Dog Breeds

"In these three genes, I think it is more likely that these alleles have arisen through mutation." — Jean Lightner, creationist geneticist, on dog coat types, Answers in Depth, 2009¹

"Dawkins is exactly right—dog breeding is a wonderful example of the power of selection acting on hidden random mutations." — Michael Behe, Darwin Devolves, 2019²

Summary

Dogs are the largest selection experiment people have run on a mammal.  In about two centuries breeders produced the Chihuahua and the Great Dane from a species whose wild relative is the grey wolf, and both sides of the creation debate cite the result.  This article asks what was watched, which gene variants did the work, where each came from, and what it cost.  Each variant is scored with the four classes of Measuring Biological Information:  gain, loss, modification or transfer.

The findings, in order of strength:

1. Breed formation is documented history, not inference. Most modern breeds were founded with closed stud books between about 1830 and 1900, and genome data reproduce the written crosses closely.^{3 4} The count is about 350, not 400.⁵ The breeds were cut from about two dozen older breed types.⁴
2. Breeders mostly sorted variants that already existed. The small-size allele at the IGF1 locus is the ancestral state in wild canids and sits in a 53,000-year-old wolf.⁶ The two gene copies that shorten legs, the three coat variants and the extra starch-digestion genes are all older than the breeds.^{7 8 9}
3. But "variation the wolf already carried" overstates it. Many breed-defining variants are absent from wolves and arose in dogs after domestication and before the breeds. The creationist geneticist Jean Lightner says so of the coat alleles: they "have arisen through mutation."¹ At least one variant, the whippet's myostatin deletion, probably arose inside the breed period.¹⁰ New copies of the FGF4 gene are still appearing in dogs today.¹¹
4. The breed-defining variants are losses, weakened genes and extra copies. Of thirteen variants scored here, four are losses, two are modifications down, four are modifications up, three of them by extra copies of a gene, and three are unmeasured. None is shown to be a gain. This site's foundation article scores the FGF4 retrogene as a gain; the evidence read here says it adds more of a signal the tissue already makes, which is a modification.
5. The cost was large and measurable. The average breed is as inbred as the offspring of a brother and sister.¹² Breed dogs carry 2 to 3% more harmful gene variants than wolves.¹³ Of 594 dog gene variants with a known single-gene effect, 541 cause disease.¹⁴ The copy that shortens the legs of beagles and French bulldogs raises the odds of spinal disc disease about fifty-fold.⁷
6. The silver fox experiment, the one domestication watched from the start, also ran on existing variation. Its founders came from fur-farm stock already bred in captivity for decades and already tamer than wild foxes.^{15 16}
7. Selection limits are real but not general. In the Illinois corn experiment, begun in 1896, selection for high oil has run more than a hundred generations "with no sign of a plateau". The low-oil line stopped at a biological floor.^{17 18}

Verdict.  Dog breeds are strong evidence that selection can reshape a mammal in two centuries, and good evidence for the creationist claim that the reshaping mostly uses and damages what is already there.  They are poor evidence for the origin of new function.  But the young-earth reading has two holes.  Much of the material was made by mutation in dogs, which Lightner concedes, and the ages of the older variants come from dated fossils that the young-earth model has not answered.  Summarising all of this as "loss of information" is also wrong:  extra gene copies and turned-up genes are not losses.  They are copies that wrote no new sequence.

Part 1: What was watched

The breeds are recent and written down

Dogs are old.  Breeds, in the modern sense, are not.  A breed today is a closed population:  a dog is registered only if both parents were registered, and it is judged against a written standard.  Greger Larson and colleagues put the date plainly in 2012:  "Modern breeding practices, focused on distinct breeds with strict aesthetic requirements and closed bloodlines, only emerged in the 19th century, and claims for the antiquity (and long-term continuity) of modern breeds are based upon little or no historical or empirical evidence."¹⁹  Their summary:  "The vast majority of modern breeds were only created in the past 150 y".¹⁹  Bridgett vonHoldt's team dated the same process to "the Victorian era (circa 1830–1900)".³

That history can be checked against DNA, and it passes.  Heidi Parker and colleagues at the National Human Genome Research Institute genotyped 1,346 dogs from 161 breeds in 2017.  In 146 of the 161 breeds, 91%, every dog grouped with its own breed in every one of 100 bootstrap trees.⁴  They then measured how much DNA pairs of breeds share in long stretches.  Long shared stretches mean a recent cross, because recombination breaks them up a little each generation.  Calibrated on crosses recorded in stud books between 35 and 160 years ago, the method dated undocumented crosses:  "the Golden Retriever was separated from the Flat-coated Retriever in 1895 and the written history of the Golden retriever dates to crosses between multiple breeds taking place between 1868 and 1890 ( Figure 5b ), a near perfect match."⁴  So the breed period is one of the few stretches of evolutionary change where the pedigree was written down while it happened and the genome can be read afterwards.

