Dun

Crossing two of them

What it looks like on each base

Its gene carrot

Gene key

horsegenetics.dun

Alleles

D d1 d2

Outcomes

wild, primitive-marks, dun

Combinations

6, landing on 3 outcomes

Default allele

d2

Wild frequency

D 1 in 24, d1 1 in 10 per allele

Founder draws

1 × nextFloat()

Deterministic

no — the markings are rolled per horse

Epigenetic draws

1 × nextLong() + 7 × nextFloat(), on both marked outcomes

Dilution

PigmentField.diluteNeutral — red first, then black

keep red / black (D)

0.93 of the visible red; 0.42 of the black

keep red / black (d1)

0.95 of the visible red; black never touched

dorsal half-width

1.5 body units, ×0.80–1.30 per horse

leg bars

joint 0.56, spread 0.48, spacing 3.2, duty 0.42 — D only, per leg

accessory chances

bars 0.82 per leg, shoulder bar 0.55, face 0.40

Two dominance orders over one locus

This is the gene that most obviously needs a table rather than a label. Dilution reads D > d1 = d2; marking reads D = d1 > d2. Two orders over the same three alleles, and no single word covers both — which is exactly why there is no dominance property in this model, only an outcome per combination.

So d2 is the only allele that draws nothing, and it — not the old catch-all d — is the gene’s default allele: the token a code with no dun segment reads as. A horse can carry a dorsal stripe and not be a dun, which is what d1 is for.

One dilution and one mask

Phase 1 is downward-only — a natural gene can never add pigment back — so nothing here can literally paint a dark line onto a coat. It does not need to. A primitive marking is countershading: “a stripe darker than the body” and “a body lighter than the stripe” are the same picture. So the whole gene is one dilution and one mask. The mask says, per texel, how much of the dilution to lerp back off, and every dark thing a dun has is a region where that number is high.

double mark = CoatRegions.dorsalStripe(skin, part, point, m.dorsalHalfWidth());
mark = Math.max(mark, pointRegion(skin, part, point));            // midtstol, ears, muzzle, lower leg
mark = Math.max(mark, alreadyAPoint(f.red(px, py), f.black(px, py)));   // a bay's black, wherever it climbed
if (accessories) {
    mark = Math.max(mark, m.shoulder() * SHOULDER_DEPTH * CoatRegions.shoulderBar(...));
    mark = Math.max(mark, m.face()     * faceMarking(skin, part, point, m));
    int leg = CoatRegions.LEGS.indexOf(part);
    if (leg >= 0 && m.bars()[leg] > 0) {
        mark = Math.max(mark, m.bars()[leg] * BAR_DEPTH * CoatRegions.legBar(...));
    }
}
f.diluteNeutral(px, py, lerp(keepRedBody, 1f, (float) mark),
                        lerp(keepBlackBody, 1f, (float) mark));

Both marked outcomes run that same painter and differ only in their constants and in how much of the list they draw. There is no special case anywhere, and the whole locus stays inside the restrict-only contract.

Grullo, and the order the two pigments come off in

The dilution is PigmentField.diluteNeutral, not the warm dilute that cream and champagne use. The two are opposites and the difference is an order: red down first, then black.

A black horse is (red = 1, black = 1) — a full load of pheomelanin that the eumelanin above it hides, visible nowhere but on the gradient’s bottom row. Scale the two pigments together and the black comes off first, unmasking that red on the way out and walking the sample diagonally into the golds; a “grullo” built that way is a milk-chocolate horse. So the dilution scales the visible red instead — red * (1 - black) — and stores back whatever reproduces it against the black that is left:

black' = black * keepBlack
red'   = red * (1 - black) * keepRed / (1 - black')

Same invariant whiten keeps, and for the same reason, with the two pigments free to move by different amounts. On a chestnut nothing is masked and it collapses to red * keepRed; on a black horse the numerator is 0, red goes straight to 0, and the sample slides down the gradient’s neutral column into the blue-greys where grullo actually lives.

Dun takes black off. It barely touches red.

