Science-in-brief: Genomic and transcriptomic approaches to investigation of equine diseases

It is now over ten years since the publication of the first draft of the horse reference genome of the Thoroughbred mare, Twilight (EquCab2.0) ¹. In 2018, this initial draft was followed by extensive re-sequencing and assembly efforts to improve the horse genome’s contiguity and composition, resulting in EquCab3.0 ². These reference genomes enabled the development of genomic tools that have resulted in unprecedented advances in equine research (reviewed in ³). Genomics and transcriptomics, together with other areas of “omics” sciences are extremely powerful tools which are likely to rapidly and comprehensively transform understanding of equine disease in future. To highlight Equine Veterinary Journal’s interest in and contributions to this exciting field, we have created an online collection of our recent omics articles. This accompanying editorial draws on work published in EVJ and other journals to demonstrate how genomic and transcriptomic approaches have been used to investigate equine diseases.

Genomics

The availability of an equine reference genome led to the development of three generations of single nucleotide polymorphism (SNP) chips, which have been used to map traits and diseases. In 2011, the first- and second-generation SNP arrays, containing 54,602 and 74,500 SNP markers, respectively, became available (reviewed in ⁴). In 2017, a third-generation SNP array was developed that increased the number of genetic markers to 670,805 SNPs ⁵. This array has successfully identified genomic regions containing genetic variants responsible for monogenic traits in the horse (i.e. the nonsense variant in ST14 associated with Naked Foal Syndrome in the Akhal-Teke ⁶ and variants in KRT25 and SP6 responsible for curly coat in horses ⁷). While invaluable discoveries, the application of performing genome-wide association studies (GWAS) for more complex traits, such as osteochondrosis (OC), susceptibility to infectious disease and neurologic conditions was an exciting prospect.

One of the most crucial components of any GWAS is stringent phenotyping of both case and control populations ⁸. This focus on rigorous phenotyping was highlighted in the recent GWAS for dynamic laryngeal collapse in Norwegian-Swedish Coldblooded Trotter racehorses ⁹. This dynamic airway obstruction cannot be observed during resting endoscopy and, with exercising endoscopy, is only observed when horses are exercised with tension on the bit and reins ¹⁰. To perform the first GWAS for this likely inherited condition, Velie et al. ⁹ rigorously selected their study population and only genotyped horses that had undergone dynamic endoscopy with and without poll flexion. These horses were genotyped on the third-generation 670K+ Axiom Equine Genotyping Array. After quality control, this resulted in n=33 cases and n=23 controls. While a significant region of genomic association was not achieved in this study, a suggestive region was identified on chromosome 7 (ECA7) that requires further investigation.

Other important considerations in GWAS are quality control of genotyping and the type of statistical analysis performed ⁸. Quality control consists of excluding any SNPs or individuals that do not meet a pre-defined quality threshold, as demonstrated in the study by Velie et al. ⁹. For association testing, traits can be considered quantitative or binary trait (i.e. “cases” and “controls”). There are often other important covariates that should be included in the model when testing for an association. In the study by Velie et al. ⁹, sex was included as a fixed effect even though there does not appear to be a sex-related component to the phenotype ¹¹. Additionally, population stratification, or systematic ancestry differences between cases and controls, must be evaluated in any GWAS cohort. Population stratification can result in false positive associations if not corrected ¹². Typically, GWAS studies are performed within one breed of horse to minimise population stratification. However, differences in population structure can occur even with one breed ¹². Methods to identify population structure in GWAS include computing the genomic control (λ_GC) and visualising a quartile-quartile (Q-Q plot). A value of λ_GC=1 indicates no stratification, whereas λ_GC>1 indicates stratification ¹³. In the Velie et al. study ⁹, λ_GC=1.13, demonstrating some degree of population stratification even within one breed. Visualisation of multidimensional scaling plots will also identify subpopulations within cases and controls (see Fig. 2 ⁹). To correct for this, the authors chose to perform a principal component analysis (PCA) and incorporate the first four components, which explained 86.7% of the variance, as covariates into their model. With these covariates applied, the corrected Q-Q plot (Fig. 3 ⁹) demonstrates successful modeling.

To account for the thousands of statistical tests performed in defining the P value for significant associations, stringent corrections must be applied in any GWAS. A Bonferroni testing correction (0.05 / (# SNPs tested) is considered the most stringent approach and this correction was applied in the study by Velie et al. (i.e. red line in Manhattan plot, Fig. 3 ⁹). Due to its stringency, a threshold for “suggestive” significance is often applied, which may be set at a 10% false discovery rate or arbitrarily defined from human GWAS. In this author’s opinion, using human GWAS suggestive thresholds is not appropriate in equine studies since this cut-off was based on the effective number of independent tests in the human genome if all common SNPs in HapMap were tested with direct genotyping or imputing ¹⁴. Thus, in the horse, the cut-off developed for humans would not be directly equivalent.