How many breeds, and how old

The catalogue description of this article said "four hundred breeds in two hundred years."  Both numbers need correcting.

• Count. The Fédération Cynologique Internationale, the largest registry, says: "The FCI recognises 364 breeds."⁵ Parker's 2017 paper says "more than 350 distinct breeds".⁴ The American Kennel Club recognises fewer, about 160 when Cadieu's team counted in 2009.⁸ The figure of "more than 400" appears in a 2018 Journal of Creation paper by Cody Guitard, which cites Parker's paper for it.²⁰ Parker's paper says 350. Answers in Genesis and Richard Dawkins both say 500, without a source.^{21 22}
• Age. An age of 150 to 200 years is right for most breeds as registered units. It is wrong for the types they were cut from. Parker's tree has 23 well-supported clades "representing breed-types that existed before the advent of breed clubs and registries."⁴ Her team proposes "a two-step process of breed creation beginning with ancient separation by functional employment followed by recent selection for physical attributes."⁴ Giant size, for example, "may have been one of the earliest traits by which breeds were segregated thousands of years ago."⁴

So the fair summary is about 350 breeds, most registered in the last 150 to 200 years, carved from a few dozen older working types.

What was and was not watched

Watched:  the founding of breeds, the crosses between them, the closing of stud books, and changes in appearance over the last century.  In one case a trait was noticed as it spread.  Owners of racing whippets began reporting unusually muscular "bully" whippets within the breed, and the cause was traced in 2007.¹⁰

Not watched:  the origin of almost any variant.  Where a variant came from is inferred from which dogs, wolves and other canids carry it and from the length of DNA shared around it.  To make an origin observed, someone would have to show the variant absent in dated older samples of the same line and present later.  For the ancient variants this has been done in part with dated ancient DNA, which is the next best thing.  For the breed-era variants it has not been done at all.

Part 2: The variants that built the breeds

A surprise of dog genetics is how few genes do the work.  Adam Boyko and colleagues mapped 57 traits across 80 breeds in 2010 and found that "a small number of quantitative trait loci (≤3) explain the majority of phenotypic variation for most of the traits we studied."²³  Their explanation is how breeds were made:  "many of the modern breeds were created during the Victorian Era where novelty was a focus of selection and breeders favored the preservation of discrete mutations."  A single variant "could be placed on a variety of genetic backgrounds by crossing".²³  vonHoldt's team called it "an iterative process that drew on a limited genetic toolkit to create remarkable phenotypic diversity."³  The toolkit is small enough to list.

Body size

Dogs span a 40-fold range in body mass, more than any other mammal species.⁶  In 2007 Nathan Sutter's team found that "A single IGF1 single-nucleotide polymorphism haplotype is common to all small breeds and nearly absent from giant breeds".²⁴  IGF1 makes a growth hormone, insulin-like growth factor 1.  The locus explains about 15% of body-size variation, within one breed and between breeds.^{24 6}

Sutter's team thought the small-size marker was the new, or derived, state.²⁴  Jocelyn Plassais and colleagues settled it in 2022 with 1,431 genomes from 13 canid species, including ancient dogs and wolves.  The working variant is a single base in IGF1-AS, an antisense RNA that pairs with IGF1.⁶  The small-size allele is the old one:  "all canids (coyotes, jackals, African wolves and hunting dogs, dholes and foxes) possessed the small allele in a homozygous state, suggesting that the small allele is the ancestral state."⁶  It was found "in a 53,000-year-old Pleistocene Siberian wolf".⁶  The new allele is the large one, which "likely arose in wolves more than 53,000 years ago".⁶  Their summary:  "this major regulator of canid body size nearly vanished in Pleistocene wolves, before its recent resurgence resulting from human-imposed selection for small-sized breed dogs."⁶

Score.  Making small dogs at IGF1 required no mutation.  Breeders raised the frequency of an allele that wolves still carry.  That is sorting, which this series does not score.  The mutation in the story is the large allele, which arose in wolves long before dogs and whose molecular effect is not measured.

