What the gene does, the three different ways a variant breaks it, where each of the six characterised variants sits, and what happens when a cat inherits two
The material on this page is drawn from the Global Cat Health Survey 2026, which is where its sources and its declarations of interest are set out in full.
Every dominant blue eye variant named so far is a variant of one gene, PAX3. That fact is repeated everywhere and explained almost nowhere, and it leaves a breeder holding a test result that names an allele with no way to find out what the allele actually did. This page is that explanation. It is written from the four published papers and it says, at each point, what those papers state and what they do not.
Every cat carries two copies of every chromosome, one from each parent, and therefore two copies of PAX3. Nothing on this page makes sense without that, and it is the step most explanations leave out.
A cat with a dominant blue eye variant has one mutated copy and one entirely normal copy. That is what the word heterozygous means on a test result, and it is what this page means by one working copy instead of two: it is counting chromosomes, not counting variants.
Mutated means the sequence has changed, and that is a statement of fact rather than of consequence. What the change costs the cat is not the same in every case, and this page keeps the two apart:
The cat does not experience a mutation. It experiences a shortage.
PAX3's job is to make a protein that switches on the machinery sending pigment cells out across the embryo, to the coat, to the iris, to the inner ear. The amount of that protein matters. Two working copies make enough to finish the journey. One working copy makes about half, and half is enough for most of the body but not enough to complete the last and hardest part of the trip: the far edges, where the pigment cells arrive last and fewest. The iris is one of those edges. So is the odd patch of coat, and so is the inner ear.
Three different things happen inside the cell, and they have the same effect on the cat. The table below is the interpretation: what the mutation does, and what that leaves you looking at.
| Variant | What happens in the cell | What that means for the cat |
|---|---|---|
| DBECEL DBEALT | The gene is intact and the protein it makes is perfectly normal. What has been disturbed is the set of switches in intron 4 that decide how much is made and when. So the cell makes the right protein and not enough of it, or not at the moment it was needed. | Fewer pigment cells set off, and they run out before the journey ends. They reach the body and the head, which is why the cat is a normal colour. They do not all reach the iris, so the eye stays blue. Sometimes they do not reach a patch of skin, which is the small white marking. Sometimes they do not reach the inner ear. |
| DBEAGO DBEMARU DBERE | The gene is read and translated, but a stop signal appears in the middle of the instruction, so the protein is built halfway and then abandoned. PAX3 works by gripping DNA and switching other genes on; a half-built protein cannot grip anything. It is made, and it is inert. | Exactly the same picture. The cell has spent the effort and got nothing back, so the supply of usable protein is the same as if that copy were absent. Blue or odd eyes, little or no white, sometimes deafness. |
| DBENAD | There is nothing to read. The gene has been deleted from that chromosome, so no message is made and no protein is built. This is the simplest of the three and the one with the fewest steps. | Again the same picture. The cat has one working copy and the shortfall it produces is the shortfall every row of this table produces. |
The Igor determinant is on that figure too, with one cell left blank. The evidence places it at PAX3: four named alleles have been excluded from the line by assay, the one cross it has made to a named variant behaved as two variants at a single locus behave, and the phenotype is the PAX3 phenotype. So a cat of that line has the same shortfall as the rows above it. What is not published is which of the three routes it takes, and when it is, only that one cell changes.
Six of the eight were characterised in peer-reviewed work and two were not, and the difference is who did the work rather than how good it is. An academic laboratory publishes because publishing is how its results are checked. A commercial laboratory sells a test and has no such obligation.