That one asymmetry is what makes the three bases come out as different as they do from a single pair of numbers — and it is why keepRed sits so close to 1.

The points, and how a bay dun keeps them

A dun dilutes the body coat. The points do not go with it — that is what makes a bay dun a tan horse with black legs rather than a uniformly faded one, and a grullo a blue-grey horse with a black mane. So the mask is 1 on the ears and muzzle outright, 1 down the centre of the mane and the tail, and 1 up the bottom POINT_LEG_SOLID of each leg before fading out over the next POINT_LEG_FADE — a soft transition rather than a ring.

That handles a grullo, whose points are only points because of where they are. It does not handle a bay, whose black climbs its legs and face by a rolled amount that can reach most of the way up. For that there is one more line:

private static double alreadyAPoint(float red, float black) {
    return Math.sqrt(black * (1.0 - red));
}

The pigment model says it exactly. A gene that paints a point paints it absolutely: BayCoat sets red = 0, black = 1, and that pair — black with no red under it — occurs nowhere else. A black horse is (1, 1): black with a full load of masked pheomelanin, which is what makes it a body colour and not a point, and which is what dun is supposed to dilute into grullo. Multiply and the two separate cleanly: 1 for a point, 0 for a black body, 0 for a chestnut, and the partial values through a bay’s leg ramp turn out to be exactly the fade the mask wants. The square root shortens that fade — without it the top of a seal bay’s black leg diluted to a grey cuff between the tan body and the black point, and no real horse has one of those.

The midtstol: why the mane is not a flat point

The dorsal stripe does not stop where the hair starts. Forelock, mane centre, spine and tail centre are one continuous system on a dun — the Norwegian name for the mane's share of it is the midtstol, and the tail's is the halefjær. So the mane and the tail get a dark band down the middle with pale guard hairs either side, not a rectangle of undiluted base colour.

private static double midtstol(Skin skin, Part part, BodyPoint point) {
    return HairPattern.centreStripe(skin, part, point, MIDSTOL_HALF_WIDTH);
}

It reuses HairPattern.centreStripe, the shape the healer's red line down the mane already uses, so the band lands on the part's own axes rather than the world's — a mane and a tail are both rotated out of true by their rest pose, and neither “stripes along X” nor “along Y” is right for both.

The half-width is set by the mesh

MIDSTOL_HALF_WIDTH is 0.25 and that is arithmetic, not taste. A mane is four texels across, so the only two distances from the centre that exist are an eighth of the span and three eighths of it. centreStripe feathers over halfWidth ± halfWidth/2, and 0.25 puts those two edges exactly on those two rows: the inner pair is fully dark, the outer pair is fully pale. At 0.20 the feather caught the dark rows too and a bay dun's mane came out at 97 % black instead of black — invisible in a screenshot, wrong in the pigment field.

The guard hair is an overshoot, scaled by the dilution

The reference calls the outer hairs cream, silver-white or yellowish. A mane diluted only as far as the barrel reads as one colour at riding distance, so the guard hair is diluted past the body by GUARD_HAIR — but scaled by how far that channel is being diluted at all:

float over = hair ? (float) (GUARD_HAIR * (1.0 - mark)) : 0f;
f.diluteNeutral(px, py,
        lerp(keepRedBody,   1f, (float) mark) * (1f - over * (1f - keepRedBody)),
        lerp(keepBlackBody, 1f, (float) mark) * (1f - over * (1f - keepBlackBody)));

That scaling is what keeps it a dun feature. A d1 horse takes no black off anything, so its guard factor is exactly 1 and the painter stays the byte-for-byte no-op the tests pin. Guard hair is pale because the dilution is stronger there — it has no reason of its own.

Every dun, not only the Fjord

The Fjord is famous for this because the breed clips the mane upright so the pale-dark-pale cross-section is unmissable, and because a pale body makes a dark centre shout. The marking is not breed-specific and it is not a separate locus — a dun is a dun. So it is drawn on every D horse here, at a contrast that follows the base colour: near-black down a grullo's mane, dark red down a red dun's, and most dramatic on a cream-diluted dun, exactly as the field guides describe. The upright clip itself is grooming, and grooming is not modelled.

