wide, multi-gene, selection-inference test of the genomic basis of sexual dichromatism in primates. Prior primate work is single-gene (MC1R) and framed as coat color, not sexual dichromatism; no published study or 2023–2026 preprint tests it as a network-level, per-origin question. The contribution is a **shift in the unit of analysis** — from "dichromatic species" to "independent origin of dichromatism" — and the finding that the trait is polygenic and non-*MC1R* in every origin tested. (Whether the architecture is *shared* or *heterogeneous* across origins is set up by this design but is not yet powered to resolve; see *Limitations*.) ::: ## Dichromatism is not one event — it is ~15 Sexual dichromatism is not a single event to be pooled across species: pooling every dichromatic species into one foreground and asking "does selection intensify in dichromatic lineages" presupposes a shared, convergent architecture — the very thing in question. Reconstructing the trait's history from the raw 238-species trait table and tree, by ancestral-state reconstruction, shows why the independent origin, not the species, is the right unit: ![Hair dichromatism arose ~15 times independently across the primate tree](comparative-genomics/results/figures/fig_origins.png) The 24 genomically-sampled dichromatic species collapse to only **14 independent origins** (≈15 across the full tree; one, *Presbytis hosei*, has no genome). Of the 14: - **11 are single genomic tips** — each a genuinely separate evolutionary gain. - **3 are multi-tip radiations** descended from one ancestral gain each: *Trachypithecus* (8 tips), *Nomascus* (3), *Eulemur macaco/flavifrons* (2). So **10 of the 24 pooled tips (42%) are within-clade replicates of just 3 events**, and *Trachypithecus* alone carried a third of the foreground weight. A pooled test reporting *n* = 24 was effectively an *n* = 14 analysis dominated by one clade. ![The pooling problem, quantified](comparative-genomics/results/figures/fig_pooling.png) ## Selection is scattered, not concentrated With aBSREL fitting a dN/dS to every branch, the pattern on *TFAP2A* — one of the pooled "certified" pigmentation hits — is telling: episodic selection appears on **4 of 21 dichromatic tips and 8 of 67 monochromatic tips**. It is scattered across the tree, not concentrated on dichromatic lineages. *TFAP2A* is a widely re-used gene under episodic selection in many primates; being dichromatic does not predict selection on it. ![TFAP2A dN/dS across the primate tree — the bird figure's grammar, inverted. Branches painted by dN/dS (ω = 1 neutral at the color center), dichromatic tips marked red, episodic selection starred. In galliforms (Nadeau 2007) one gene, MC1R, showed elevated dN/dS specifically in the dichromatic clades; here TFAP2A is conserved almost everywhere and the selected branches (★) are scattered across dichromatic and monochromatic lineages alike.](comparative-genomics/results/figures/fig_tfap2a_tree.png) ::: {.column-margin} Compare Nadeau et al. (2007), Fig. 1 — the same tree-with-dN/dS-and-dichromatism layout for galliform birds, where the signal *does* track the dichromatic clades because it is essentially one gene. ::: The companion strip plot below separates the same result by dichromatic vs. monochromatic status, on a −log₁₀(corrected p) axis: ![TFAP2A: episodic selection is not concentrated on dichromatic branches](comparative-genomics/results/figures/fig_tfap2a_branches.png) ::: {.callout-tip appearance="simple"} Two branches showed an *undefined* ω from near-zero synonymous change (*Macaca tonkeana*, *Mandrillus sphinx*) — flagged as artifacts, **not** read as extreme positive selection. This is the standard aBSREL failure mode: infinite ω on a branch with dS ≈ 0 is undefined, not signal. ::: ## The three-layer test, matched to the origin structure Because only 3 origins have ≥2 tips, no single method covers all of them. The design uses two executed selection layers — matched to what the data can support — plus a third, relative-rate layer that is set up but not yet run. The unifying question each layer asks is the coupled-network one: **for a given origin, is it the pigmentation module or the hormone module that carries the selection signal** — and does that differ from origin to