| Variant | Established by | What that means for a test |
|---|---|---|
| DBECEL | Abitbol, Couronné, Dufaure de Citres and Gache, Animal Genetics 2024 | Published, so the sequence and the assay are in the open literature and any laboratory can offer it |
| DBEALT | Abitbol, Dufaure de Citres, Rudd Garcés, Lühken, Lyons and Gache, Animals 2024 | Published |
| DBERE | First reported by Rudd Garcés and others, G3 2024, and named DBERE in the Animals paper of the same year | Published. Note that a product sold as a deafness test for this line is a DNA test for the variant and not a hearing test |
| DBEAGO | Abitbol, Cloquell, Kaczmarska, Holmes, Lühken and Macaulay, Animal Genetics 2025 | Published |
| DBEMARU | Abitbol, Eiselé and Lipovenko, Animal Genetics 2026 | Published |
| DBENAD | Abitbol, Eiselé and Lipovenko, Animal Genetics 2026 | Published. The test reads across the break points, so it is a different assay from the others and must be asked for by name |
| The Igor determinant | The exclusions that place it at PAX3 come from the same laboratory's published work, where cats of this line were genotyped against four named alleles and returned wild type each time | No test is on sale. A commercial panel can exclude the named variants from a cat of this line. No panel on the market reports this determinant, so nothing a breeder can buy confirms it |
| DBEcDc DBExHW | PetGeno, a commercial laboratory in China, which identified DBEcDc in August 2025 and DBExHW in September 2025 and stated both publicly in October 2025. No publication exists for either. The laboratory names PAX3 for DBExHW and names no gene for DBEcDc | A test is on sale for DBEcDc and none for DBExHW, from the laboratory that found both. A test cannot be built without a sequence, which is why this register treats DBEcDc as identified rather than claimed. The gene, the position and the mechanism are public for neither, so nothing about them can be checked against a paper and neither appears on the figures above |
Where this register can name a laboratory it does so, because a reader should be able to see whose work a statement rests on. Where a variant rests on a commercial identification rather than a publication, that is recorded plainly rather than smoothed over: it is a real difference in what can be checked.
One caution about a natural inference. Cats of the Igor line and cats of the Nadeya line look alike. That is not evidence that the two share a mechanism, because every variant on this page produces cats that look alike: that is the whole point of the section above. Appearance cannot separate a deletion from a stop codon from an insertion, and it cannot group them either.
The practical consequence is the one worth carrying away. Three different molecular events, and a cat carrying any of them looks the same as a cat carrying any other. The eye cannot tell you which variant a cat has, the coat cannot, and neither can a hearing test. Only a DNA test naming the variant can.
Where the shortfall shows, and why it is those places. Pigment cells start in the embryo's neural crest and travel outwards. The parts they reach last and in smallest numbers are the first to miss out when there is not enough PAX3 to see them all the way: the iris, the tips and edges of the coat, and the inner ear. That is why one gene produces a blue eye, a white marking and sometimes deafness, and why those three travel together.
Everything above describes molecules. This is what it amounts to in an animal, which is the only reason any of it matters. The three consequences of the shortfall are not equal, and they should not be spoken of as though they were.
| What the shortfall causes | What it means for the cat | What the evidence says |
|---|---|---|
| A blue or odd eye | No welfare consequence is reported for the eye itself. The iris lacks pigment; the eye is built normally otherwise, and no paper in the survey reports impaired vision in any dominant blue eye cat of any variant. It is a cosmetic difference, and it is the one everybody notices. | No vision problem has been reported. That is an absence of reports and not a study of vision, and no paper here set out to test sight. |
| A white marking | No welfare consequence in itself. A patch of skin the pigment cells did not reach. Its significance is as a sign rather than a symptom: the amount of white is a rough guide to how far the shortfall went, and the cats with the most white are the ones that carry two affected copies. | Consistently reported as minimal to none in single-copy cats. Extensive white appears in the two-copy cats, which is the pattern the survey watches for. |
| Deafness | This is the welfare question, and it is the only one of the three that is. Pigment cells are needed in the inner ear for it to work at all. Unilateral deafness a cat compensates for and an owner may never notice; bilateral deafness changes how a cat can live, is permanent, and cannot be treated. | Not raised in DBECEL carriers across 47 hearing tests, and not associated with DBEALT carriers. Confirmed in DBERE carriers in one study and absent in ten further cats in the next. No hearing data at all exists for DBEAGO or DBEMARU. |
So a cat with one affected copy is, on everything published, a healthy cat. It sees, it is not in pain, it is not shortened in life, and for most of these variants its hearing has not been found to be at greater risk. The blue eye is not an illness and the survey has never said it was.
The welfare harm in this subject is concentrated almost entirely in cats with two affected copies, and there it is severe and documented: kittens born white and deaf, kittens dead at birth with cleft palates and contracted limbs, and kittens lost in the womb and never born at all. Those come from one decision, which is mating two carriers together, and nothing about the individual cat causes them.