The reference this came from proposed a fictional F “Fjord mane pattern” modifier plus three hidden quantitative loci gating the effect. That is a new system for a marking every true dun already has, and it would have let a D/_ horse with the wrong modifiers show no mane stripe — which the same reference says is not how dun works. Declined; see philosophy on inventing loci.

The marking checklist, region by region

What a field guide looks for, and what the gene draws for it. Only the dorsal stripe is diagnostic; everything below it is a “sometimes”, which is why the accessories are rolled per horse.

Leg bars: strokes, not bracelets

The old field was a plain function of y — evenly spaced, equally strong, and identical the whole way round the limb — so every leg wore the same three rings, low down on the cannon. Every clause of that is wrong about a real dun. The rewrite is four separate irregularities stacked on one band phase:

On top of that, a window peaks at BAR_JOINT and dies out both at the pastern and where the leg meets the body. That is the same fraction of the leg box the point mask is fading out at, which is not a coincidence: the dark lower leg ends at the knee, and the bars start there, on the pale leg above it, which is exactly where a real dun’s read best.

The markings are incomplete on purpose

Leg barring, shoulder bars and forehead rings occur on some duns. A genuine dun may have a striking dorsal stripe and nearly invisible shoulder bars, or clear leg barring and no cobwebbing at all. So each accessory carries a propensity on the expressing copy — a stored number tested against that accessory’s chance, so the tendency is heritable and a line can be bred toward it. A horse regenerates the same markings every session, and a foal that inherits the copy inherits its dam’s.

/** absent on (1 - chance) of horses; on the rest, a faint smudge ramping up to full. */
private static double accessory(float roll, double chance) {
    if (roll >= chance) {
        return 0;
    }
    double t = (chance - roll) / (chance * 0.55);
    return t > 1 ? 1 : t;
}

Deliberately not a plain uniform: most of the duns that have a shoulder bar have a hint of one rather than a painted stripe. That ramp lives in the painter, and what is stored is the propensity it reads — seed (the marking field), dorsal_width, shoulder, face, and bar per leg. Both marked outcomes read the same values even though d1 only uses the first two, so a d1 horse that later gains a D copy keeps the stripe it had and gains the accessories it was always carrying.

d1: a marking with no dilution

A non-dun horse is by definition undiluted, so d1 cannot paint a dark line on it. It does not need to — it runs the same painter with constants chosen to move as little as possible:

• Black is never touched (keepBlack = 1), which is the whole of what “non-dun” means here. Taking black off is the half of the dilution that actually changes a horse’s colour, so an allele that does not dilute must not do it.
• Only the visible red is taken, and only 5% of it — less than D’s 7%, which is itself barely anything. A chestnut lightens a whisper, a bay body less, a true black or a bay’s points not at all.

Founder frequencies

The six weights are written out rather than computed, so one row can be retuned without disturbing the others. They are Hardy-Weinberg at p(D) = 1/24 and p(d1) = 1/10. The three D rows still sum to what the old two-allele table gave, so adding d1 split the non-dun population rather than making duns rarer.

So roughly one wild horse in five carries primitive markings without being a dun — common enough to be a familiar sight, and the reason the locus is worth three alleles rather than two.

The gene, and two different things near it

TBX3 is T-box transcription factor 3 — a protein that regulates when and where other genes switch on during development, with broad roles across mammals including limb and skeletal development. What makes it interesting here is that different regulatory variants near the same gene affect two very different traits: the dun coat, and (in certain Chinese horse and pony populations) a substantial component of wither height. The familiar dun test is about the first of those and predicts nothing about the second.

The original mapping put the coat locus in a roughly 200 kb interval on chromosome 8 containing TBX3. On EquCab3 the gene is annotated around 20.51–20.53 Mb; older papers report positions near 18.2 Mb because they used an older assembly. That is an assembly-version mismatch, not two loci. Imsland et al., PMC4731265

The mechanism, which is inside the hair

Dun is not simply less melanin. In a dun horse most body hairs have pigment packed into only about 25–50% of the hair cortex, giving a radially asymmetric, partly diluted shaft. Hairs in the dorsal stripe stay evenly and intensely pigmented, so the stripe stays dark.

