Variants in FtsJ RNA 2′-O-Methyltransferase 3 and Growth Hormone 1 are associated with small body size and a dental anomaly in dogs

Significance

Through the study of a dental anomaly we identified a locus strongly associated with body size in the Shetland Sheepdog. Within this locus are variants in two genes: a substitution in FtsJ RNA 2′-O-Methyltransferase 3 (FTSJ3) and a splice donor insertion in Growth Hormone 1 (GH1). We demonstrated that the GH1 variant causes an abnormal splicing pattern that is also observed in dominant forms of human pituitary dwarfism. Interestingly, the FTSJ3 variant is estimated to have the greatest impact on height and weight and this gene has not been previously characterized in body size traits. Both derived alleles are found in high frequencies in very small “toy” breeds but are entirely absent from larger breeds.

Abstract

Domesticated dogs show unparalleled diversity in body size across breeds, but within breeds variation is limited by selective breeding. Many heritable diseases of dogs are found among breeds of similar sizes, suggesting that as in humans, alleles governing growth have pleiotropic effects. Here, we conducted independent genome-wide association studies in the small Shetland Sheepdog breed and discovered a locus on chromosome 9 that is associated with a dental abnormality called maxillary canine-tooth mesioversion (MCM) (P = 1.53 × 10⁻⁷) as well as two body size traits: height (P = 1.67 × 10⁻⁵) and weight (P = 1.16 × 10⁻⁷). Using whole-genome resequencing data, we identified variants in two proximal genes: FTSJ3, encoding an RNA methyltransferase, and GH1, encoding growth hormone. A substitution in FTSJ3 and a splice donor insertion in GH1 are strongly associated with MCM and reduced body size in Shetland Sheepdogs. We demonstrated in vitro that the GH1 variant leads to exon 3 skipping, predicting a mutant protein known to cause human pituitary dwarfism. Statistical modeling, however, indicates that the FTSJ3 variant is the stronger predictor of MCM and that each derived allele reduces body size by about 1 inch and 5 pounds. In a survey of 224 breeds, both FTSJ3 and GH1 variants are frequent among very small “toy” breeds and absent from larger breeds. Our findings indicate that a chromosome 9 locus harboring tightly linked variants in FTSJ3 and GH1 reduces growth in the Shetland Sheepdog and toy breed dogs and confers risk for MCM through vertical pleiotropy.

In dogs, maxillary canine-tooth mesioversion (MCM) describes an upper canine tooth that is displaced forward toward the nose, also known as a lance canine (Fig. 1A, compare to normal dentition in Fig. 1B) (1, 2). One or both maxillary canines may be affected. MCM can cause traumatic occlusion, ulceration of the upper lip, and/or periodontal disease and may require extraction or orthodontic repositioning (1). MCM is rarely observed outside of the Shetland Sheepdog (2, 3), a breed that also has a high prevalence of hypodontia, a condition characterized by one or more congenitally missing teeth (4).

Fig. 1.

MCM in the Shetland Sheepdog. (A) The maxillary canine (upper C) of an affected dog is closer to the third incisor (I3) and rostral to the mandibular canine (lower C). The long axis of the affected canine (indicated by the vertical dashed line) is rostral compared to B normal dentition in an unaffected dog. (C) Manhattan plot of MCM GWAS using 39 cases and 39 controls. The −log₁₀ P values for 117,053 SNPs are plotted on the y axis against chromosome position. The black horizontal line represents the threshold for Bonferroni significance. The position and P value of the lead SNP are included.

The Shetland Sheepdog is a small breed with a standard height of 13 to 16 inches at the shoulder. Derived alleles of IGF1, GHR, HMGA2, IGF1R, SMAD2, and STC2 explain about half of reduced body size observed across dog breeds (5). “Small” alleles of the first four aforementioned genes have been observed in Shetland Sheepdogs in various genotypic combinations, indicating the alleles are not fixed in the breed (5). In general, smaller breeds tend to have more derived alleles at these loci than larger breeds, but even the smallest “toy” breeds are not fixed across these loci (5).