One qualification that applies to every row above. Where a column says no harm has been reported, that is a statement about what has been looked for. Two of the six variants have no hearing data of any kind and none of the six has been the subject of a study of vision or of lifespan. Not found is not the same as not there, and this page does not use one to mean the other.
Half is enough for most of the body and not enough for the end of the journey. Pigment cells begin in the embryo's neural crest and travel outwards, and the places they reach last and in the smallest numbers are the first to miss out when the supply runs short: the iris, the edges of the coat, and the inner ear. That is why one gene produces a blue eye, a white marking and sometimes deafness, and why those three travel together.
Three different routes, and the same destination: one functioning PAX3 instead of two. That is why they can share a column in the tables on this page, and it is why this page says mutated rather than faulty when it means all six at once.
So the deleted copy carries nothing. It cannot also carry DBECEL or DBEMARU, because there is no gene sitting there to carry them. And the other copy is normal, so it carries nothing either. One cat, one altered chromosome, one normal chromosome, and no stack of variants anywhere.
The same is true of all six. A DBEMARU cat has one PAX3 with a stop codon in exon 5 and one normal PAX3. A DBEALT cat has one PAX3 with an insertion in intron 4 and one normal PAX3. They differ in what happened to the one affected copy, and they are identical in having exactly one normal copy left. That is why they all produce similar-looking cats.
So a cat does not carry several variants stacked on one gene. It carries one change on one of its two copies. A cat with DBENAD has PAX3 deleted from one chromosome and a normal PAX3 on the other; the deleted chromosome cannot also hold DBECEL or DBEMARU, because there is no gene sitting there to hold them. A cat with DBEMARU has one PAX3 with a premature stop in exon 5 and one normal PAX3. The six differ in what happened to the affected copy, and they are identical in having exactly one normal copy left. That is why they produce cats that look so much alike.
The exception, and it is the dangerous one, is a cat that inherits a different variant from each parent. Then both copies are altered and there is no normal one. That case is set out under what happens when a cat inherits two, and it is why two dominant blue eyed cats of different lines are not a safe mating.
The two dangerous rows are the bottom two, and both of them need two carrier parents. A carrier bred to a cat with two normal copies cannot produce either of them. That is the whole of the breeding rule in one picture.
One point of vocabulary, because it causes more confusion than anything else on this subject. PAX3 is a normal gene. Every cat has two copies of it and always has, in the way every cat has a heart. A cat does not "carry PAX3"; what a dominant blue eye cat carries is a variant of PAX3, which is one of its two copies in an altered or a deleted state.
So it takes one parent, not two, to produce a carrier. The other parent passes an ordinary PAX3, as any cat would. The cats in these lines are not homozygous: they are heterozygous, one affected copy and one normal one.
Two carrier parents is the situation to avoid, and it produces the bottom two rows of the figure. If both pass the same variant the kitten is homozygous; if they pass different ones it is a compound heterozygote. In both cases the kitten has no working PAX3 at all, and those are the kittens that were born white and deaf, died at birth, or were never born.
PAX3 encodes a transcription factor: a protein whose job is to switch other genes on and off. One of the genes it regulates is MITF, which is the master control for the cells that make pigment. Those cells, the melanocytes, do not start life in the skin or the eye. They arise in the neural crest of the embryo and migrate outwards to the coat, to the iris, and to the inner ear.
That migration is why one gene reaches three apparently unrelated things. Where the pigment cells arrive in the iris, the eye is coloured; where they do not, it is blue, because a blue eye is not a pigment but the absence of one. Where they arrive in the coat, there is colour; where they do not, there is white. And where they fail to populate the stria vascularis of the inner ear, that ear does not work. The blue eye, the white marking and the deafness are three views of one process, which is why a blue eye in the survey is followed by a hearing question and not only a colour one.
