Polymorphisms in the CTLA4 promoter sequence are associated with canine hypoadrenocorticism

Background Canine hypoadrenocorticism is an immune-mediated endocrinopathy that shares both clinical and pathophysiological similarities with Addison’s disease in humans. Several dog breeds are overrepresented in the disease population, suggesting that a genetic component is involved, although this is likely to be polygenic. Previous research has implicated CTLA4 as a potential susceptibility gene. CTLA4 is an important regulator of T cell function and polymorphisms/mutations in CTLA4 have been associated with a number of autoimmune phenotypes in both humans and rodent models of autoimmunity. The aim of the current study was to undertake a case:control association study of CTLA4 promotor polymorphisms in three dog breeds, cocker spaniels, springer spaniels and West Highland white terriers (WHWT). Results Polymorphisms in the CTLA4 promoter were determined by PCR and sequence-based typing. There were significant associations with three promoter haplotypes in cocker spaniels (p = 0.003). A series of SNPs were also associated with hypoadrenocorticism in cocker spaniels and springer spaniels, including polymorphisms in predicted NFAT and SP1 transcription factor binding sites. Conclusions This study provides further evidence that CTLA4 promotor polymorphisms are associated with this complex genetic disease and supports an immune mediated aetiopathogenesis of canine hypoadrenocorticism.


Plain English summary
Around one in 500 dogs are affected by Addison's disease (hypoadrenocorticism), a lifelong, life threatening, disease that also less commonly affects humans. Addison's disease is caused by a lack of cortisol (stress hormone) and other hormones due to destruction of the adrenal gland that makes these hormones. The most common cause of adrenal gland destruction is autoimmunity, meaning the body's immune system attacks its own gland, similar to type 1 diabetes or rheumatoid arthritis. The reasons why this happens are not fully understood.
In this study we looked at genetic differences in dogs specifically in the area controlling expression (promotor region) of a gene, CTLA4. CTLA4 is vital to immune system function. Specifically, CTLA4 acts as a brake on the immune system, so less CTLA4 means less of a brake and therefore possibly more autoimmunity. We showed changes in this promotor region are associated with increased risk of Addison's in springer spaniels and cocker spaniels. This highlights CTLA4 as likely one of the genetic reasons why some dogs get this disease and others do not.
By better understanding the underlying genetic reasons for Addison's disease in dogs we aim to be able to better help dogs with the disease and hopefully help understand the human disease better too.

Background
Canine hypoadrenocorticism, in common with human autoimmune Addison's disease (hAAD), results from a deficiency in production of steroid hormones from the adrenal gland [1]. Clinical signs associated with adrenal insufficiency are varied and dogs may present in an 'Addisonian crisis' with electrolyte disturbances, collapse and shock. Hypoadrenocorticism is a potentially fatal disease, which appears to be under diagnosed [2,3]. The pathology in the adrenal gland is consistent with an autoimmune aetiology, characterised, at least in the early stages, by immune cell infiltration, similar to that seen in hAAD [2,4]. Furthermore, circulating autoantibodies, which can be long lived, have been demonstrated in a subset of dogs, and are linked to a susceptibility genotype [5,6].
In man, the primary susceptibility locus associated with hAAD, is the human leukocyte antigen (HLA), encoding major histocompatibility complex (MHC) molecules [7]. Variation in other immune response genes have also been shown to be involved in susceptibility to hAAD, including the protein tyrosine phosphatase, non-receptor 22 (PTPN22) gene, which is involved in intracellular T cell receptor signalling [8,9] and cytotoxic T-lymphocyteassociated protein 4 (CTLA4) [8,10]. Due to its important role in immune regulation and tolerance, CTLA4 been posited as a general autoimmune susceptibility locus by several authors [10][11][12][13]. CTLA4 polymorphisms have been extensively studied in the human population and consistent associations have been found with hAAD [14][15][16] including a Europe-wide meta-analysis which demonstrated a significant association between hAAD and a SNP in exon 1, within the signal peptide of the coding sequence [17] and associations with a promoter polymorphism and hAAD in the Norwegian population [16,18]. However, the relationship between CTLA4 polymorphisms and hAAD is not found in all ethnic groups [14,19] and it is likely that within different populations, different polymorphisms exert different effects within a complex genetic landscape [18,20].