Five other size genes add to IGF1.  Maud Rimbault's team typed seven markers at IGF1, IGF1R, GHR, HMGA2, SMAD2 and STC2 in 500 dogs of 93 breeds, and defined the ancestral allele from wolves, red wolves and coyotes.  "We observed that the derived alleles at all markers correlated with reduced body size".²⁵  Together the markers explain "46%–52.5% of the variance in body size of dog breeds".²⁵  "Derived" here means different from the red wolf and coyote.  It is not a date.  In grey wolves, "the ancestral alleles greatly predominated," which implies that some derived alleles occur in wolves too.²⁵  A 2019 whole-genome study raised the total to 14 genes explaining "as much as 95%" of breed weight variation.²⁶  One of them makes dogs larger by breaking something:  "In large dogs, we uniquely observe a single base insertion in LCORL that causes a premature stop, truncating 600 bp of a long isoform of the protein."²⁶

Score.  The derived size alleles at the five genes are modifications, mostly unmeasured.  The LCORL insertion is a loss.

Muscle

In 2007 Dana Mosher's team traced the "bully" whippet to the gene for myostatin, a signal that limits muscle growth.  Bully whippets "carry two copies of a two-base-pair deletion in the third exon of MSTN leading to a premature stop codon at amino acid 313."¹⁰  The stop removes "63 aa from the predicted 375-aa protein."¹⁰  Dogs with one copy "are, on average, more muscular than wild-type individuals" and "are significantly faster than individuals carrying the wild-type genotype in competitive racing events".¹⁰

The deletion sits on a haplotype "identical to haplotype seven except for the deletion itself", which means it is young, and it was not found in greyhounds, the whippet's closest relative.  The authors allow that it may be "a relatively new mutation in the purebred dog population."¹⁰  The whippet breed dates from the late 1800s.¹⁰

Score.  Loss, useful to racers in one copy.  It is the clearest case in dogs of a breed-defining variant that probably arose within the breed period.  Its origin is inferred, not watched;  whippet DNA from before 1950 would show whether it is absent in the early breed.

Short legs: the FGF4 retrogenes

The dachshund, basset hound and corgi owe their short legs to a gene copy that landed in a new place.  It is the case the evolution side cites most, and Michael Behe's list of dog mutations leaves it out.²

What happened.  In 2009 Parker's team found a roughly 5,000-base insert on chromosome 18 in 175 short-legged dogs from 19 breeds.  It was absent from all 204 dogs of 41 breeds without the trait.²⁷  The insert is a retrogene:  a copy of the FGF4 gene made from its processed message and pasted back into the genome by the enzymes of LINE-1 retrotransposons.  It lacks the gene's introns.  It has "all exons … with no alterations in the coding sequence, as well as the 3’ UTR and poly-A tail characteristic of retrotransposition of processed mRNA".²⁷  FGF4 makes fibroblast growth factor 4, a signal that helps pattern the growing limb.  Parker's team proposed that "atypical expression of the FGF4 transcript in the chondrocytes may be causing inappropriate activation of one or more of the fibroblast growth factor receptors such as FGFR3."²⁷  Over-activation of the FGFR3 receptor also causes the commonest human dwarfism, achondroplasia.²⁷

In 2017 Emily Brown and Danika Bannasch at the University of California, Davis found a second copy on chromosome 12.  It "segregated with limb length and had an odds ratio of 51.23 (95% CI = 46.69, 56.20) for IVDD", intervertebral disc disease.⁷  It is common in beagles, French bulldogs and cocker spaniels, and dachshunds carry both copies.⁷

What is new and what is copied.  The protein-coding sequence is copied exactly.  So is more than the coding sequence.  When Brown's team sequenced both inserts in full, each carried the same 5′ untranslated region of the parent gene, including its transcription start site.⁷  They conclude that "it is more likely that the CpG island included in the retrogene is driving expression" than any promoter at the landing site.⁷  Parker had written in 2009 that "Neither the introns nor the upstream promoter sequences of the gene were present in the insert," and had proposed that neighbouring LINE and SINE elements "likely provide the regulatory machinery".^{27 27}  Brown's full sequence shows that part of the gene's own control sequence came with the copy.  What is new is the location and the extra dose.

Where is the copy switched on?  Mostly where the parent already is.  Parker's own test found the parent gene's message in the growing leg bones of normal fetal dogs.²⁷  Brown's team measured total FGF4 message and found "a 19.47× higher (P = 0.03) and 2.16× higher (P = 0.03) expression of FGF4 in neonatal IVD and VB," disc and vertebral body, in dogs with two copies of the chromosome 12 insert.⁷  A gene already at work in a tissue is now made at many times its normal level there.