A gene is not one continuous instruction. It is written in blocks. The exons are the parts kept in the final message and translated into protein. The introns between them are cut out before translation, but they are not empty: they carry the switches that decide how much protein is made and where. Both things matter here, and the variants divide neatly between them.
| Where the variant lands | What it does | Which variants |
|---|---|---|
| In an exon | Changes the protein itself. A letter is deleted or swapped and the instruction is cut short, so the protein comes out truncated and cannot do its job. The cell ends up with one working copy where it should have two. | DBEAGO, DBEMARU, DBERE |
| In an intron | Changes no protein at all. The insertion lands beside the switches that regulate the gene, so the protein is normal and there is less of it, or it appears in the wrong place at the wrong time. This is why a non-coding variant is not a harmless one. | DBECEL, DBEALT |
| Neither: the gene is removed | A rearrangement of several million bases deletes the whole gene from one chromosome, along with thirteen of its neighbours. Again one working copy instead of two, but by subtraction rather than by damage. | DBENAD |
All three routes arrive at the same place, which is the point worth holding on to: one functioning copy of PAX3 instead of two. Geneticists call that haploinsufficiency, and it is why six variants that look nothing like each other in the laboratory produce cats that look very much alike in the flesh.
Exon and intron, in plain terms, because tier three cannot be read without them. A gene is not one continuous instruction. It is written in pieces, and the cell reads all of them and then throws part of the text away before anything is built. The pieces that are kept are the exons: strung together they spell out the protein, and PAX3 has eight of them, the numbered boxes on the diagram. The pieces cut out are the introns: the gaps between the boxes. Nothing an intron says reaches the protein. That is why the same gene can be altered in two quite different ways. A variant that lands in an exon changes the protein itself, usually by cutting it short. A variant that lands in an intron changes no protein at all, because the intron was going to be discarded anyway; what it does instead is disturb the switches an intron carries, which set how much protein the copy makes. Neither is the milder kind. Both leave the cat with one copy pulling its weight instead of two, which is the whole of the effect and is set out above. An intron is not empty space and a non-coding variant is not a lesser fault.
Adapted from Abitbol M, Eiselé T, Lipovenko E, Additional PAX3 variants for dominant blue eyes in cats, Animal Genetics 2026;57:e70199, Figure 1g, which draws the wild-type allele above the DBE allele on chromosome C1. Tier two follows that model. The three-tier zoom, tiers one and three, the placement of the five variants that lie inside the gene, and the two-copy panel are the survey's, drawn from the four papers cited on this page.
† How the 1 747 is derived, so it can be checked. DBECEL is published as an insertion at position 206 974 029 (Abitbol and others, Animal Genetics 2024;55:670, p.671). DBEALT is published as an insertion at position 206 975 776 (Abitbol and others, Animals 2024;14:1845, p.11). Both are on chromosome C1 against the Felis_catus 9.0 reference genome, so the two are directly comparable. 206 975 776 − 206 974 029 = 1 747. No paper states this distance, and none states that the two variants are hard to tell apart; the subtraction is the survey's and is shown here so a reader can repeat it. Every other number on this figure is quoted from the paper it comes from.
Every cell below is a quotation or a direct reading from the paper named in the last column. Where a paper does not state something, the cell says so rather than filling the gap.
| Variant | Where | What it is | Effect on the protein | Founder, breed, country | Named |
|---|---|---|---|---|---|
| DBEAGO | exon 2 | A single base deleted, which throws the rest of the instruction out of step. PAX3:c.160del | Truncated at p.(His54ThrTer108). Loses 78 per cent of the protein, the largest loss of any variant here | Not named in the paper. Maine Coon, the Agostino line | 2025, Abitbol and others, Animal Genetics 56:e70020 |
| DBECEL | intron 4 | A 395 base insertion of a FERV1 endogenous retrovirus LTR at position 206 974 029. Non-coding | None. No amino acid changes. It sits beside the regulatory elements CNE2 and CNE3 and acts on them | Roxi, an outbred cat from Kazakhstan. The Céleste was built on this variant | 2024, Abitbol and others, Animal Genetics 55:670 |
| DBEALT | intron 4 | A 433 base insertion of an RD-114 endogenous retrovirus LTR at position 206 975 776. A different virus, a different length. Non-coding | None. Sits beside CNE3 and CNE4 | Fyodor, found in the streets of Ust-Kamenogorsk, Kazakhstan, in the 1990s | 2024, Abitbol and others, Animals 14:1845 |