We undertook a genome-wide association study (GWAS) and whole-genome resequencing to identify genetic risk factors causing MCM and uncovered a strong correlation with body size. We identify variants in Growth Hormone 1 (GH1) and FtsJ RNA 2′-O-Methyltransferase 3 (FTSJ3) that are strongly associated with MCM, height, and weight and demonstrate the functional consequence of a GH1 splice-site insertion. Our data suggest that the locus harboring GH1 and FTSJ3 variants is a major determinant of body size in Shetland Sheepdogs and toy dog breeds and confers risk for MCM through vertical pleiotropy.

Results

MCM Is Associated with Single-Nucleotide Polymorphisms on Chromosome 9.

We obtained genetic material and phenotypic information from 86 Shetland Sheepdogs having MCM (51 bilateral, 33 unilateral, and 2 unknown laterality) and 144 Shetland Sheepdogs with correctly positioned canines (SI Appendix, Table S1). There was no statistical difference between males and females among cases (51 male, 35 female; P = 0.10). Unilateral cases were more likely to have the left canine affected (23 left, 8 right; P = 0.01). Bilateral cases were more likely to have hypodontia (n = 27) than unilateral cases (n = 11) (P = 0.01). Fourteen dogs unaffected by MCM were also missing teeth. The teeth most commonly reported as missing were the maxillary second premolars, with no side preference.

We conducted a GWAS for MCM using 39 cases, 39 controls, and 117,053 single-nucleotide polymorphisms (SNPs) after filtering and identified a single region of association exceeding Bonferroni significance on chromosome 9 (12753481, P = 1.53 × 10⁻⁷; Fig. 1C and SI Appendix, Table S2). There was no evidence of genomic inflation (lambda = 1.03). Sixty-seven percent of cases were homozygous for the risk allele of the lead SNP (chr9:12753481), while 31% were heterozygous. The risk allele was also present among the control dogs: 8% were homozygous risk and 72% were heterozygous, likely indicating a complex pattern of inheritance. Using the lead SNP, we calculated regional linkage disequilibrium (LD) (r² > 0.5) to define a conservative 1.9-Mb candidate interval from 11.6 Mb to 13.5 Mb on chromosome 9 harboring about 30 genes.

FTSJ3 and GH1 Harbor Candidate Causal Variants.

We initially searched for candidate causal variants using a VCF file generated from 722 dogs, including 1 Shetland Sheepdog having bilateral MCM, 2 Shetland Sheepdogs with unknown dentition, 665 dogs of various pure and mixed breeds, and 54 wild canids. The affected Shetland Sheepdog was homozygous for the risk allele at the lead SNP and for the associated haplotype across the candidate interval. The affected dog possessed no unique variants, even when excluding the other two Shetland Sheepdogs from the analysis, indicating that the causal variant is likely a polymorphism. Manual scanning of the critical interval in IGV revealed no major structural changes unique to the Shetland Sheepdog.

We then used whole-genome resequencing data from the aforementioned Shetland Sheepdog with bilateral MCM to identify variants that were 1) homozygous, 2) present within the coding regions or splice sites of genes within our critical interval, and 3) absent from the genomes of five Collies, a medium- to large-sized, genetically similar ancestor of Shetland Sheepdogs (6) in which MCM has never been described. We identified two synonymous SNPs, one nonsynonymous SNP, and one splice-site variant and considered the latter two to be potentially deleterious. A substitution, g.11775131T > C, occurs in FTSJ3 (FTSJ3^mut, hereafter) and predicts a p.Lys797Glu missense in a highly conserved position of the canine protein (XP_005624307; SI Appendix, Fig. S4). This change is categorized as “probably damaging” by in silico programs PANTHER (1500) and PolyPhen2 (0.979). GH1 harbors a single base insertion in the splice donor of exon 3: g.11833343_11833344insA (GH1^mut, hereafter). We also further investigated a g.14561174A > G substitution in AXIN2, located 1 Mb outside of our critical interval, because mutations in the orthologous human gene result in dental anomalies (7). The AXIN2 variant predicts a p.Lys103Arg missense variant, a change categorized as probably damaging by PANTHER (456) and benign by PolyPhen2 (0.02). While we cannot exclude the possibility that a noncoding variant is contributing to MCM, we focused the remainder of the study on these three potentially deleterious coding variants.