In dogs, genetic susceptibility to hypoadrenocorticism has been evidenced through breed predispositions and further revealed by pedigree analyses [21][22][23][24][25][26]. More focussed dissection of underlying genetic factors has shown dog leukocyte antigen (DLA) class II variation to be significantly associated with hypoadrenocorticism [5,[27][28][29]. Whilst the significance of these findings have been contested [30,31] it is the case that fine mapping techniques have confirmed MHC associations in human studies as well as revealing further associations and the importance of epistasis [32].
In early genetic studies of canine hypoadrenocorticism, microsatellite markers in the vicinity of CTLA4 were shown to be associated with hypoadrenocorticism in Portuguese Water Dogs (PWDs) and Nova Scotia duck tolling retrievers [24,33]. However, when the sequence around CTLA4 was analysed in PWDs, three haplotypes were identified that did not segregate convincingly with hypoadrenocorticism, although further details were not provided [24]. More recent candidate gene studies, undertaken across several breeds, have revealed a SNP in the CTLA4 promoter region, to be associated with hypoadrenocorticism in springer spaniels but not in other breeds [34,35]. Although there are no reported SNPs in the canine CTLA4 coding sequence, the promoter region has been characterised in some detail [36,37], revealing 20 SNPs and three indels within the space of 1.6 kB upstream of the start codon. These variants segregate into 17 distinct haplotypes. Significant allele and haplotype associations have been reported between the CTLA4 promoter region and diabetes [36], although not for immune-mediated haemolytic anaemia [37]. This high degree of polymorphism in the canine CTLA4 promotor has the potential to impact on protein expression, which may, in turn, influence its immune regulatory function and susceptibility to autoimmunity. Our hypothesis is that CTLA4 plays a role in the immunopathogenesis of canine hypoadrenocorticism and the study aimed to examine CTLA4 promotor alleles and haplotypes in selected breeds.

Materials and methods
Samples consisted of the residual volume of blood samples following completion of diagnostic testing at the Royal Veterinary College (RVC) Diagnostic Laboratories (Hertfordshire, UK) or NationWide Laboratories (NWL) (Poulton-le-Fylde, UK); blood samples were not specifically taken for this study in line with the ethical approval for the project and United Kingdom law. Furthermore, due to the nature of the ethical approval for this project (i.e. submitted samples were required to be de-identified) and to comply with UK data protection regulations, it was not possible to explore the relatedness of the animals, whose DNA samples were genotyped in the study. Further genomic DNA (gDNA) samples from dogs affected by hypoadrenocorticism were provided from the UK DNA Archive for Companion Animals (Universities of Liverpool and Manchester). A cohort of affected dogs comprised the populations described in previous genetic analyses in hypoadrenocorticism, [34,35]. The control (unaffected) dogs are unique to this paper and were accessed from the RVC Genetic Archive. The cocker spaniels affected with hypoadrenocorticism are common to both studies; ten affected springer spaniels are unique to this study and four affected WHWT are unique to this study.
Dogs affected with hypoadrenocorticism were identified from clinical and diagnostic records with pre-and post-ACTH stimulation test results consistent with corticosteroid deficiency (both < 27.6 nmol/L) and no known history of recent steroid administration. Genomic DNA from breed matched control dogs over 9 years old with no known history of endocrinopathy or immunemediated disease were selected from the RVC Genetic Archive. All samples were handled and processed according to local laboratory protocols and procedures until they entered the study at which point they were stored at − 20°C in RNase, DNase, DNA and PCRinhibitor free polypropylene tubes. Transfer of samples between locations was performed frozen on dry ice.
The RVC has ethical approval for residual clinical material, taken for diagnostic purposes, to be used for research with informed owner consent. NationWide Laboratories has approval for utilising clinical material for development of diagnostic assays, provided that anonymity is maintained and data protection is observed.
PCR products were separated by horizontal gel electrophoresis in 2% agarose dissolved in Tris/borate/ EDTA (TBE) buffer (both from Bioline) containing 1.5 μL/mL SafeView™ Nucleic Acid Stain (NBS Biologicals Ltd., Huntingdon, UK) and visualised under 590 nm UV light (ImageMaster VDS®, Pharmacia Biotech/GE Healthcare, Buckinghamshire, UK). Amplicons were excised and DNA extracted using the GenElute Gel Extraction Kit (Sigma-Aldrich, UK) according to the manufacturer's instructions. PCR products were submitted for Sanger sequencing (Source Bioscience, UK) using the PCR primers and two additional primers for internal sequencing (CTLA4 Promoter Sequencing 5′-TCTGTT TGCCTGTCAGTCTCC-3′ and CTLA4 promoter Sequencing 2 5′-TGCAGTTAATGCCTTAAGGGA-3′). Chromatograms were analysed using BioEdit Sequence Alignment Editor Software. Previously documented SNPs and indels were analysed and haplotypes assigned for each dog (Table 1) [37]. The call rate for all markers was 100%, achieved through repeat PCR and sequencing when necessary.