| DBEMARU | exon 5 | One letter changed, creating a stop signal in the middle of the instruction. XM_019838731.3:c.784C>T | Truncated at p.(Arg262*). Loses 46 per cent of the protein | Marusja, an outbred domestic shorthair of no known ancestry, Russia, 2019 | 2026, Abitbol, Eiselé and Lipovenko, Animal Genetics 57:e70199 |
| DBERE | exon 6 | One letter changed, again creating a premature stop. PAX3:c.937C>T | Truncated. The papers give no p. notation and no percentage for this one | Rociri Elvis, a purebred Maine Coon, the Netherlands | 2024, named in Animals 14:1845, variant first reported by Rudd Garces and others |
| DBENAD | the gene is deleted | A rearrangement of 6 289 500 bases: a deletion of 2 314 301, an inversion of 3 878 523, then a deletion of 96 677 | The whole gene is gone from that chromosome, with thirteen neighbouring genes. One working copy remains | Ermine Trace Nadeya, a purebred British Shorthair, Russia, born 12 October 2016 | 2026, Abitbol, Eiselé and Lipovenko, Animal Genetics 57:e70199 |
| The two variants below are set out in the same columns so that the gaps can be seen rather than described. Both were identified by a commercial laboratory in China that does not publish. The gene is now named for one of them, and for neither is there a position, a sequence or a paper. They are here because a breeder may be shown either name on a result sheet, and because leaving them off would suggest the list of variants is closed. | |||||
| DBEcDc | Not stated for this variant. See the note below the table: the laboratory's own numbering points at PAX3 without stating it | Not stated. The laboratory calls it a mutation site and gives no position, no length and no class | Not stated. Whether it changes the protein at all is unknown | Chinese domestic cats, China. Subsequently detected in Abyssinians of one cattery in China, which makes this the only variant of the eight to enter a registered breed with no publication behind it | No publication. Identified August 2025 and stated publicly by the laboratory in October 2025. A test is on sale |
| DBExHW | PAX3. The laboratory's own material calls this the sixth identified PAX3 mutation. That is a stated attribution and not a paper, and it is the only thing about this variant that anyone outside the laboratory can repeat | Not stated. No exon, no intron, no position, no length, no class. It can be named on the gene and still not be drawn on it, which is why it is absent from the figures above | Not stated. | Black domestic cats, China. A heterozygote of the line is written out by the laboratory as N/DBE6, the same notation every other dominant blue eye result carries | No publication. Identified September 2025 and stated publicly by the laboratory in October 2025. No test is on sale, not even from the laboratory that found it |
What the two rows do and do not mean. They do not mean the variants are doubtful. The laboratory states that both follow the same dominant inheritance as the published variants, it sells a test for one of them, which cannot be built without a sequence, and it now names the gene for the other. What is missing is the ability of anyone else to check any of it. If the sequences are published the empty cells fill in an afternoon.
One inference, marked as an inference. The laboratory calls DBExHW the sixth identified PAX3 mutation, and its own numbering runs DBE1 CEL, DBE2 RE, DBE3 ALT, DBE4 AGO, DBE5 cDc, DBE6 xHW. If the sixth is at PAX3 then the five below it are, and the fifth is DBEcDc. That is a reading of a count rather than a statement about DBEcDc, so this page does not record the gene as established for it. Note also what the count leaves out: it stops at six, and DBENAD and DBEMARU, both published in 2026 and both at PAX3, are not in it. The numbering is the laboratory's own and is not the field's, which matters to any breeder handed a result sheet reading variant 5 of 6.
The other five variants leave PAX3 in place and alter it: a letter changed, a letter removed, a piece of retrovirus dropped into an intron. DBENAD does not alter PAX3. It takes it out. On one of the cat's two copies of chromosome C1, a stretch 6 289 500 bases long is rearranged, and the gene is inside the part that is deleted.
It has three parts and they do three different things.
| Part | Size | What it does |
|---|---|---|
| The first deletion | 2 314 301 bases | This is the one that matters. PAX3 is inside it, so PAX3 is gone from that chromosome. So are thirteen other genes. |
| The inversion | 3 878 523 bases | Nothing is lost here. The block is still present, turned back to front. It is the largest of the three parts and the least consequential |
| The second deletion | 96 677 bases | Also gone, but no gene is described in that stretch, so nothing known is lost with it |
It is the obvious question, and the answer has two halves that point in different directions.
For PAX3 itself, no, and that is what dominance means. The effect is set by how much working protein the cat can make, and a copy cannot contribute less than nothing. A copy whose protein stops halfway through contributes nothing usable; a copy that has been deleted contributes nothing at all. Zero is zero. DBENAD and DBERE arrive at the same place by different roads, and the cat cannot tell them apart.