Score.  Modification, up, by copy.  No new coding sequence was written, and no control relationship that did not exist before has been shown.  The unique functional sequence created is zero.  This differs from the score in Measuring Biological Information, which treats the retrogene as a gain by copy because it "carries no promoter of its own and is expressed from its new surroundings."  That reading followed Parker 2009, before Brown's full sequence.  If future work shows the copy switched on where FGF4 is normally silent, the score should move to gain.  Behe's own 2010 definitions place "the simple duplication of features, without divergence of activity" under modification, so his scheme agrees.²⁸

How old.  Both copies are older than the breeds.  Parker's team found a single shared haplotype around the chromosome 18 copy in all 19 breeds, "before the division of early dogs into modern breeds."²⁷  Brown's team argues from depictions of short-legged dogs "dating back over 4,000 y".⁷  That is an inference from pictures, not a molecular date.  But neither copy is a wolf variant.  Kevin Batcher and colleagues searched 1,125 dog genomes and 141 wolves in 2020 and found five more FGF4 copies, for seven in all.  "The FGF4 retrocopies identified in domestic dogs were identical to domestic dog FGF4 haplotypes, which are distinct from modern wolf FGF4 haplotypes, indicating that these retrotransposition events likely occurred after domestication."¹¹  One of the new copies turned up in three pit bull terrier mixes, all treated for disc disease.¹¹

Such copies are common.  In 293 canid genomes Batcher's team later counted 1,911 polymorphic gene copies, "1236 of which appeared identical to their parent genes", about 54 per dog and 13 times the human rate.^{29 29}  New ones form in dogs "at around four retroCNVs per 100 yr".²⁹  Breeders picked up two that changed the legs.

Coat

In 2009 Edouard Cadieu and colleagues found that three variants explain most coat types:  "combinations of alleles at … three genes … explain the observed pelage phenotype of 95% of dogs sampled, which include 108 of the ~160 American Kennel Club (AKC)–recognized breeds."⁸

• RSPO2, wire hair and "furnishings" (moustache and eyebrows). A 167-base insertion in the untranslated tail of the gene's message. Dogs with it had "a threefold increase in RSPO2 transcripts."⁸ Score: modification, up. Jean Lightner, writing for Answers in Genesis, agrees: "the RSPO2 mutation was not a loss-of-function mutation, since more mRNA was detected than normal."¹
• FGF5, long hair. "a highly conserved Cys is changed to Phe (Cys95→Phe)".⁸ FGF5 stops the hair-growth phase, and in mice and cats long hair comes from broken FGF5. Score: loss, by analogy; not assayed in dogs.
• KRT71, curl. One amino-acid change, Arg151Trp, in a hair keratin. Score: modification, unmeasured.

"None of the mutations we observed were found in three gray wolves or the short-haired dogs".  The authors' "finding of identical haplotypes surrounding the variants in all dogs displaying the same coat type suggests that a single mutation occurred for each trait and was transferred multiple times to different breeds through hybridization."⁸  Three wolves is a small sample, but the three variants are shared across unrelated breeds and are not seen in wild canids.  They arose in dogs and are older than most breeds.

Hairlessness in the Mexican and Peruvian hairless dogs and the Chinese crested is "a frameshift mutation within the FOXI3 coding sequence", and it costs teeth as well as hair.³⁰  Score:  loss.  Dominant black coat comes from a three-base deletion in a beta-defensin gene, CBD103, whose product acts on the pigment receptor.  It is at least 12,779 years old at the lower 95% limit, and it later passed from dogs into North American wolves.^{31 31}  Score:  modification, unmeasured, pending the binding work that would say whether the deletion creates a new contact.

Skull

Short-faced breeds such as the pug and the bulldog carry two variants read here.  In BMP3, a bone-growth signal, a phenylalanine at position 452, where the gene family "nearly invariably" has an aromatic amino acid, is changed to leucine, a change "predicted … likely damaging".  It is "nearly fixed among extreme brachycephalic breeds."^{32 32}  In SMOC2, a fragment of a LINE-1 element sits inside an intron and "promotes the utilization of cryptic splice sites".  Dogs with two copies had "an ∼5-fold reduction in total SMOC2 mRNA".^{33 33}  The fragment is on 91.5% of short-faced dogs' chromosomes and 2.1% of others, and accounts for 36% of face-length variation.³³

Score.  BMP3:  modification, probably down.  SMOC2:  modification, down.  An inserted element is not automatically a gain;  here it lowers the output of a working gene.

Starch

Dogs digest starch better than wolves because many carry extra copies of AMY2B, the pancreatic amylase gene.  Maja-Louise Arendt's team showed that serum amylase activity rises with copy number.³⁴  In 51 wolves, 49 had the usual two copies, one had one and one had three.  In 392 dogs the median was 10, with a range of 1 to 22.^{9 9}  High copy numbers track the spread of farming, and dingoes and Arctic dogs mostly lack them.⁹  Ancient DNA fixes the timing:  expansion is seen "from as early as the 7th millennium cal. BP", not yet fixed in early farming dogs, and "did not become widespread until several thousand years after the first appearance of starch-rich agricultural lifestyles."^{35 36}

Score.  Modification, up, by copy.  Zero unique sequence.  It predates the breeds by thousands of years on radiocarbon dates, and it happened in dogs, not wolves.