The developmental sequence appears to be:

1. In a growing follicle, TBX3 is expressed asymmetrically in a subset of hair-bulb keratinocytes.
2. That asymmetry changes follicle differentiation and locally reduces expression of KITLG, a major signal supporting melanocyte migration and survival.
3. Melanocytes and pigment therefore occupy only part of the developing cortex.
4. The finished hair looks diluted, while selected regions keep darker hairs and form the primitive markings.

That explains why dun looks so patterned: it is controlled at a microscopic radial level inside individual hairs as well as by where on the body the follicle sits.

Three alleles, one dominance order

D > nd1 > nd2, where D is dun, nd1 is non-dun with possible primitive markings, and nd2 is ordinary non-dun with no genetically expected markings.

Some laboratories write d1 and d2 in lower case; same alleles. The mod uses that second convention. A single D is enough to produce dun, so a D/nd2 horse is visibly dun and transmits D to only half its foals, while D/D transmits to all of them.

Why a dorsal stripe is not proof of dun

A horse can show a countershading stripe, seasonal shading, or an nd1-associated marking without being genetically dun — easily misread in buckskins, bays, red horses and horses carrying other dilutions. Going the other way, the darker the base the harder some markings are to see: in the original work, markings were easier to recognise against red pigment than against dark eumelanin. Phenotype alone misleads; the DNA test is the clean answer.

What it looks like on each base

The most reliable hallmark is the dark dorsal stripe from mane to tail. Others: leg barring (zebra bars), a shoulder cross, dark ear edging, facial masking, cobwebbing or forehead marks, neck and shoulder shadows, dark tail barbs, and dark regions around the muzzle and vibrissae. Expression is variable — not every dun horse shows every marking clearly, which the mod reproduces deliberately in the incomplete-marking rolls.

What the coat locus does not do

There is no good evidence that D/nd1/nd2 changes build, bone, conformation, muscle mass, mature height, frame, movement quality or athletic phenotype. The coat alleles alter a hair-follicle enhancer’s activity in a restricted developmental context, and the foundational paper specifically reported no recognised limb-development or other pleiotropic differences between dun and non-dun horses.

So a Fjord’s compact sturdy build, a Quarter Horse’s muscling and a Konik’s hardiness are not consequences of the dun genotype. They are breed-level outcomes from many genes, selection history, management and environment. The mod agrees: dun paints and contributes nothing to the trait layer.

The separate height signal

A different regulatory region upstream of TBX3 has been associated with stature in Chinese horses and ponies. The strongest site, TBX3-EN2, is an A/G SNP at ECA8:20,644,555; in a panel of 763 Chinese horses and ponies the G allele went with greater wither height, and the model estimated it could account for up to about 20.3% of height variation — roughly 10 cm — in that population. PMC8796118

The functional work makes it credible: reporter assays found the G allele increased enhancer activity and TBX3 expression; TBX3 expression was higher in the placentas of larger Yili horses than in Ningqiang ponies; and deleting the corresponding enhancer segment in mice reduced TBX3 expression in developing limbs and produced shorter limb bones. The A allele was common in ponies — 65.6% across seven pony breeds — and fixed in the 13 sampled Przewalski’s horses; G averaged 81.2% across 18 Chinese horse breeds, ranging from 21.1% in Debao ponies to 85.7% in Yili horses.

Health, jumping, speed

Dun is a pigmentation trait, not a disorder. OMIA classifies it as not disease-related, and the foundational study found no developmental, limb or systemic effects attributable to the coat variants.

That needs distinguishing from the wider biology of the gene. In humans, damaging one copy of TBX3 causes ulnar-mammary syndrome — limb, tooth, gland and genital abnormalities. That does not mean a dun horse carries a disease mutation: the equine coat variants leave the protein intact and modify hair-follicle expression rather than removing the gene’s developmental function. It is the cleanest example on this wiki of why “this gene causes X in humans” is not an argument about a horse.

