DNA samples were originally isolated from blood by a sodium perchlorate extraction technique (18). These samples were obtained from remaining aliquots of previous studies (19). Samples from the following breeds (with numbers of samples) were used during the development of the multiplex panel: Quarter horse (18); Arabian (16); Thoroughbred (13); Paint (11); Standardbred (10); Miniature horse (6); Morgan (6); Haflinger (4); Percheron (4); Pinto (4); American Saddlebred (4); Appaloosa (3); Hanoverian (3); Friesian (2); Tennessee Walker (2); Belgian (1); Anglo-Arab (1); Morab (1); American Paso Fino (1); Paso Fino (1) and mixed breed horses (12). Additional samples from affected and unaffected Friesian horses (not counting the two Friesians listed above) were collected from blood samples or tissues from horses that died after developing megaesophagus. Lastly, some additional samples were collected from hair roots from an Arabian horse breeding farm to determine usability of this DNA source. DNA was isolated from 20 hair roots using QIAGEN DNeasy Blood and Tissue kit as per the manufacturer's instructions. DNA samples from several other species (mouse, cat, dog, cow, and goat) were generously supplied by other labs on campus. DNA samples from these other species were examined, as is commonly done with microsatellites that might be used for forensic purposes. The rule for examining other species is based on the human forensics Scientific Working Group on DNA Analysis Methods (SWGDAM) validation guideline 3.2 “species specificity” (https://www.swgdam.org/publications). The goal is to prevent misinterpretation of misidentified or mixed DNA samples. Although this guideline has not been a requirement in animal forensic testing, it is nevertheless often followed during the validation process. DNA samples were collected under IACUC approval.

We had previously noted that tetraSTRs with Smith-Waterman (SW) scores > 450 for dogs and humans (UCSC Genome Browser prior to 2012) had useful heterozygosity (>0.80 of the maximum expected heterozygosity in both humans and purebred dogs; Venta, unpublished results). We used the same strategy to identify tetra- and pentaSTRs in the horse genome using the UCSC Genome Browser's table browser function. In later stages of the work, a second critical feature of the STRs was identified with respect to heterozygosity. The longest uninterrupted sequence (LUS) had to be eight or greater for the SW score to adequately predict the heterozygosity for a given population. In order to produce useful primer sets as conveniently as possible, STRs were chosen that also had a few hundred bp of unique flanking DNA (i.e., areas that do not contain repetitive elements such as SINEs or LINEs) that was less likely to produce off-target amplicons. Amplicons that would include other STRs (e.g., diSTRs and monoSTRs) that would confound the interpretation of the tetraSTR were also avoided. In addition, only one STR was chosen per horse chromosome, except for chromosome 4 for which two STRs were chosen that are 68 Mb apart.

When potentially useful tetra- and pentaSTRs were identified near the candidate genes (generally within 2 Mb), the sequences plus several hundred bp of flanking sequence were searched for PCR primer pairs using Primer3 software (http://bioinfo.ut.ee/primer3-0.4.0/). Many primers were chosen for which the 3′ end finished in two adenines (AA) for reasons that will be discussed later. As primer sets accumulated, additional effort was put into identifying sets that would not encompass overlapping ranges with other markers labeled with the same fluorescent dye. When marker ranges were discovered to overlap, new primers were designed to adjust the size range for one of the markers that showed the overlap. Forward (F) primers were labeled with either 6-FAM (hereafter referred to as FAM), or HEX, both from IDT (idtdna.com, USA) or NED (from ABI, USA, obtained through Thermofisher.com, USA). Unlabeled primers were purchased from IDT. Primer sequences used in the final multiplex panel are listed in Table 1 and other primers are listed in Supplementary Tables S1, S2. Primers were initially tested for robust amplification and good heterozygosity as single amplicons using a universal primer labeling system (see below in section Universal and directly labeled primers). Many reverse primers included an additional pigtail or a single “g” to suppress split peaks (20, 23). Occasionally (e.g., for the amelogenin marker) extra bases were added to avoid overlap with other markers or pull-up peaks. Primers were also checked using Autodimer for interactions that might occur during multiplex analysis [(24); https://www-s.nist.gov/dnaAnalysis/]. Our primary criteria were that no pairwise primer interactions were found that had a score of seven or more (AutoDimer output), and that the reverse primer did not template more than two bases at the 3′ end within the forward primer. Some care was also taken to keep the LUS shorter than 16 to reduce the problem of large stutter bands (25).