Part 3: Old variation or new mutation?

The table sorts each variant by where it came from.  "Wolf" means the variant is found in wolves or other wild canids.  "Dog, pre-breed" means it is absent or rare in wolves but shared across unrelated breeds, so it arose in dogs before the breeds split.  "Breed era" means it probably arose after about 1800.

Most breed-defining variants predate the breeds.  That supports the creationist description of breeding as selection among variants already present, and Parker's two-step history.

Most of them are not wolf variants.  Only the IGF1-AS small allele is shown in wild wolves, and CBD103 may be.  The leg-shortening copies, the coat variants, the hairless allele and the amylase expansion arose in dogs after domestication.  So the catalogue line "almost all of it from variation the wolf already carried" is wrong.  The accurate line is "almost all of it from variation dogs already carried before the breeds," and much of that variation was made by mutation in dogs.

At least one arose during the breed period, and new copies keep arising.  The whippet deletion is the best case, and the rare FGF4 copies show the event behind the dachshund's legs still happening.

Part 4: The cost

Inbreeding

A breed is closed by definition, so every breed inbreeds.  Danika Bannasch and colleagues estimated inbreeding from the genotypes of 49,378 dogs in 227 breeds in 2021.  "The average inbreeding based on genotype across 227 breeds was Fadj = 0.249".¹²  They put it in context:  "the breeding of two first cousins produces F = 0.0625, two half siblings F = 0.125 and two full siblings or parent-offspring F = 0.25."¹²  "Strikingly few breeds (N = 12) had low inbreeding values (< 0.10)."¹²  Insurance records showed that "Smaller less inbred breeds were healthier than larger more inbred breeds."¹²

The load of harmful variants

Two studies measured how many harmful variants dogs carry compared with wolves.  Fernando Cruz, Carles Vilà and Matthew Webster found in 2008 that the ratio of protein-changing to silent variants "was around 50% greater for SNPs found in dogs than in wolves".³⁷  Clare Marsden and Kirk Lohmueller's team used 90 whole genomes in 2016:  "on average, dogs have 2–3% higher genetic load than gray wolves."¹³  They traced it to the population bottlenecks of domestication and breed formation, "rather than recent inbreeding," and found that the regions breeders selected hardest carried extra baggage:  "regions of the genome implicated in selective sweeps are enriched for amino acid changing variants and Mendelian disease genes."¹³  Their advice cuts against the whole method of breed standards:  "maintaining a large population size, rather than just avoiding inbreeding, is a critical factor for preventing the accumulation of deleterious variants."¹³

Two to three percent is a real cost, not a collapse.

Disease

The Online Mendelian Inheritance in Animals database at the University of Sydney lists more inherited traits for dogs than for any other animal, 1,019.  Of 594 dog variants with a known likely causal effect on a single-gene trait, 541 are for single-gene diseases.¹⁴  That ratio has a bias built in:  veterinary genetics looks hardest for variants that make dogs sick.  But the absolute number is the point.  More than five hundred known disease variants ride in a species that breeders have kept in small closed pools.

Some costs come with the breed-defining variant itself.  The chromosome 12 FGF4 copy that shortens beagles also raises the odds of disc disease 51-fold, and Brown's team notes that "the number of animals with this intervertebral disc disease mutation across the globe is in the millions."^{7 7}  The FOXI3 allele that makes a hairless dog also takes its teeth.³⁰  The SMOC2 variant that shortens the face is part of the skull form that brings "respiratory distress and ocular trauma" to short-faced breeds.³³  Creation Ministries International made the same point in 2010, under the title "Parade of mutants".³⁸

Part 5: The silver fox experiment

Dogs were domesticated before anyone was taking notes.  Foxes were not.  In 1959 the Soviet geneticist Dmitry Belyaev began breeding silver foxes, a colour form of the red fox, for tameness alone at the Institute of Cytology and Genetics in Novosibirsk.¹⁵  Lyudmila Trut, who headed the research group, described it in 1999:

• Founders. "He began with 30 male foxes and 100 vixens, most of them from a commercial fur farm in Estonia. The founding foxes were already tamer than their wild relatives."¹⁵
• Selection. "typically not more than 4 or 5 percent of male offspring and about 20 percent of female offspring have been allowed to breed."¹⁵
• Response. "By the tenth generation, 18 percent of fox pups were elite; by the 20th, the figure had reached 35 percent. Today elite foxes make up 70 to 80 percent of our experimentally selected population." Elite foxes are "eager to establish human contact, whimpering to attract attention and sniffing and licking experimenters like dogs."¹⁵ This took "40 years and 45,000 foxes".¹⁵
• Side effects. "In our foxes, novel traits began to appear in the eighth to tenth selected generations": white patches, floppy ears, curled tails.¹⁵

Trut's team read this as selection on behaviour that "may accelerate phenotypic variations."³⁹

In 2020 Kathryn Lord, Greger Larson, Raymond Coppinger and Elinor Karlsson challenged the story.  The Russian fur farms had been stocked from Prince Edward Island in Canada, where foxes had been bred in captivity since the late 1800s and already selected for coat and temperament.  "In reality, the experiment started with a fox population from eastern Canada that had been captive and purpose-bred since the late 1800s".¹⁶  "Mitochondrial DNA analyses of 24 Farm-Fox Experiment foxes showed 100% of haplotypes derived from Canadian foxes".¹⁶  A 1922 photograph of Prince Edward Island farm foxes being hand-fed is their evidence that friendliness and white spotting came first.¹⁶  Their conclusion:  "This history reframes the behavioral selection in the Farm-Fox Experiment as selection on standing genetic variation and explains the rapid behavioral change (within ten generations)."¹⁶

The genome agrees.  Anna Kukekova's team sequenced tame, aggressive and ordinary farm foxes in 2018 and found "103 regions with either significantly decreased heterozygosity in one of the three populations or increased divergence between the populations."⁴⁰  Lower heterozygosity means that selection pushed existing variants toward fixation.  No new mutation was reported.

What the foxes show.  Sixty years of hard selection on a trait with real genetic variation produced a dramatic change in behaviour, and the change drew on variation already in the founders.  Trut herself estimated that "about 35 percent of the variations in the foxes' defense response to the experimenter are genetically determined."¹⁵  It is the strongest watched case for the creationist description of domestication as sorting, and it contradicts nothing the mainstream claims about a short, intense program.

Part 6: What artificial selection shows

Darwin's use of it

Darwin began the Origin with pigeons, cattle and dogs.  His summary:  "The key is man's power of accumulative selection: nature gives successive variations; man adds them up in certain directions useful to him."⁴¹  He guessed wrong about dogs.  "I think it highly probable that our domestic dogs have descended from several wild species."⁴¹  Genome data show one grey-wolf-like source.³  The dog case is stronger for his argument than he knew.

The creationist and design reading, in its own words

The young-earth organisations read the breeds as sorting and loss within a created kind.

• Creation Ministries International, 2010. The wolf-like ancestor "had the genetic diversity that allowed people to breed dogs as different in size as the Chihuahua and the Great Dane." Selection means "genes have been lost from the population—i.e. there has been a loss of genetic information, the opposite of what microbe-to-man evolution needs in order to be credible." And "Other breeds, as will be shown, are the result of downhill mutations."^{38 38 38}
• Answers in Genesis, 2022. Ken Ham: "the original dog kind that came off the ark must have had a massively heterozygous genome," and "we have selected for mutations (basically “mistakes”) that we prefer."^{42 42} Tom Hennigan in Answers magazine: "The only reasonable explanation for such complex, elegant, preprogrammed genetic systems is that they were designed that way."²¹
• Institute for Creation Research, 2009. Brian Thomas, on the coat genes: the first dogs "would have had more genetic information than their descendants, with built-in potential for producing variations. And wolves seem to still have just that."⁴³
• Jean Lightner, 2009. On the same coat genes, eight days later: "Theoretically, God could have created several alleles for different coat characteristics. In these three genes, I think it is more likely that these alleles have arisen through mutation. First, it doesn’t look like these alleles exist in other canids (wolves, coyotes, etc.)." And: "On a molecular basis, all of these changes are degrading the genome, not building it."^{1 1}
• Michael Behe, 2019. "we see that the great majority of dog mutations unwittingly selected by us humans are very likely to be damaging, degrading, or outright loss-of-FCT ones."² Behe does not say the variants were created. He accepts that they are "hidden random mutations" and argues that they are mostly broken genes.²

Checked against the primaries, these hold up unevenly.