Data were organised in Microsoft® Excel© 2013 version 15 (Microsoft Corporation, Redmond, WA, USA) and statistical analyses performed using PLINK version 1.07 1 (Shaun Purcell) and IBM© SPSS© Statistics for Windows, Version 24.0 (IBM Corp, Armonk, NY, USA). An exact method was used for the testing of Hardy Weinberg equilibrium, due to the presence of rare genotypes [38]. Categorical data were analysed using contingency tables, with Fisher's exact test used for comparisons. Permutation testing (10,000 permutations using the max (T) method in PLINK) was used to correct for multiple testing when individual markers were assessed. The cut off for significance was set at p < 0.05.
Sex and age information was only available for 10 springer spaniel cases, seven female and three male, which was not significantly different to the control population (p = 0.168) comprising of 24 female and 33 male dogs. A significant age difference was found in springer spaniels (p < 0.001) with controls (572; 468-728 wk) older than cases (364; 156-520 wk). The medical conditions for which the control population presented included 15 dogs with a neurological condition, including seven with intervertebral disc disease (IVDD) and three with epilepsy, four dogs had an orthopaedic complaint, four infections, two each with neoplasia, cardiac disease, chronic renal failure and a toxic insult. Several dogs had more than one concurrent medical complaint.
Sex and age information was only available for 12 of the cocker spaniel cases, ten female and two male, which was significantly different to the control population (p = 0.049) comprised of 23 female and 35 male dogs. A significant age difference was found found in the cocker spaniels (p = 0.001) with controls (520; 468-728 wk) older than cases (286; 156-598 wk). The medical conditions for which the control population presented included 20 dogs with neoplasia, 12 with IVDD, six with cardiac disease, two each with epilepsy and orthopaedic disease. Several dogs had more than one concurrent medical complaint.
Sex and age information was only available for 14 of the WHWT cases, 10 female and four male which was not significantly different to the control population (p = 0.051) comprised of 11 female and 20 male dogs. A significant age difference was found between the WHWT groups (p < 0.001) with controls (588; 486 -780wk) older than cases (217; 13-464 wk). The medical conditions for which the control population presented included nine dogs with respiratory disease, of which seven had idiopathic pulmonary fibrosis, five dogs with renal disease, four with orthopaedic disease, four with neurological disease and two each with oesophageal foreign bodies and trauma. Several dogs had more than one concurrent medical complaint.
Minor allele frequencies are shown in Table 2, showing some degree of disparity between the three breeds. All variants were in Hardy Weinberg equilibrium for case, control and whole populations for cocker spaniels and springer spaniels. For WHWT, 13/23 variants were not in equilibrium for the case population, whilst being in equilibrium within the control population (Table 3); due to this the WHWT haplotypes and nucleotide variants were not assessed further.

Genotype and haplotype associations
Testing for association of genotype (p = 0.669) and haplotype (p = 0.291) with hypoadrenocorticism in springer spaniels did not reveal any overall significant associations, although haplotype 2 was overrepresented in post-hoc testing (Table 4). No effect of homozygosity was found (p = 0.474). For cocker spaniels, there were significant associations between disease phenotype and CTLA4 promoter genotype (p = 0.006) with dogs homozygous for haplotype 8 more likely in controls (43/51) than cases (10/ 19) (p = 0.0034) no other significant genotype associations were found. Cocker spaniel also had significant associations with haplotype (p = 0.003) with haplotypes 3 (OR = 4.59, p = 0.0247) and 12 (OR = 11.8, p = 0.0193) associated with increased odds of hypoadrenocorticism and haplotype 8 (OR = 0.20, p = 0.0038) associated with decreased odds of hypoadrenocorticism (Table 5). Cases were more likely to be heterozygous (p = 0.011).