In order to test the hypothesis that one of several candidate genes underlies the megaesophagus phenotype in Friesian horses, tetraSTRs were identified flanking these genes. The candidate genes selected were AAAS, COL4A5, COL4A6, FH, GDNF, HMGA2, NOS1, NOS2, NOS3, KIT, MED12, PTPN22, RASSF1, RET, SPRY2, SPRY4, and VIPR1. Tetra- or pentaSTRs were identified using the UCSC Genome Browser and SW scores above 450. Primers were designed to amplify these repeats using Primer3 (http://bioinfo.ut.ee/primer3-0.4.0/). Primer sets were then examined in DNA samples from affected and unaffected Friesian horses, as well as samples from eight horses of other breeds (Appaloosa, Arabian, Belgian, Quarter horse, Miniature horse, Percheron, Thoroughbred, and a mixed breed horse). Amplicons were labeled with a fluorescently labeled M13 tag sequence as an inexpensive way to implement high resolution genotyping (see below under Universal primers).

Sequences were aligned from Genbank records AB091793.1 (amelogenin sequence on the horse X chromosome) and AB091794.1 (amelogenin sequence on the horse Y chromosome) using the SeaView alignment software (21). Regions were selected surrounding a gap in the Y sequence compared to the X sequence and for which primers exactly matched both the X and Y chromosomes. Primer sequences are shown in Table 1.

Primer sets were initially amplified as single amplicons using the following conditions: 10 mM Tris (pH 8.3), 50 mM KCl, 1.5 mM MgCl₂, 0.1 mM dNTPs, 0.02 U Amplitaq (Qiagen, USA) per μl of final reaction volume and ~10 ng horse DNA (quantified by nanodrop and Qubit) in a 25 μl reaction. Cycling conditions were 1 min 94°C, 2 min 59°C, and 3 min 72°C for 50 cycles. Potential variability was assessed by running a 2% agarose gel and examining allelic and heteroduplex bands from a sample of Friesian and other horse breeds. For multiplex reactions, two sets of conditions were used. The first set was 10 mM Tris (pH 8.3), 50 mM KCl, 2.0 mM MgCl₂, 0.2 mM dNTPs, 0.04 Amplitaq Gold (Qiagen, USA) per μl, a primer mix to produce the final concentrations given in Table 1, and 10 ng horse DNA in a 25 μl reaction. Multiplex cycling conditions were the same as for single amplicons, except that the annealing temperature was 57°C. The second set of conditions used the Multiplex Microsatellite Type-It kit (Qiagen, USA) (12.5 μl) with 4.8 μl of primer mix and 2 μl (10 ng/μl) of DNA. The cycling conditions were 5 min 95°C for the initial denaturation and hot-start Taq activation step, then 30 s 95°C, 1.5 min 60°C, and 1 min 72°C for 40 cycles, and final extension at 68°C for 90 min. Primer concentrations were adjusted to bring the peak heights for all markers to an acceptable balance (generally less than a three-fold difference among different markers as assessed by homozygous peak height ratios). Cross-species amplifications (the species are given above under “DNA samples”) were also examined with no horse DNA included and amplified under the same conditions for the final multiplex panel.

In order to reduce the expense of testing markers for variability, several universal primers were used. These included the M13 primer, and primers C and D of Blacket et al. (22). After we discovered that at least the M13 universal primer caused primer-dimer among some horse-specific primer sets, a switch was made to universal primers that had the dinucleotide AA at the 3' end (Supplementary Table S2) (26). Because primer-dimer was a significant problem in the multiplex panel with the original primers that were already directly labeled, many reverse primers were redesigned to have AA at the 3′ end.