For the size enhancer there is likewise no established equine health syndrome or soundness risk. The mouse enhancer-deletion experiments show why TBX3 can affect limb growth; they are not evidence that naturally occurring horse genotypes cause orthopaedic disease, hoof disease, infertility, metabolic disease or reduced lifespan.

Jumping. No validated evidence that any of these variants predicts scope, technique, carefulness, bascule, soundness under jumping load or competition success. TBX3’s developmental role makes it tempting to speculate about limb architecture and biomechanics; that inference has not been demonstrated in horses, and would need replicated genotype-phenotype associations controlling for breed, height, training, rider, pedigree, competition level and other athletic loci.

Speed. No robust evidence that the dun alleles affect sprint speed, stamina, earnings, stride mechanics or gait. The speed-gene discussion centres on MSTN. The Chinese size variant could in theory correlate indirectly with body dimensions where it affects height — which is not causing speed, and taller is not inherently faster.

Breeds and frequency

The coat work genotyped 1,814 animals from more than 45 breeds, confirming that TBX3 variation is broad rather than confined to one “primitive” breed. Breed-by-breed frequencies remain unevenly measured, and many figures circulating online rest on phenotypes, registry preferences or small samples rather than population surveys.

OMIA documents the dun phenotype or the causal variants in the American Trotter, Arabian, American Bashkir Curly, Faroes Pony, Norwegian Fjord, Gotland Pony, Icelandic Horse, Mongolian Horse, Polish Konik, Quarter Horse, Shetland Pony, Swedish Warmblood and Vyatka. Variant-specific reporting is narrower:

• D in Curly, Fjord, Icelandic, Polish Konik, Quarter Horse and Vyatka.
• nd1 in American Trotter, Arabian, Curly, Faroes Pony, Gotland Pony, Mongolian, Quarter Horse, Shetland Pony, Swedish Warmblood and Vyatka.
• nd2 is the reference-genome allele, widespread, and the most prevalent allele among common domestic breeds.

The distinction between “known to occur” and “common” matters here more than anywhere. For Arabians, Swedish Warmbloods, American Trotters and Shetlands, documentation of nd1 does not mean many individuals are visibly dun — nd1 is a non-dun allele that merely preserves subtle markings.

The mod uses D at 1/24 (4.2%) and d1 at 1/10 (10%), so a founder is more than twice as likely to carry the marking allele as the dilution — which matches the qualitative picture above, where nd1 turns up in far more breeds than D does. No published global figure exists to check either number against.

Evolution, and the open questions

Well established. Dun is likely the ancestral equid coat state, seen in Przewalski’s horses and other wild equids, and plausibly adaptive camouflage — inferred from ecology and phylogenetic distribution rather than directly quantified. The D and nd1 haplotypes predate domestication: ancient DNA identified a roughly 42,700-year-old horse carrying D/nd1 and a roughly 4,400-year-old horse inferred nd1/nd1. nd2 is more recent, was not observed outside domestic horses in that study, and likely rose during domestication alongside selection for more intensely pigmented, non-camouflage colours. And all three are regulatory changes — they alter where TBX3 is expressed in a growing follicle, not the protein’s amino-acid sequence.

Speculative or incomplete. The camouflage advantage is plausible and often stated, and its magnitude and historical setting are unresolved. Researchers have proposed that spatial TBX3 regulation could illuminate the evolution of zebra stripes — TBX3 is not established as the zebra-stripe gene, which is worth knowing next to the mod’s own zebra genes. Work in dun Mongolian horses is exploring differential TBX3 skin expression in relation to Bider markings, an active area that does not change the central point: dun expression is patterned and region-specific rather than on/off. Claims that dun changes temperament, durability, athleticism or “primitive hardiness” are unsupported unless they can separate TBX3 from breed ancestry and selection. And the TBX3-EN2 height finding needs validation outside the Chinese populations before it can be a general stature predictor.