• Thomas's claim about wolves is contradicted by the paper he was reporting. Cadieu's team found none of the three coat variants in wolves.⁸ Lightner read the same paper correctly.
• Behe's list is mostly right. It overstates twice. He calls all three coat variants "likely damaging", but the RSPO2 variant triples the gene's message.^{2 8} He writes that "All the mutant genes decrease the size of a dog", following Rimbault's derived-allele calls.² At IGF1-AS, the most important size locus, the small allele turned out to be the ancestral one, and a truncating LCORL variant makes large breeds larger. He also cites Marsden's genome-wide load study for "the great majority of dog mutations unwittingly selected by us humans", but that study measured harmful variants riding along with selected regions, not the molecular class of the selected variants themselves.¹³
• CMI's size illustration, a medium-sized ancestor carrying both a large and a small allele, matches what Plassais found at IGF1-AS.
• "Loss of information" does not describe the amylase copies, the FGF4 copies or the RSPO2 insertion. They are extra doses of existing genes. They wrote no new sequence, which supports the unique-sequence argument, but they are not losses.

The mainstream reply, at full strength

The mainstream reply has four parts.

1. Two hundred years is nothing. Breeding in small closed pools is a short, strongly directed experiment, and the big visible effects a breeder can use in a few generations are more often breaks than builds. Behe's own First Rule, which his critics largely accept as a statement about rates, predicts exactly this; see Measuring Biological Information.
2. New mutations did arise and were used. The FGF4 copies, the coat variants, the hairless allele and the amylase expansion all arose in dogs, after domestication, and were then picked up by breeders. Lightner concedes the coat case. New FGF4 copies are still arising.¹¹
3. Limits reflect small closed pools, not a barrier. Marsden's team traced the load to small population size and advised "maintaining a large population size".¹³
4. Long selection in large populations keeps going. The standard example is corn.

The Illinois corn experiment

In 1896 Cyril Hopkins at the University of Illinois began selecting corn for high and low oil content in the kernel.  The top or bottom 20% of ears were chosen every year.  By generation 89, "the populations had changed from 4.7% oil to 19.3% in IHO and 1.1% in ILO", the high and low lines.  Selection "continues to this day for IHO with no sign of a plateau in response."^{17 17}  That is about four times the starting oil content, far outside the range of the founding variety.

The limits were in the downward lines.  The low-oil and low-protein lines "have probably reached lower biological limits for these traits owing to their poor germination frequencies and lack of change in recent cycles".¹⁸  Low-oil selection was stopped at generation 89 "because of poor viability".¹⁷

Where did a century of response come from?  The mainstream is split.  Bruce Walsh argued in 2004 that "the majority of response in IHO and ILO was due to new mutation," because the lines' effective size was only about ten and the starting variation would have been used up quickly.  Peter Keightley noted that responses fed by new mutation tend to jump, while those fed by standing variation are smooth, like the Illinois lines.  Cathy Laurie's team, who mapped more than 50 loci, concluded:  "Unfortunately, the genetic architecture results do not seem to have a direct bearing on the importance of new mutations."^{17 17}  Stephen Moose and colleagues found the remaining diversity "too high to be explained by mutation".¹⁸

So the corn does not settle the question of new information.  It does settle a narrower one.  A claim that artificial selection reaches a general ceiling within a century is false.  Upward selection kept working for over a hundred generations, even in small lines.  Where limits appeared, they were physiological floors, like a kernel with no oil, and they came with sick plants.  The dog breeds' costs look like the second kind:  the breeds at the extremes of size and face shape are among the sickest.^{12 12}

Part 7: The ledger

The series asks of each organism how many mutations were tried and what they bought.⁴⁴

What it bought.  Adaptive variants shown to have arisen in the breed period:  one, the whippet myostatin deletion, a loss.  New FGF4 copies of the period:  several, with no benefit shown.  Gains of function:  none shown.  Unique functional sequence created:  none shown.

Why the zero is a weak test.  Four reasons, each against reading much into it.  First, the dog-generation count is a ceiling.  The population was smaller in 1800, and about three quarters of today's dogs are free-roaming, where no breeder selects anything.⁴⁵  Second, the relevant search is inside each breed, and breeds are small closed pools whose sizes were not found for this article.  That cell is left empty rather than guessed.  Third, breeders select looks, and the fastest way to change a look is often to break or weaken something, as Behe's rule predicts.  Fourth, the variants were found by mapping traits breeders fixed, so a gain that did nothing visible would not be found.  The honest reading is that about 10¹² mutations in the world's dogs produced no documented gain under selection aimed at appearance.  That is a real data point for the thesis that gains are rare.  It is not a measured rate.

What the ledger cannot price.  Several changes are unmeasured and could hide gains:  the KRT71 and CBD103 variants, the five derived size alleles, the dog behaviour variants not read for this article, and the 97 retrocopies whose frequencies differ sharply between breed groups.²⁹  Any of them could turn out to be a gain.  So the ledger is a lower bound on what dogs produced, and that cuts against the young-earth side.