We genotyped all three candidate variants in 78 MCM cases and 125 controls. FTSJ3^mut and GH1^mut were strongly associated with MCM (P = 8.2 × 10⁻¹⁶ and 1.7 × 10⁻¹³, respectively), while the AXIN2 variant showed a weaker association (P = 8.5 × 10⁻⁷). We thus eliminated AXIN2 as a candidate causal gene.

We investigated the frequencies of FTSJ3^mut and GH1^mut using publicly available genomes from 1,049 dogs representing 224 different breeds (SI Appendix, Table S3). Across all breeds, the derived alleles had frequencies of 4.5% and 3.7%, respectively. In addition to the Shetland Sheepdog, these alleles were found only in the following toy breeds: Affenpinscher, Biewer Terrier, Chihuahua, Miniature Pinscher, Papillion, Pomeranian, Toy Poodle, Cavalier King Charles Spaniel, Toy Fox Terrier, and Yorkshire Terrier.

MCM Is Highly Correlated with Body Height and Weight.

Because growth hormone is necessary for normal body growth (8), we looked for a correlation between body size and MCM. Affected dogs were significantly shorter and weighed less than controls (P = 4.6 × 10⁻⁶ and 5.6 × 10⁻¹¹, respectively) (Fig. 2 A and B). As female Shetland Sheepdogs are often smaller in height and weight compared to males (P = 0.007 and 0.22, respectively in our cohort), we confirmed that the correlations were significant within each sex (SI Appendix, Fig. S2).

Fig. 2.

Height and weight correlate with MCM. Boxplots illustrate (A) lower heights in affected dogs (n = 63, mean = 13.97 inches) compared to controls (n = 104, mean = 14.93 inches) and (B) reduced weights in affected dogs (n = 62, mean = 16.20 pounds) compared to controls (n = 99, mean = 21.97 pounds). Two-sample t test P values are reported above. Manhattan plots for (C) height and (D) weight GWASs using 34 cases and 37 controls. The −log₁₀ P values for 117,053 SNPs are plotted on the y axis against chromosome position (CanFam3.1). The black horizontal line represents the threshold for Bonferroni significance. The positions and P values of the lead SNPs are included.

Because of the strong correlation between MCM and body size, we conducted another GWAS with height, weight, and sex as covariates (34 cases vs. 37 controls). The chromosome 9 signal was diminished but persisted as the most significant association (chr9:12753481, P = 1.04 × 10⁻⁶) (SI Appendix, Fig. S3A). Independent GWASs for height and weight indicate the locus on chromosome 9 affects both phenotypes (P = 1.67 × 10⁻⁵ and 1.16 × 10⁻⁷, respectively) (Fig. 2 C and D). The impact of this locus on either phenotype cannot be detected when we include MCM as a covariate, consistent with a shared cause (i.e., pleiotropy) (SI Appendix, Fig. S3 B and C) (9).

GH1^mut Causes Exon 3 Skipping.

To further assess GH1 as a candidate gene, we sought to determine if the splice-site insertion alters gene splicing. We expressed mutant and wild-type gene fragments in HEK293 cells and evaluated splicing patterns through complementary DNA (cDNA) sequencing. Cells carrying the wild-type fragment produced a primary band at 475 bp. The mutant cells yielded the wild-type band and a brighter, smaller band of 358 bp (Fig. 3). Sequencing confirmed that the larger band represented normal splicing of exons 2 through 5. The 358-bp band is a spliceoform in which all 117 bp of exon 3 are skipped, predicting a protein lacking 39 of 216 amino acids. Proteins resulting from GH1 transcripts without exon 3 have been identified in human patients with growth hormone deficiency type II and shown to act in a dominant negative manner (8). Both cell lines also yielded a small amplicon of 196 bp that sequencing confirmed to be an isoform in which exons 3 and 4 are skipped (10).

Fig. 3.

Amplification of GH1 cDNAs. Agarose gel electrophoresis of a minus reverse transcriptase control (−RT, lane 2), empty vector control (EV, lane 3), wild-type GH1 cDNA (GH1⁺, lane 4), and GH1 cDNA harboring the splice-site insertion (GH1^mut, lane 5). The primary wild-type band at 475 bp (1), mutant band at 358 bp (2), and faint wild-type band at 196 bp (3) were each extracted and characterized via Sanger sequencing. The splicing pattern of each isoform for exons 2 to 5 is depicted below the gel image.