origin? | Layer | Method | Covers | Status | Question | |---|---|---|---|---| | Per-origin | HyPhy **RELAX** | the 3 multi-tip origins | **run** | for *Trachypithecus*, *Nomascus*, *Eulemur* — is the intensified module the pigmentation axis, the hormone axis, or both, and is it the **same** module across origins (convergence) or **different** (heterogeneous, branch-specific)? | | Per-branch | HyPhy **aBSREL** | all 14 origins (each single-tip origin = one branch) | **run** | which individual lineages carry episodic selection, at which genes? | | Relative-rate | **RERconverge** | all origins at once | *planned, not yet run* | across the tree, do the origins accelerate the same genes — no per-origin foreground needed? | The two executed layers (RELAX, aBSREL) are what the results on this page rest on. The codon alignments for RERconverge are staged (`comparative-genomics/results/full_panel_117/`), but the relative-rate scan has not been run and is not reported as a result here. ## The cross-taxon contrast | | birds | primates | |---|---|---| | genetic basis | genes of major effect, clade- and pigment-specific — *MC1R*/melanin in galliforms (Nadeau 2007), *BCO2*/carotenoid in finches (Gazda 2020) | polygenic; no single gene of major effect (*MC1R* is **not** the hit) | | pigmentation signal sits at | a named large-effect locus per clade | melanoblast development / signaling (KITLG–EDN3–TFAP2A) — no single dominant node | | architecture across origins | not systematically tested clade-wide | polygenic; shared-vs-heterogeneous **underpowered to resolve** (χ² p = 0.42) | | hormone coupling | present — dichromatism genes hormonally regulated (Griffith et al. 2026) | the two modules co-evolve with each other (concordant selection, tested) but that coupling does not track the trait (correlated-evolution test: independent model preferred) | | trait stability | — | labile; loss ≈ 9× faster than gain | ## How the study was built This is the shape of the analysis in one place; the detailed pipeline and its limits follow below. **The question.** Does the genetic basis of primate sexual dichromatism rest on a gene of major effect, as the best-studied bird cases do — *MC1R*/melanin in galliforms ([Nadeau et al. 2007](https://doi.org/10.1098/rspb.2007.0174)) and *BCO2*/carotenoid in finches ([Gazda et al. 2020](https://doi.org/10.1126/science.aba0803))? **The data and tools.** A 238-species primate hair-trait table and a dated primate tree; a curated selection panel of **80 genes — 27 pigmentation and 53 sex-hormone** (expanded to ~105 for the per-origin and cross-species-GWAS layers after the pigmentation module was widened); and ~117 primate genome assemblies. Trait history comes from ancestral-state reconstruction (`ape`/`phytools`, ARD/ER Mk models, stochastic mapping); coding sequences from homology-based extraction (miniprot), codon alignment from MAFFT, and selection inference from the HyPhy suite (RELAX per origin, aBSREL per branch) — run on an HPC cluster from a committed, parameterized protocol. A relative-rate layer (RERconverge) is designed and its alignments are staged, but it has not yet been run and is not among the results reported here. **The finding.** Primate sexual dichromatism is polyphyletic (~15 origins), labile (loss ≈ 9× gain), and polygenic with no single shared signature — unlike the avian genes-of-major-effect model. A phylogenetic correlated-evolution test finds selection on the pigment and hormone modules concordant with *each other* but not with the trait, and a phylogeny-controlled association scan flags one suggestive candidate (*AKR1C4*, hormone module — see Limitations) rather than an established locus. **The design choice that makes it work.