The foxes.  45,000 foxes over 30 to 35 generations produced one of the most striking behavioural changes on record.¹⁵  No new variant has been tied to it.

Objections and responses

"Dog breeds show evolution in action.  If breeders did this in two centuries, nature can do anything in millions of years."  Dawkins and Neil deGrasse Tyson both make this argument.²²  Breeds show that selection changes shape fast.  None of the breed-defining variants read here is a new function, so the extrapolation from shape to new organs is exactly what breeds do not test.

"Breeding is all loss of information."  This overstates the creationist case.  The amylase copies, the FGF4 copies and the RSPO2 insertion increase the output of working genes.  They wrote no new sequence, which supports the argument that nothing new was made, but they are not losses.  And Lightner is right that many variants came from mutations in dogs, not from created wolf diversity.

"The FGF4 retrogene is a new gene, so evolution made new information."  The copy's coding sequence is identical to the parent's, and it carried part of the parent's control region.  What changed is its location and the dose of FGF4.  Evidence that the copy is active where FGF4 is normally silent would move it from modification to gain.

"The dates are circular.  The variants look old only because evolutionists assume long ages."  Some of the evidence does not depend on dates at all.  That the FGF4 copies sit on dog-specific haplotypes, and that the coat variants are missing in wolves, is a matter of which animals carry them.  The ages do depend on dating:  the 53,000-year-old wolf with the small IGF1-AS allele and the Neolithic amylase dogs rest on radiocarbon.  Guitard's creationist model places the domestication of dogs between about 2521 and 2200 BC.²⁰  That model must fit the wolf-held IGF1-AS variants, the post-domestication FGF4 copies, and an amylase expansion that followed farming by thousands of radiocarbon years into a few centuries.  No creationist paper found attempts that.

"Breeders hit limits, which proves that kinds are fixed."  The limits dogs show are the limits of small closed pools and of physiology.  Corn under selection for high oil has not plateaued in over a century.  A limit shown for dogs is not a limit shown for life.

"Two hundred years is too short to expect a gain.  The verdict is unfair to evolution."  This is correct.  Dogs are a weak test of the origin of new function.  The organism benchmarks with larger populations, such as the Long-Term Evolution Experiment and Plasmodium falciparum, carry that load.

"The silver foxes prove that selecting tameness produces the domestication syndrome."  The founders came from long-bred Canadian farm stock that already had white spotting and unusual friendliness, and the response was fast because the variation was already there.

"Behe is refuted, because the retrogene is a gain he left out."  By the scoring used here it is not a gain, though he should have addressed it.  His conclusion survives in weaker form:  most breed-defining variants are losses or down-regulations, and the rest are extra copies.

Issues and unanswered questions

• Where the FGF4 copies are active in the embryo is unknown. That single fact decides between modification and gain.
• Old whippet DNA from the early 1900s would turn the myostatin origin from inferred to observed.
• Breed population sizes are needed for a per-breed ledger. None was found.
• Supplemental Table 7 of Rimbault 2013, which gives the frequencies of the derived size alleles in grey wolves, was not read.
• Three papers were read only as abstracts: Axelsson 2013 on starch, Drögemüller 2008 on hairlessness, and Cruz 2008 on load. Candille 2007, which would say whether the black-coat deletion creates a new binding contact, was not read.
• Behaviour. The variants linked to dog friendliness, which Behe cites, were not read for this article.
• A creationist computation is missing. With about 20 new mutations per dog per generation and the observed rate of new gene copies, how many FGF4-type events are expected in 4,500 years of dogs? Lightner's position, that the coat alleles arose by mutation after the Flood, invites that calculation.

Verdict

Dog breeds are the best-documented case of fast, directed change in a mammal.  About 350 breeds were registered in 150 to 200 years, and the genome confirms the stud books.  The variants that made them are few and mostly older than the breeds.  Those read here are losses, weakened genes and extra copies of working genes.  None is shown to be a new function, and the one breed-era variant is a broken gene.  The creationist and design readings are right about that, and right about the cost:  inbreeding at the level of a sibling mating, more harmful variants than wolves, and hundreds of known disease genes.

They are wrong in two places.  The material was not mostly wolf variation.  Much of it came from mutations in dogs after domestication, as Lightner concedes, and new gene copies are still arising.  And "loss of information" misdescribes the copies that turned genes up.  Those added no new sequence and removed nothing.  The ages of the older variants rest on radiocarbon dates that the young-earth model has not answered.

For the series' thesis, dogs add a modest data point:  roughly 10¹² mutations in two centuries yielded no documented gain, under selection that was not looking for one.  Dogs show what selection on standing variation does quickly.  They do not test what mutation builds slowly.

Sources