FTSJ3^mut Impacts Body Size in an Additive Fashion.

FTSJ3 and GH1 are separated by ∼58 kb on chromosome 9 and their alleles are predictably in high LD (D′ = 0.99). We observed three haplotypes in our population (Fig. 4A). The absence of an FTSJ3^mut-GH1⁺ haplotype suggests that the GH1 mutation occurred first and that recombination between chr9:11775131 and chr9:11833343 is rare. The FTSJ3⁺-GH1^mut haplotype was uncommon but found significantly more often in dogs having normal dentition (P = 0.04).

Fig. 4.

Understanding the roles of FTSJ3, GH1, and IGF1 in MCM and body size. (A) Odds ratios (OR), 95% CIs, and Fisher’s exact two-tailed P values are calculated for each of the three observed haplotypes for MCM. Haplotypes with FTSJ3⁺ are protective, while the haplotype with both FTSJ3^mut and GH1^mut confers risk. (B) Height and weight averages (calculated using all available data) and MCM risk are plotted for the four most frequent genotypic combinations of FTSJ3 and GH1. One copy of each derived allele (n = 74) markedly reduces height and weight compared to wild type (n = 40; P = 0.003 and 0.0002, respectively). Compared to heterozygotes, homozygosity for the GH1^mut allele alone (n = 23) does not significantly alter height (P = 0.3) or weight (P = 0.7). Among GH1^mut homozygotes, adding a second copy of FTSJ3^mut (n = 67) significantly reduces height (P = 0.0002) and weight (P = 7.8 × 10⁻⁵), illustrating that GH1 is not solely responsible for the variation in body size and disease risk attributed to this locus. (C) AIC values are shown for different genotypic and/or phenotypic predictive models of MCM. The small AIC value for FTSJ3 + weight (bold) indicates a more parsimonious fit to the disease data. (D) Probability of disease plotted against FTSJ3 genotypes at two extremes: heavy (27 pounds) and light (12 pounds) shows the importance of weight on risk for MCM. (E) Probability of disease for all combinations of FTSJ3 and IGF1 genotypes shows a greater role for IGF1 among FTSJ3 heterozygotes.

To determine if both alleles or just GH1^mut impact body size, we investigated height and weight within two subpopulations that were fixed at one locus but variable at the other (Fig. 4B). We found that the heights and weights of dogs having only one risk allele at FTSJ3 were significantly greater than those having two copies, suggesting that FTSJ3^mut reduces body size in an additive fashion. When FTSJ3^mut is neutralized, we found no significant differences in the heights and weights of dogs having one vs. two GH1^mut alleles. This is consistent with a dominant negative effect of the GH1 mutant protein. We did not have the genotypic combinations in our population to enable us to study the impact of having at least one GH1^mut allele vs. homozygous wild type.

FTSJ3 and IGF1 Are Highly Predictive of MCM.

To evaluate genotypic and phenotypic contributions to MCM, we performed statistical modeling using Akaike information criterion (AIC), where smaller numbers indicate the best and most parsimonious fit to the disease data (Fig. 4C). Consistent with our association statistics, we found that FTSJ3 is a better predictor of MCM than GH1 and that weight is a better predictor of MCM than height. While FTSJ3 alone is a better predictor of MCM than weight alone, the best predictive model includes both FTSJ3 and weight. In homozygosity, FTSJ3^mut confers a strong risk for MCM, regardless of weight. However, in heterozygosity, weight governs risk, such that a lightweight dog has a much higher risk than a heavy dog (Fig. 4D).

We then considered the impact of alleles of other known body size genes segregating in the Shetland Sheepdog: IGF1, GHR, HMGA2, and IGF1R. Using associated SNPs from the BeadChip and our GWAS cohort (39 cases vs. 39 controls), we found that the derived allele of IGF1 was significantly overrepresented among cases (P = 0.04), and even more so when we only consider dogs heterozygous for FTSJ3^mut (P = 0.016). The derived allele of IGF1R was significantly overrepresented among controls (P = 0.04) but was not significant in the heterozygous subset. We detected no significant differences in the allele frequencies of GHR or HMGA2.