Although a significant region of genomic association for dynamic laryngeal collapse was not identified in the Velie et al. study, the rigorous study design and analysis provide a candidate region for replication studies. Replication in an independent cohort is essential for any GWAS (as reviewed in ^8,15). Genetic risk factors for OC and OC dissecans (OCD) have been extensively evaluated using SNP-based technologies, as reviewed by Naccache et al. ¹⁶. Using first-, second- and third-generation SNP platforms, OC/OCD genetic risk has been evaluated across warmblood breeds, Thoroughbreds, French Trotters, Standardbreds and South German Coldbloods for fetlock, hock and stifle OC/OCD. Putative OC/OCD-associated SNPs and single nucleotide variants (SNVs) were identified on ECA1, 2, 3, 4, 6, 10, 14, 16, 18, 26, 27, 28, 29 and 30 (summarised in ¹⁶). Replication studies should be conducted in an independent dataset drawn from the same population as the GWAS, with the same SNPs or SNPs in high linkage-disequilibrium with the GWAS-identified SNP tested ¹⁵. To replicate the OC/OCD associations in the warmblood, n=440 Hanoverian warmblood horses, phenotyped by the standardised radiological examination for prepurchase evaluations, were genotyped for the previously identified 26 putative genetic variants ¹⁶. Sex was again included as a fixed effect and, in this study, a mixed linear model was used to account for population stratification ¹⁶. Mixed models can model not only population structure but also family structure and cryptic relatedness, which may not be captured by incorporating PCs as covariates ¹³. The random animal effects are based on a phenotypic covariance matrix, which is modeled as a sum of both heritable and non-heritable variation (summarised in ¹³). Although 670K markers were not tested in this study, correction for multiple testing is still required, which resulted in two genetic variants remaining significant for total OCD, hock OCD and fetlock-OCD ¹⁶. The strongest OCD-associated SNP (BIEC-808543 (rs68603064)) at ECA3:105 547 002) is located 63 kb upstream from the gene LCORL (ligand-dependent nuclear receptor corepressor-like gene). Excitingly, LCORL has been associated with equine body size ^17–22, which is a known risk for OC/OCD ²³. The second OCD-associated SNP (rs68945244) at ECA14: 73 865 948 is within the gene MCTP1 (multiple C2 and transmembrane domain containing 1), which, to date, has no functional relationships reported with cartilage or bone tissue metabolism but warrants further study in equine OC/OCD.

While the study by Naccache et al. ¹⁶ highlights how a replication study can confirm genetic variants, a recent GWAS by Dunuwill et al. demonstrates how replication studies can also refute previous findings ²⁴. In 2018, a GWAS by Brosnahan et al. identified a significant SNP in platelet-related gene that was associated with equine herpesvirus type-1 induced myeloencephalopathy (EHM) ²⁵. This 2018 study genotyped n=61 EHM cases and n=68 controls, including both experimentally infected and naturally exposed horses from 2001–2012. Cases were defined as horses that developed neurological signs including ataxia, urinary bladder paralysis or recumbency following EHV-1 infection whereas controls were horses that had developed non-neurological signs, including fever and respiratory disease, following EHV-1 infection. Horses were genotyped on both the first-generation (SNP50; 59 horses) and second-generation (SNP70; 70 horses) Illumina arrays and overlapping markers retained after quality control, resulting in 37 064 usable SNP markers. Breed, age and sex of horses were not reported in this study but the uncorrected λ_GC=1.30 indicated population stratification. A mixed linear marker analysis was performed to account for this stratification, reducing the λ_GC to 1.07. After Bonferroni correction, the initial GWAS did not identify any significant associations. Based on suggestive regions on ECA6, 10 and 25, 128 additional SNP markers were selected, genotyped using a custom Sequenom array (GeneSeek) and added into the analysis, resulting in 37 173 SNPs. Using additive and dominant models, there were no significant associations on the GWAS. However, a single marker was significant using a recessive model. This SNP (BIEC-946397) at ECA6:31 120 293 is located in an intron of the gene TSPAN9 (tetraspanin 9). This family of genes have role in platelet activation ²⁶ and, with the lesions of EHM including virus-induced endothelial damage, haemorrhage and thrombosis ²⁷, provided an exciting potential candidate gene for genetic testing of EHM susceptibility.