Specific allele associations
To further elucidate the role of specific polymorphisms within associated haplotypes in cocker spaniels and to determine whether any polymorphisms segregated with hypoadrenocorticism outside full haplotypes in springer spaniels, SNPs and indels were analysed individually. In springer spaniels 14/23 markers segregated with disease, 11/14 were in linkage disequilibrium and the minor alleles were associated with hypoadrenocorticism (OR 2.16, p = 0.026). Of the further 3/14 markers, for SNP7 (OR = 2.5, p = 0.008) and SNP9 (OR = 2.19, p = 0.027), the minor allele was associated with hypoadrenocorticism and for indel3, the minor allele was associated with decreased odds of hypoadrenocorticism (OR = 0.5, p = 0.031) ( Table 6). Three markers segregated with hypoadrenocorticism in the cocker spaniels, with the minor allele in SNP13 (OR = 11.88, p = 0.016), SNP14 (OR = 3.61, p = 0.031) and indel3 (OR = 4.43, p = 0.006), associated with an increased risk of hypoadrenocorticism (Table 7).

Discussion
Within canine hypoadrenocorticism, two previous studies analysed genetic variants in the region of the CTLA4 locus [24,33] and two studies focussed on a single CTLA4 promoter SNP as a susceptibility marker (SNP 16 in this study) [34,35]. This study was therefore designed to conduct a more detailed analysis of variation in the CTLA4 promoter region in dogs with hypoadrenocorticism. Two populations of dog with hypoadrenocorticism in this study, cocker spaniels and springer spaniels, overlap with previous analyses [34 , 35]. However, this study has expanded case populations, novel control populations and has extended the analysis from a single SNP to a wider, previously described, promoter region [37]. The control populations used in this study were selected based on a more stringent phenotyping compared to prior studies [34,35]. Despite a more rigorous inclusion/ exclusion criteria, the phenotyping and selection of unaffected dogs was still limited by the information available from the available clinical records. It is possible that dogs might have had historical or concurrent disease(s) not recorded in the electronic patient record. Incorrect phenotyping in this study would lead to the erroneous inclusion of individuals with immune-mediated conditions; which might decrease the power of this study to identify a difference rather than bias towards a type one error. In this study the case phenotyping is robust given the nature of the diagnostic criteria; however, any incorrect phenotyping would also bias towards a type two error.
Permutation analysis was used to correct for familywise error in this study as well as false discovery rate correction. Given the analysis of a single locus with several highly linked markers, this choice seems appropriate and caution must be used in interpreting the results.
Genotyping errors from low call rates and missing samples can limit or bias genetic analyses with these effects exaggerated in candidate gene studies with relatively small numbers [39]. Every effort was made to identify and call each polymorphism in this study leading to a 100% call rate eliminating this as a source of bias.
The WHWT case population selected for this study was found not to be in HWE. Inferences of significant associations with disease can made regarding deviation from HWE in case but not control populations [40,41]. However, deviation from HWE can also be caused by a number of factors, including genotyping errors, population stratification and sampling methods. Sampling in this study was performed through two centres, therefore population stratification is possible, with local genetic pools of dogs contributing to these differences. As outlined in the methods, no further exploration of relatedness of dogs was possible. These data should therefore be interpreted with caution, both due to the small sample size and possible stratification issues [40]. It remains possible that the loss of HWE in the case population reflects an association of the SNPs and indel2 with hypoadrenocorticism.
Springer spaniels were not found to have significant associations between hypoadrenocorticism and genotype, heterozygosity or haplotypes. Significant associations were found with several variants, primarily with SNPs making up haplotype 2 which was overrepresented in post-hoc haplotype analysis (Table 4). Haplotype 2 only differs from haplotype 3, overrepresented in the closely related cocker spaniels with hypoadrenocorticism, by a G to A substitution at SNP6 ( Table 1). The potential significance of possibly biologically relevant individual variants is discussed later.