Two alleles for each marker were chosen for Sanger sequence analysis. This was accomplished either by using the amplification primers if they were far enough away from the STR to allow the complete repeat to be sequenced, or by designing new primers that were further away from the repeat (Supplementary Table S2). Sequences were aligned with the reference genome using the SeaView software package and the repeat numbers were counted (21).

STR loci were named according to their approximate chromosomal base locations based upon the Broad/equCab2 assembly, although for brevity in the text, they are generally just designated by the particular chromosome upon which they reside. Motifs were based on the upper strand (all markers show the same strand orientation in the reference genome builds EquCab 2 and 3). Repeat motifs were based upon the designation by Tandem Repeats Finder (TRF) as given in the UCSC Genome Browser (27). We have found these reported motif sequences to conform well to the recommendations published by the DNA Commission of the International Society for Forensic Genetics (28, 29). The two alleles for each marker were bidirectionally Sanger sequenced and the repeat numbers were counted. Alleles were then named by the number of repeats sequenced. Repeat numbers for the other alleles contained in this report were inferred from this data.

Stutter peaks were compared to allele peaks to determine stutter rates. Ten peaks were analyzed per marker. Alleles were only used if they were homozygous, or heterozygous and for which the two alleles were more than one repeat apart, and if the allele or stutter peak did not coincide with another peak that had a pull-up peak of the same color.

After manual correction of rounding errors from main peak sizes, alleles and genotypes were analyzed using the Genalex analysis package (30, 31). For example, output from the high-resolution genotyping system using the local Southern method might give 100.48 base and 100.52 base alleles for different horses, which are both considered to be the same allele (and they were manually corrected to be the same allele). Only 16 markers for this analysis were used because it was discovered after the analysis of all of the horses that one marker (Eca28) had apparent null alleles in more than 50% of the Quarter horses, Thoroughbreds, and Paints. After the main analysis, these primers were redesigned and the new designs worked in all horses (those primers are reported in Table 1). The Cervus program was also used to check the power to exclude for parentage (32).

A 17-plex panel of horse tetra- and pentaSTRs was developed, plus an amelogenin (amel) marker that can be used to distinguish the sex of the horse (Supplementary Figure S1). The average number of alleles per marker was 8.3 ± 2.3. The average observed heterozygosity (Ho) and expected heterozygosity (He) for all markers for different breeds can be found in Table 2. Allele frequency distributions are shown in Figure 1. The probability of identity (PI; the probability that two genotypes from unrelated horses will match at random) within a given breed ranged from 1.1 x 10⁻¹¹ to 1.5 X 10⁻¹² and the PI among siblings (PIsib, the chance complete matching of genotypes of siblings) ranged from 1.4 x 10⁻⁵ to 7.2 x 10⁻⁶. These values were calculated for Paint, Quarter horse, Standard bred, Thoroughbred, and Arabian horses, where sample sizes of 10 or greater were available. Other breeds were genotyped, but the sample sizes were insufficient to produce meaningful calculations.

Although there was some weak ability to identify horse breeds based upon the output of Genalex analysis (mean correct assignment 65%), some horses clustered with the wrong breed. One marker, Eca12, had some tendency to produce split peaks, despite the presence of a pigtail on the reverse primer so, in addition to a prolonged final extension, some care was necessary to infer the genotype. The multiplex panel produced good results with both blood and hair-root derived DNAs (data not shown).

Genalex identified six markers in the horse 17-plex panel that could potentially contain null alleles [based on Hardy-Weinberg equilibrium (HWE) calculations]. Those markers that were not consistent with HWE across horse breeds, were three markers (Eca05, Eca21, and Eca22) not conforming to HWE in two breeds and the other three markers (Eca01, Eca02, and Eca03) in one breed each (data not shown).