** Taking the independent *origin* of dichromatism, not the dichromatic *species*, as the unit of selection testing — the only unit at which a test can ask whether the architecture is shared across origins or heterogeneous. ## Limitations - Per-origin RELAX is statistically powered only for the 3 multi-tip origins; the 11 single-tip origins are covered by aBSREL (per branch). A relative-rate layer (RERconverge) is designed and its alignments are staged but it has not yet been run, so single-tip origins currently rest on aBSREL alone rather than being treated as missing data. - Two phenotypic origins (*Presbytis hosei*, *Hoolock hoolock*) have no genome in the 117-species set, so they contribute to the origin count from the phenotype tree but not to the selection scans. A targeted genome-expansion round (a *Presbytis* proxy plus a *Hoolock* genome) was scoped for exactly this gap but has not been run; the selection results here are on the 117-genome set as it stands. - The pigmentation-vs-hormone **set-level** contrast is not significant (permutation *p* ≈ 0.87) and is **not** claimed. The apparent hormone-tilt in the pooled selected set (≈59% hormone) is consistent with the panel's own 2:1 hormone:pigmentation gene ratio (binomial *p* ≈ 0.17), not with preferential selection on either module. - **A phylogeny-controlled cross-species association scan finds one suggestive gene.** Treating the origins as a cross-species GWAS — dichromatic/monochromatic as case/control, per-lineage dN/dS as the per-gene marker, and phylogenetic covariance (per-gene Pagel's λ) as the stratification control — the strongest association is *AKR1C4*, a sex-hormone-module steroid reductase, with elevated dN/dS on dichromatic lineages. It is consistent in **direction** across three tests (PGLS, canonical binary phyloglm, and a tree-structured permutation, perm *p* ≈ 5e-4), but its survival of multiple-testing correction is **model-dependent**: it clears BH under PGLS (*p*~BH~ ≈ 0.046) but not under the more appropriate binary model, phyloglm (*p*~BH~ = 1.0). Since PGLS is an anticonservative linear-probability model, this is a **single suggestive candidate**, not an established locus: one hit among ~100 genes with only ~24 cases, and the binding constraint is the number of independent dichromatic lineages, which no method can inflate. See the [cross-species GWAS notebook](notebooks/16_cross_species_gwas.ipynb). - **Shared vs. heterogeneous architecture is currently underpowered to resolve.** A χ² homogeneity test across the 11 origins with any detectable selection does not reject a single shared architecture (*p* ≈ 0.42; Monte-Carlo *p* ≈ 0.45), and the two "pure-module" poles (*Eulemur*, *Pithecia*) are each defined by only 1–2 selected genes — a small-sample funnel effect, not a measured contrast. Gene count per origin is ≈92% explained by foreground-tip count. The established finding is that the trait is **polygenic and non-*MC1R***; "different genes in different lineages" is a hypothesis this data cannot yet confirm or reject. A rank-based cross-lineage test (e.g. PicMin) on the 14 origins, and a network-subgraph test (e.g. signet), are the powered next steps — see `internal/lit_review/phylo-grn/`. ## Reproducibility The two results notebooks are the executable version of this page, each self-contained and rendered from stored outputs: - **[Sexual dichromatism synthesis](notebooks/15_dichromatism_synthesis.ipynb)** recomputes the lability and origin-count analyses from the tree, and reports the per-origin module-balance, circular-tree, and sister-pair contrasts. - **[Cross-species GWAS](notebooks/16_cross_species_gwas.ipynb)** runs the phylogeny-controlled association scan (PGLS + phyloglm) and produces the *AKR1C4* result. - **[Correlated-evolution test](https://github.com/tinalasisi/pigmentation-gene-network/tree/main/comparative-genomics/analysis/coevolution_test)** computes the Pagel's-λ, transition-rate, and `fitPagel` results above from frozen inputs in its own `data/` directory (no network, no cluster). `quarto render coevolution_test.qmd` reproduces every number and the figure; the notebook ends with an assertion block that fails if any value drifts from the committed ones (verified: λ = 0.6547, loss ≈ 9.1× gain, fitPagel independent model preferred for both modules). The full pipeline (miniprot → MAFFT → RELAX / aBSREL; RERconverge staged but not yet run), the 14-origin foreground map, the scoped genome-expansion roster, and the exact cluster run protocol are in [`comparative-genomics/`](https://github.com/tinalasisi/pigmentation-gene-network/tree/main/comparative-genomics) and the handoff notes under `internal/handoffs/notes/`. The selection run executes on the University of Michigan Great Lakes cluster; only small summary tables are committed and pasted back — genomes and alignments stay on scratch.