The IGF1 association with MCM strengthened in an expanded cohort of 65 cases and 100 controls (P = 0.0007). We therefore added IGF1 genotypes into the model with FTSJ3 and weight and observed that weight was no longer a significant factor. A risk prediction plot using FTSJ3 and IGF1 genotypes illustrates an additive effect (Fig. 4E). Consistent with an additive model, individuals with the fewest derived alleles across the three genes have greater heights and weights and the lowest frequencies of MCM, while dogs having the most derived alleles are shorter, weigh less, and have higher frequencies of MCM (Table 1).

Table 1.

Distribution of genotypic combinations across dental and morphometric phenotypes in 65 MCM cases and 100 controls

FTSJ3	GH1	IGF1	Unilateral	Bilateral	Controls	95% CI*	P value*	Mean weight, pounds	Mean height, in
+/+	+/+	+/+	0	0	12	0–0.20	0.045	26.53†	15.92†
+/+	+/+	+/mut	1	0	18	0–0.21	0.005	23.64†	15.22
+/+	+/+	mut/mut	0	0	5	0–0.37	0.083	26.40	15.91
+/+	+/mut	+/mut	0	1	0	—	—	18.70	15.80
+/mut	+/mut	+/+	2	0	18	0.02–0.28	0.002	21.46†	14.99
+/mut	+/mut	+/mut	3	3	20	0.13–0.45	0.021	21.92†,‡	15.06‡
+/mut	+/mut	mut/mut	3	6	4	0.42–0.88	0.19	17.85†,§	14.48§
+/mut	mut/mut	+/+	0	2	2	—	—	22.70	14.91
+/mut	mut/mut	+/mut	1	1	8	0.05–0.49	0.064	19.16§	14.20§
+/mut	mut/mut	mut/mut	0	0	1	—	—	21.00	15.25
mut/mut	mut/mut	+/+	5	7	6	0.45–0.84	0.153	16.54‡	13.79§
mut/mut	mut/mut	+/mut	5	9	4	0.55–0.92	0.028	14.67	13.53
mut/mut	mut/mut	mut/mut	4	11	2	0.67–0.97	0.007	13.40§	13.12§

^*Calculated for genotypes observed at least five times in all 65 MCM cases and 100 controls. Significant statistics are bolded. Laterality status was unknown for one case with the genotype +/mut +/mut +/mut.

^†One control had measurements that were not available.

^‡Two cases had measurements that were not available.

^§One case had measurements that were not available.

FTSJ3 Explains Most Weight Variance in Shetland Sheepdogs.

When evaluated simultaneously (e.g., height as predicted by FTSJ3 and GH1), the individual contributions of FTSJ3 and GH1 to height and weight cannot be disentangled because the genes are in high LD. In independent models (e.g., height as predicted by FTSJ3), FTSJ3 is the most impactful, reducing body size by about 1 inch and 5 pounds with each derived allele and accounting for 37% of height variance and 73% of weight variance in our cohort. GH1 is slightly less impactful, explaining about 32% and 67% of height and weight variances, respectively. Comparatively, IGF1 explains only 7% of height and 28% of weight in Shetland Sheepdogs.

Discussion

Domesticated dogs vary in body size more than any other terrestrial species, but genetically isolated purebred populations must conform to standards that restrict variation within a breed. Each breed therefore possesses a set of polymorphisms that dictate growth, and these loci are under constant artificial selective pressures (5). While many inherited diseases of dogs correlate with body size [e.g., Legg–Calve–Perthes disease (11), osteosarcoma (12), and gastric dilatation-volvulus (13)], surprisingly few studies to date have directly attributed disease phenotypes to alleles underlying growth (14–16). In this study, we discovered that a dental anomaly in the Shetland Sheepdog correlates with small body size and that a large proportion of height and weight variation within the breed is attributed to a locus on chromosome 9 that harbors FTSJ3 and GH1.