Relatively rare micro-variants (<2% of all alleles for a marker) were found in four of the markers: Eca02, Eca15, Eca20, and Eca25. These micro-variants were visually confirmed using the “overlay all” function of the Peak Scanner software, but were not sequenced in this study because of their rarity. A single micro-variant in the Eca12 marker was at a much higher frequency, 38% of all alleles. This variant was sequenced and found to be due to a single base indel (G/-) outside of the repeat, near the edge of a poly-T run of 10 bp (the location can be seen by comparing Genbank records MN684185 and MN684186). It may be a good practice to bin the 17.3 microvariant for Eca12 with the nearest whole repeat (18) because the single base deletion is outside of the main repeat. This was suggested for certain microvariant alleles in the Mini-Dogfiler panel for the same reason (33). The frequency of these micro-variants, as well as the on-ladder alleles for all of the markers, are shown in Figure 1.

Some of the original horse primer sets used to examine candidate genes had low or no variability within the unaffected Friesian horse population, despite having SW scores that are correlated with heterozygosities higher than 65% in purebred dogs and 86% in human outbred populations (Venta unpublished results; Supplementary Table S1). This demonstrates that, at least within some mammalian genomes, in order to have reasonable variation it is first necessary to have an uninterrupted perfect repeat (also known as LUS) of at least 8, no matter how high the SW score for the particular microsatellite (25). One exception was a marker called MK21 (Supplementary Table S1). Although the longest uninterrupted tetranucleotide repeat was three for the MK21 primer set, this marker had a longer perfect repeat that was 16 bases long and was repeated five times in the reference genome. Thus, on occasion, a tetraSTR (as identified in RepeatMasker) may have variation despite not having an LUS of eight or more for the tetraSTR, and its variability may depend upon higher order repetitions within a given marker. We have observed this for a few other markers in other species as well (unpublished results).

An examination of several markers for amplification showed that the M13 primer that we used to determine the variability of markers could cause strong primer-dimer formation (Supplementary Figure S2). For this reason, a new set of universal primers was developed that reduced the probability of primer-dimer formation. This set of universal primers was used in previous studies for other species (28, 34). The basis for reducing primer-dimer formation was the design of most primers with AA at the 3′ end (35).

MK21, in addition to showing reasonable heterozygosity despite having an LUS of only 3 for the tetraSTR (as described above), also migrated aberrantly compared to its sequenced size. It was therefore not included in the final multiplex panel. It is not clear what causes the fragments to run dozens of bases too quickly, but it seemed prudent to exclude it from the multiplex panel. Another marker called 7981, which is located on chromosome 6, was excluded from the multiplex panel because different alleles either did or did not include a deletion of 12 bp outside the primary repeat. Marker 8019 on chromosome 11 was excluded because the repeat structure is complicated and would have caused difficulties in nomenclature. Marker Eca28 in the final 17-plex panel required a redesign because the original set failed to amplify in many horses, mostly within three breeds (Arabian, Thoroughbred, and Paint). The redesigned set, included in the multiplex panel, no longer had the null phenotype.

Stutter for the markers ranged from 1.2 to 11.3% (Table 1). The highest stutter was found for the marker Eca20, ~11% stutter averaged across all alleles, which had the largest LUS in the reference genome (18), and also had the largest number of observed alleles (17) among all markers in the 17-plex panel.

DNA samples from several other species were also subjected to PCR with the 17-plex panel. Zero to four small peaks (<200 bases, except for the goat) were observed among the different species examined (Supplementary Figure S3). Based upon the published results for other species, this result does not seem to be unusual, but it should be taken into account if these markers are used for certain forensic analyses (33).

Markers within 2 Mb of candidate genes were examined in a group of 13 Friesian horses that had been diagnosed with megaesophagus. If a marker showed heterozygosity for two alleles that differed by more than one repeat unit, the corresponding gene was considered to be excluded as underlying a recessive mode of inheritance for megaesophagus; for discussions of criteria and assumptions underlying exclusion analysis (6, 36). If homozygous markers were not shared among all affected horses (i.e., if they were homozygous for different alleles in different affected horses), they were excluded under a dominant or X-linked mode of inheritance. Results for excluded genes are shown in Supplementary Table S3.