For everything else on that stretch of chromosome, the question is open, and this is where DBENAD genuinely stands alone. It is the only one of the six that removes anything other than PAX3, and it removes thirteen other genes with it. That is set out immediately below. The PAX3 consequence is the same as the others; the thirteen extra unknowns are not shared by any of them.
Every other variant on this page affects one gene. This one removes fourteen. The authors examined the other thirteen and report that two of them are associated with other conditions, FARSB and MRPL44, but that in both cases the deleterious variants described are autosomal recessive, which means one lost copy would not produce the condition. No such phenotype has been seen in any cat of the line.
Their conclusion is stated carefully and this register repeats it as they wrote it: it is likely that losing one copy of the thirteen is not responsible for the effect, but the involvement of one or more of them, individually or together, cannot be ruled out. That is a sentence about a gap in knowledge rather than about a known harm, and it exists for no other variant here.
This is the difference a breeder should take away. The other five variants were found by reading the gene: sequencing its exons and the boundaries around them. That method was tried on this line first and it found nothing, because there was nothing wrong with the gene that was left. The fault was not inside PAX3; the fault was that a copy of PAX3 was missing.
Finding it needed a different kind of work: whole genome sequencing of one cat, and thirty gigabases of long-read sequencing of a second, in order to see the shape of the chromosome rather than the letters of the gene. The test now used to detect it works by reading across the break points, the joins where the deleted ends meet, because that join exists only on a chromosome carrying the variant.
It follows that a test for the other variants will not find this one, and a test for this one will not find the others. They look for different kinds of thing. A cat reported clear on a panel of point variants and insertions has not been tested for DBENAD unless the panel says so by name.
For all that, the outcome in the cat is the ordinary one. One copy of PAX3 is gone and the other is entirely normal, so the cat has one working copy instead of two, which is exactly where the other five arrive by damaging the gene rather than removing it. A cat with DBENAD carries one PAX3 and one variant, like any other carrier.
The genotyping bears that out. Thirty-nine dominant blue eyed cats of the Nadeya line were tested and every one was heterozygous; sixteen dominant blue eyed cats of other lines were negative for it. No homozygote has ever been found, and the proportions in the litters indicate they are lost before birth. The line was never outcrossed and is maintained inside the British genetic background.
One caution the authors give about their own work, recorded here because it bears on how firmly any of this can be stated: the diagram in that paper shows the break points, and further study will be needed to characterise the sequence within them in detail. The shape of the rearrangement is established; the fine detail at its joins is not yet.
DBECEL and DBEALT are both insertions in intron 4, about 1 747 bases apart. That figure is the survey's arithmetic from the two published coordinates and no paper states it. Their closeness is why the two lines were confused with each other for years and why breeder charts still attribute one line to the wrong variant.
They are nonetheless two different things and a laboratory can tell them apart without difficulty: different retroviruses, FERV1 against RD-114, and different lengths, 395 bases against 433. Each has its own assay. A test for one is not a test for the other, and a result reading wild type for DBEALT says nothing whatever about DBECEL. Several results held in this register have been over-read in exactly that way.
This page explains the gene. What follows from it for a breeding programme, variant by variant, is set out on the health page: what is known about hearing in each, what happens when two carriers meet, what the pitfalls are for each variant in turn, and how much evidence each of those statements actually rests on.
Three of the eight variants this register holds cannot be placed on the figure. DBEcDc and DBExHW, both from China, are identified by a commercial laboratory that does not publish. A test is on sale for DBEcDc, and a test cannot be built without the sequence; no test is on sale for DBExHW, so that one rests on the laboratory's statement alone. The laboratory names PAX3 for DBExHW and gives no position for it, which is enough to place it on the gene and not enough to place it on a figure. The Igor determinant is the third: the evidence places it at PAX3, four named alleles have been excluded from the line by assay, but no allele has been published and no test is on sale.
Two further things the papers do not supply, and this page therefore does not draw. No paper maps the PAX3 protein domains to its exons, so no domain diagram is given here. No paper gives a genomic coordinate for DBERE or DBEAGO, only a transcript position. Where this register needs a figure it cites the paper, and where no paper gives one it says so.