GH1 is an excellent candidate gene for a growth phenotype and was, in fact, recently suggested to harbor variants contributing to small size in dogs (17). GH1 encodes pituitary growth hormone, the release of which is the first step in the GH1–IGF1 axis that regulates cell proliferation, differentiation, and apoptosis. We demonstrate here that a single base insertion in the 3′ donor splice site of GH1 exon 3 causes incomplete alternative splicing. In humans, weak splice sites surrounding this exon increase the importance of intron 3 splicing sequences (8). Mutations that occur within the first six bases of the donor splice site or interrupt splicing enhancer elements cause exon 3 skipping (18). The mutant protein misfolds and exerts a dominant negative effect through disruption of the trafficking and stability of wild-type GH1 (19, 20), resulting in pituitary dwarfism. Orofacial anomalies are common in growth hormone deficiencies and include underdevelopment of the maxilla and mandible, delays in tooth eruption, and tooth agenesis or malposition (21, 22).

FTSJ3 has not been previously described as having a role in canine body size, although genome-wide studies have associated the locus with various height- (23) and weight-related phenotypes in other species, including humans (24–27). FTSJ3 encodes an RNA 2′-O-methyltransferase, located in the nucleolus, and is ubiquitously expressed and evolutionarily conserved (28). Since 1974 it has been known that 2′-O methyl groups are added to pre-rRNA during its synthesis (29), and it is now clear that these modifications are made by FTSJ3 (30). FTSJ3 has also been shown to add internal 2′-O methyl groups to HIV RNA (31), raising the possibility that it internally modifies cellular pre-messenger RNA (mRNA) and/or mRNA. Knockdown of FTSJ3 activity decreases the rate of cell proliferation (30), which may be a consequence of decreased ribosome production and/or mRNA supply. In any case, a decrease in the net rate of cell proliferation in developing puppies of toy breeds might account for the markedly slower rates of growth in these breeds as compared to larger breeds. In an alternative scenario, mutation of FTSJ3, or a linked noncoding variant, could disrupt the complex regulatory pattern of the GH1 locus and impact growth hormone signaling (32).

FTSJ3^mut predicts a nonconservative substitution of glutamic acid for lysine in the conserved C-terminal domain of FTSJ3, wherein binding to preribosome complexes is mediated (33). This variation was consistently better associated with MCM and body size, despite a complete lack of recombination between the FTSJ3 and GH1 derived alleles. We attribute this to an ancestral haplotype (FTSJ3⁺-GH1^mut) that was more frequent in larger-bodied unaffected Shetland Sheepdogs, illuminating a significantly lower risk of MCM in individuals having a wild-type copy of FTSJ3 and reflecting the additive effect of the mutant allele. The nonrisk allele of the lead SNP also occurs on this haplotype, suggesting that the strongest associations in the GWAS were detecting the protective factor of the FTSJ3⁺ allele.

Shetland Sheepdogs are variable at two other loci harboring genes encoding proteins downstream in the GH1–IGF1 axis. We observed that among individuals homozygous for FTSJ3^mut-GH1^mut, IGF1 had a mild effect, modestly decreasing weight and increasing risk for MCM with the addition of each derived allele. In FTSJ3^mut-GH1^mut heterozygotes, however, IGF1 was much more impactful, taking individuals from low risk in heterozygosity to high risk in homozygosity and decreasing weight by nearly 20%. An absence of growth hormone would impact the secretion of IGF1, potentially masking the effect of an IGF1 mutation. Our findings suggest that homozygosity for GH1^mut results in the release of at least some growth hormone, possibly because of incomplete alternative splicing. Heterozygosity for GH1^mut appears to permit the production of much more growth hormone, thereby allowing us to clearly see the effect of mutant IGF1. It is important to consider that “small” alleles of other body size genes not studied here may be impacting the trends we observed.

The chromosome 9 locus harboring FTSJ3 and GH1 has been associated with head and mandible lengths across breeds (34). We noted that the Shetland Sheepdog and other breeds in which MCM has been reported [e.g., Italian Greyhound and Fox Terrier (1)] are dolichocephalic, whereas the toy breeds that also possess the derived alleles are mesocephalic or brachycephalic (35). Because the frequencies of dental anomalies in dogs correlate with skull morphology (36), it is possible that the presence of MCM in the Shetland Sheepdog, and not the toy breed dogs, could be attributed to the impact of the chromosome 9 locus on their elongated skull.

Based on our data, we propose that the chromosome 9 locus demonstrates vertical pleiotropy (37), whereby its effect on dentition is mediated by its role in reducing size, specifically of the skull. Given the strong selection for height within breeds, it is plausible that multiple variants within this locus contribute to body size variation in dogs. Future studies will be necessary to determine exactly which variants, or combination of variants, in this locus are retarding growth.