The recent study by Dunuwille et al. ²⁴ however, was unable to replicate this association. Phenotyping criteria were similar, except that both cases and controls were refined to only horses from naturally-occurring outbreaks. While the number of controls used was similar (n=67), fewer cases were available for genotyping (n=27). Horses were genotyped on the third-generation SNP array (670K Axiom array) that, following quality control, resulted in 382 529 SNPs for association testing. This set of SNPs included the BIEC2–946397 SNP from the Brosnahan et al. study. Both sex and age were not significantly different between case and control groups and thus were not included in the model. Both the case and control groups included many different breeds and therefore a second GWAS was performed using only the subset of Thoroughbreds, Paints and Quarter Horses (17 cases and 38 controls). Population structure was not reported in this study (i.e. no multidimensional scaling plot, Q-Q plot or reporting of λ_GC). Regardless, no significant association was identified following Bonferroni correction. Since the Brosnahan et al. ²⁵ study identified an association using a recessive mode of inheritance, Dunuwille et al. ²⁴ also evaluated runs of homozygosity in their cohort but did not identify any significant associations. Of note, the herpesvirus genotype for the ORF30 SNP was available in only 29/94 samples in the recent study ²⁴ and not reported at all in the 2018 study ²⁵. Viral genotype can lead to higher magnitude and duration of viremia ^28,29. Additionally, older horses are generally more susceptible to EHM and thus age remains an important variable to include in future studies (reviewed in ³⁰).

Comparison of these two studies highlights the need for replication and validation of initial GWAS results prior to development of genetic marker tests for equine diseases. Importantly, the results of a GWAS provide researchers with a candidate genomic region for investigation, but these SNPs are not the causative genetic variant for the disease. The next step, following identification and validation of a genomic region via GWAS, is to perform sequencing in that region and identify putative genetic variants that could be associated with the phenotype of interest. Causation (i.e. the genetic variant directly causes disease) can be very difficult to prove in veterinary studies and often requires the creation of targeted animal models (mice, zebrafish, etc.) that demonstrate a similar phenotype when the particular gene of interest is disrupted with the specific putative mutation identified in the horse.

While not able directly proving causality, using information gleaned from similar diseases in humans and other species can support a likely genetic variant. In the recent study by Hack et al. ³¹, whole-genome sequencing of a Tennessee Walking Horse identified a missense mutation in the gene GRM6 (metabotropic glutamate receptor 6), as likely cause for congenital stationary night blindness (CSNB) in this horse. This is a unique example of how leveraging the low cost of whole-genome sequencing for one individual horse can lead to novel discoveries when coupled with an understanding of candidate genes for CSNB in humans. For this study, the researchers first excluded a known genetic mutation in TRPM1 ³² as the cause for this horse’s condition. Next, the horse was whole genome sequenced and the genetic variants (SNPs and insertions/deletions) filtered based on the location of 100 candidate genes for CSNB in humans. Subsequently, these filtered SNPs were then tested for association between the one affected horse and 29 unaffected horses, across 7 different breeds that did not include Tennessee Walking Horses. An autosomal recessive model was selected. Thus, the affected horse had to be homozygous for the alternate allele and all other horses were homozygous for the reference allele. These filtering steps identified 23 candidate SNPs, of which only one was a coding variant (GRM6; NC_009157.3:g.2655618C>T) that changed an amino acid from a threonine to a methionine. In silico protein modeling was used to demonstrate that this SNP likely impairs glutamate binding. Importantly, GRM6 is almost exclusively expressed in the retina ³³ and six missense mutations in GRM6 have been reported to cause CSNB in humans, with recessive modes of inheritance ³⁴. Therefore, there is a very high likelihood that this genetic variant can be considered causal for CSNB in this Tennessee Walking Horse. Of importance to the industry, the estimated allele frequency in Tennessee Walking Horses was 10%, with a carrier frequency of 18% ³¹. Thus, this is an important genetic variant to screen for in this particular breed.

Transcriptomics

While genomics can provide strong evidence for associations of genetic variants or genomic regions with diseases in the horse, the study of gene expression (transcriptomics) can afford insight into the underling disease pathophysiology, while also providing potential markers for testing. One important consideration when performing transcriptomics work is that the RNA that encodes for proteins (i.e. mRNA) is not nearly as stable as DNA and can therefore be much more difficult to process.

Unlike DNA, where the nucleotide sequence will be the same regardless of sample type, gene expression is cell-type specific. Most early transcriptome studies used microarray technologies to profile gene expression in specific equine tissues of interest (reviewed in ⁴). More recently, mRNA-sequencing (mRNA-seq) has become a cost-effective method of evaluating gene expression across various equine tissue types ^35,36. An important benefit of mRNA-seq is the detection of alternative transcripts from the same gene, which can assist with mapping splicing variants.