Materials and Methods

Study Population.

All samples were obtained with informed consent according to protocols approved by the Clemson University Institutional Review Board (IBC2018-13). Buccal cells or whole blood samples were collected from 230 Shetland Sheepdogs with unilateral MCM, bilateral MCM, or proper canine alignment. Pedigrees were obtained for all dogs. Dental status was verified by physical examination, dental photographs, radiographs, and/or veterinary records. During the course of the study, it became necessary to obtain height (measured at the withers) and weight data for each dog. Height and weight measurements were reported by owners for 167 and 161 dogs, respectively. Dogs were at least 1 y of age at the time of measurement (median age = 6 y; average age = 6.8 y). DNA was isolated following the Gentra Puregene DNA Isolation protocol (Qiagen) and concentration was quantified using a NanoDrop 1000 spectrophotometer (Thermo Scientific).

Genome-Wide Association Analyses.

Inclusion in the GWAS for MCM was determined based on relatedness: selected dogs were unrelated within at least two generations. We also sought to balance geographic origin, sex, and coat color/pattern. Genotyping was performed for 78 Shetland Sheepdogs at GeneSeek, Inc. using the Illumina CanineHD BeadChip containing 220,859 markers. All samples had call rates >95%. SNPs with call rates <95%, minor allele frequencies <5%, and/or significant deviation from Hardy–Weinberg equilibrium (P < 0.0004) were excluded from further investigation. All filtering and statistical analyses were conducted with SNP & Variation Suite v8 (SVS; Golden Helix, Inc.). All chromosome positions are reported in CanFam3.1.

For MCM GWASs, Fisher’s exact P values for each SNP were calculated under an additive model. Because hypodontia was underrepresented in the control GWAS population (n = 2) compared to the cases (n = 13), we confirmed our results for MCM through a second GWAS in which we excluded all but two randomly selected cases having hypodontia (28 cases vs. 39 controls) (SI Appendix, Fig. S1). Height (inches) and weight (pounds) were analyzed as quantitative values under an additive model. Covariates were considered using a linear regression under a full vs. reduced model.

LD pairwise analysis was performed to calculate r² values for the MCM lead SNP (chr9:12753481). The r² values were calculated in PLINK (38) using all controls.

Variant Analyses.

Variant filtering within a VCF file of 722 dogs (39) was performed using Golden Helix SVS and manual scanning of the critical interval was performed in IGV (40). Variants in the affected Shetland Sheepdog (accession no. SRX4036142) were filtered for allele state and against five Collie genomes (accession nos. SRX2506416, SRX2506417, SRX2506418, SRX2506419, and SRX2506420) in Golden Helix. The impact of nonsynonymous SNPs in FTSJ3 and AXIN2 were characterized using two in silico programs (41, 42): PolyPhen2 scores ranging from 0.85 to 1 and PANTHER preservation times (in millions of years) >450 were considered probably damaging. Variants of FTSJ3, GH1, AXIN2, and IGF1 were genotyped using Sanger sequencing. Primers are reported (SI Appendix, Table S4). Across breed allele frequencies for FTSJ3^mut and GH1^mut were calculated using whole genome data from 1,049 dogs (SI Appendix, Table S3).

Statistical Calculations.

Associations of alleles or haplotypes with phenotypes were assessed by Fisher’s exact tests using VassarStats (http://vassarstats.net/). Two-sample t tests in Microsoft Excel were used to evaluate differences in morphometric means between MCM cases and controls and between genotypes.

Having combined the bilateral and unilateral MCM disease phenotypes, the observations evaluated are now classified into one of two classes (i.e., affected and control). Accordingly, for the evaluation of the binary MCM phenotype, we turn to logistic regression to model the risk of disease as a function of the observed genotypes and other potential explanatory variables. Define the probability of disease as $p_i$ for the $i$-th dog and thus the logit of this probability as ${\theta }_i=log\left[p_i/\left(1-p_i\right)\right].$ Modeling the logit as a function of any explanatory variables considers this simple linear model:

$${\theta }_i=b_0+b_{w}Weigh{t}_i+b_{h}Heigh{t}_i+b_{s}Se{x}_i+{\displaystyle \sum }_{j=1}^k\left({\alpha }_j+{\delta }_j\right),$$