While tissue-level gene expression can provide unique insight into disease pathology, the heterogeneity that exists in many tissues both in health (i.e. brain) and disease (i.e. influx of immune cells into the liver, etc.) can obscure findings due to differing cellular compositions. Technologies now allow for the profiling at a single cell level, permitting the assessment of biologic properties at unprecedented resolution. In the study by Karagianni et al. ³⁷, the transcriptome of equine alveolar macrophages was assessed using microarray technology with 30,559 probes. First, the investigators compared the transcriptome of alveolar macrophages, collected via a bronchoalveolar lavage from n=5 healthy horses euthanised for reasons unrelated to airway disease, to the transcriptome of peritoneal macrophages, collected via peritoneal lavage from another n=3 healthy horses euthanised for reasons unrelated to gastrointestinal disease. The authors identified 451 differentially expressed transcripts associated with the alternative M2 macrophage phenotype in alveolar macrophages and a hybrid M1/M2 profile in peritoneal macrophages. While these results provide the necessary first step to define cellular transcriptomes in the horse, an ideal comparison would have been between macrophage groups collected from the same horses to account for the biologic variability among individuals.

The investigators then performed a second experiment where they treated the alveolar macrophages with LPS and compared pre- versus post-LPS transcriptome profiles ³⁷. In this experiment, 240 differentially expressed transcripts were identified that included most well-known inflammatory genes. Excitingly, through the comparison of these results with similar studies in humans and mice, it was demonstrated that the LPS-induced gene expression profile of equine alveolar macrophages more closely resembles that of human, justifying the use of the horse as an appropriate model for human inflammatory airway research. Of note, 63/66 orthologous genes demonstrated the same type of dysregulation in horse alveolar macrophages vs. humans (i.e. same change in direction with LPS stimulation).

One of the difficulties encountered in the study by Karagianni et al. ³⁷ is the lack of a completely annotated dataset of transcripts for the horse. This problem is encountered both with microarray and RNA-seq experiments. Ongoing efforts to improve both gene annotation and also to incorporate epigenetic marks to understand gene function, similar to the human and murine ENCODE projects, are currently underway in the horse (reviewed in ³⁸). These publicly available tools will further enhance our ability to evaluate gene expression at a tissue and single-cell level and compare with results in other species.

While critical for the study of gene expression in disease, mRNAs can also act as biomarkers. In the recent study by Page et al. ³⁹, targeted candidate genes were profiled via RT-qPCR in catastrophically injured racing Thoroughbreds (n=107) and compared to profiles from non-injured horses sampled pre-race (n=374) or post-race (n=205). The goal of this study was to identify mRNA biomarkers that were significantly associated with catastrophic injury, which could potentially lead to future screening tools to prevent injury. By collecting blood in Tempus® Blood RNA tubes, the investigators were able to isolate total quality RNA from all horses. Genes to profile included those that encoded proteins involved in inflammation, bone repair and remodeling, tissue repair and response to injury. Of the 21 genes analysed, the investigators used a subset of samples from 37 horses with pre-race and post-race samples collected to exclude 12 genes that changed expression based on exercise alone. The statistical model used included type of race, jurisdiction, sex and age with a binary outcome (injured or non-injured). Three genes demonstrated significant differences between groups: insulin-like growth factor 1 (IGF1) and matrix metallopeptidase 2 (MMP2) were significantly elevated in most injured horses, regardless of injury classification, whereas and interleukin 1 receptor antagonist (IL1RN) was significantly decreased in horses with catastrophic sesamoid injuries. Since the blood samples were collected from injured horses within 30 minutes of injury and mRNA expression is not significantly altered until ≥ 6 hours in experimental models of injury across species ^40–42, these markers may identify future Thoroughbreds at risk for catastrophic injury. One consideration is that many “not injured” mRNA profiles overlapped with injured profiles for these mRNA biomarkers (Fig. 3, ³⁹). While these horses may still be at risk for catastrophic injury and were simply fortunate that this did not occur in that particular race or in the following three months where the horses were followed, increased specificity may be required in order to implement decisions on whether or not a horse should race that day based on mRNA biomarker testing.

Another class of RNAs, microRNAs (miRNAs), bind to target mRNAs and regulate gene expression and are routinely profiled in many human diseases ⁴³. To date, three miRNA studies have profiled equine musculoskeletal disease (tendon injury, laminitis, osteochondrosis) but none were related to equine stress fractures (reviewed in ^44,45).

While equine genomics and transcriptomics continues to evolve, improvements in the annotation of the equine genome will undoubtably accelerate the rate of discovery. With the need for large sample sizes of well-phenotyped horses to study most complex diseases, equine genomics and transcriptomics research will likely become increasingly collaborative, similar to the current status of human genomics initiatives. Aligned with this collaborative effort is the strong need for publicly available genomic and transcriptomic data that is available to all researchers. Only through the promotion of open science can the field of equine genetics and genomics advance and reproducibility be achieved ^46,47.

Data availability statement

Data sharing is not applicable to this article as no new data were created or analysed in this study.