where $b_0$ is an unknown constant common to all dogs, $b_w$ is the regression coefficient of body weight on MCM, $b_h$ is the regression coefficient for height, $b_s$ is the regression coefficient on sex (for males = 1 and 0 = females), and ${\alpha }_j$ and ${\delta }_j$ are the additive and dominance contributions, respectively, of the j-th SNP of the k SNPs being considered in the analysis. Estimation of the unknown model parameters is accomplished with the glm command in the public domain language R (43). Of course, various reduced models can also be considered. In the comparison of these models, that with the lowest AIC value (44) represents the best and most parsimonious fit to the disease data.

The potential impact of the observed SNPs on the measurements of height and weight was also evaluated. Height and weight were assumed to be normally distributed variables conforming to a straightforward linear model for the i-th dog:

$$y_i=b_0+b_{s}Se{x}_i+{\displaystyle \sum }_{j=1}^k\left({\alpha }_j+{\delta }_j\right),$$

where $b_0$ is an unknown constant common to all dogs, $b_s$ is the regression coefficient on sex (for males = 1 and 0 = females), and ${\alpha }_j$ and ${\delta }_j$ are the additive and dominance contributions, respectively, of the j-th SNP of the k SNPs being considered in the analysis. Estimation of the unknown model parameters is accomplished with the lm command in the public-domain language R (43). As before, comparison of models was facilitated through the AIC (44).

Cloning of GH1 Fragments into pcDNA Vector.

GH1 fragments were amplified from the genomic DNAs of two Shetland Sheepdogs, an affected dog homozygous for FTSJ3^mut-GH1^mut, and a control dog homozygous for FTSJ3⁺-GH1⁺, using Herculase II Fusion DNA Polymerase (Agilent) and primers with sequences that overlapped with the pcDNA3.1 vector (Invitrogen) (SI Appendix, Table S4). The pcDNA3.1 vector was amplified with Phusion polymerase (Thermo Scientific). GH1⁺ and GH1^mut fragments were cloned into the pcDNA3.1 vector using Gibson assembly (New England Biolabs) following the manufacturer’s instructions. Insertion of GH1 fragments into the vector was verified by Sanger sequencing.

Transfection of GH1 into HEK293 Cells.

HEK293 cells were grown at 37 °C with 5% CO₂ in Dulbecco’s modified Eagle’s medium (Gibco) with 10% fetal bovine serum. HEK293 cells were transfected with 4.0 µg of GH1⁺ or GH1^mut plasmid DNA using 12.0 µL of Lipofectamine LTX (Thermo Fisher) following the manufacturer’s instructions. Cells were harvested 48 h posttransfection. RNA was isolated from cells using GeneJET RNA Purification Kit (Thermo Scientific) following the manufacturer’s protocol. Five hundred nanograms of RNA was reverse-transcribed using MLV-RT (Optizyme) with a dT(20) primer at 42 °C for 1 h.

Sequencing of GH1 cDNA.

The cDNAs were amplified using Phire Green Hot Start II DNA Polymerase (Thermo Scientific) following manufacturer guidelines. Primers were designed to amplify the last four exons of GH1, with the forward primer located in exon 2 and the reverse primer located in exon 5 (SI Appendix, Table S4). Minus reverse transcriptase and empty vector controls were also run. PCR products were visualized via gel electrophoresis. Products were purified using the E.Z.N.A. Gel Extraction Kit (Omega Bio-Tek) and verified by Sanger sequencing.

Genotyping of Other Body Size Loci.

Previously associated Illumina SNPs in the following genes influencing small body size in dogs were genotyped in the SNP chip data from our GWAS population: IGF1R (chr3:41758863 and 41849479), GHR (chr4:67040898), IGF1 (chr15:41221438), and HMGA2 (chr10:8183593) (5, 45).

Data Availability.

Whole-genome resequencing data from affected Shetland Sheepdog, five control Collies, and all other dog genomes used herein have been deposited in the Sequence Read Archive (all accession numbers are given in SI Appendix, Table S3). Nucleotide sequences for Growth Hormone 1 spliceoforms are deposited in GenBank under accession numbers MT499773 to MT499775.