In cocker spaniels, a homozygous 8/8 genotype was found to be associated with an absence of hypoadrenocorticism. Following this, haplotypes 3 and 12 were both identified as risk haplotypes and haplotype 8 as a protective haplotype with affected dogs more likely to be heterozygotic. Haplotype 3 was not identified as a risk haplotype for diabetes mellitus in a previous study [36]. However, the closely related haplotype 2 was associated with diabetes mellitus in Border terriers with an opposite association in West Highland white terriers. Haplotype 12 has not been identified as a risk haplotype previously but reported as protective for diabetes in Samoyed and miniature Schnauzers, and neutral in WHWT and crossbreed dogs. Haplotype 8, a protective haplotype for indel3 + 0.14 -0.49 + 0.12  hypoadrenocorticism in cocker spaniels, has also been reported as protective for diabetes in Border terriers; however, it was neutral in Labradors, miniature Schnauzers, West Highland white terriers and crossbreed dogs and associated with an increased risk of diabetes in Samoyeds. Across eight breeds and three diseases, the same SNPs and haplotypes have been variously described as conferring risk, being protective or neutral [36,37]. There are a number of possible reasons for these findings. Firstly, these risks may be disease specific with hypoadrenocorticism risks changing differently to those for IMHA or diabetes or the gene may or may not be involved in the aetiology and pathogenesis of any specific disease.
The same haplotype could have a different relative risk profile between different breeds; for example, if haplotype 3 is relatively protective to haplotype 8 and haplotype 8 is relatively protective to haplotype 12, then haplotype 8 will appear as either increasing or decreasing risk depending on the other haplotypes present in the breed and in what proportions. Breed specific genetic backgrounds might also influence the effect of CTLA4 promoter variation, due to epistasis. A specific example of epistasis and CTLA4 is known in humans, between CTLA4 and TNF-alpha. A specific TNF-alpha variant is only associated with an altered risk of primary biliary cirrhosis when there is an AA genotype, not AG or GG, at the − 318 CTLA4 promoter polymorphism [42]. Breed differences may be similar to those seen in different ethnic groups in the human population, e.g. different CTLA4 3'UTR (AT) n associations are apparent in Japanese and Portuguese patients with SLE [43], or different associations of CTLA4 polymorphisms between distinct populations with Addison's disease [18]. The interactions between breed, specific epistasis and disease are likely to be complex and require deeper genetic analysis in large cohorts of well phenotyped dogs to understand.
Although transcription factor-binding sites are not fully conserved between species [44] cross-species comparisons can be used to identify potential regulatory sites [45], possibly due to conserved core binding motifs [46]. There has been little research undertaken on the nature of canine response elements and transcription factor binding characteristics, therefore interpretation of effects of SNPs on binding sites was undertaken based upon conserved aspects of promoter biology sourced from the JASPAR database using the online Transcription Element Search System (TESS; www.cbil.upenn.edu/tess) and literature searches. Since computational methods used alone can deliver high false positive rates [47,48] below we have focussed on selected sites of potential biological significance.
Similar to the findings for springer spaniels and hypoadrenocorticism in this study, SNP13T has previously been associated with increased risk of diabetes in Labradors [34]; however, this variant was also associated with a decreased diabetes risk in miniature Schnauzers and Samoyeds. SNP13 is located at the start of a GATA-1 binding site [49]. GATA-1 has been shown to upregulate CTLA4 transcription and to be able to initiate a substantial component of the Treg transcriptional signature, when expressed in conjunction with FoxP3 in nonregulatory murine T cells [50]. CTLA4 promoter SNP14 has not previously been associated with disease risk, however unusually it was found to change minor allele status between breeds [36]. SNP14 lies adjacent to a core SP1 promoter site. A polymorphism in the promoter region of ICOS in humans is associated with an SP1 site [51] and SP1 is known to affect CIITA transcription in dendritic cells [52]. SP1 also appears to bind upstream of Foxp3 in nTregs, which indicates its potential importance in immune responses. SP1 binding can protect CpG sites from methylation [53] and the SNP14A  polymorphism removes a CpG motif from the canine CTLA4 promoter. This is the first association between the CTLA4 promoter indel3 and canine disease, with the presence of a C at indel3 being the most significantly associated risk marker for hypoadrenocorticism in cocker spaniels (p = 0.0097) and springer spaniels (p = 0.031). Indel3 is situated between a potential NFAT binding site [54] and a putative FoxP3 site [55], which are separated by 12 or 13 bases depending on the indel3 status (Fig. 1). NFAT typically binds to DNA with a partner, e.g. FoxP3 [56] and both FoxP3 and NFAT response elements are also located close together in the CTLA4 promoter in humans [55,57]. In mice, FoxP3 and NFAT co-operate in upregulating CTLA4 expression and disrupting interaction between these two transcription factors impairs CTLA4 transcription, impacting on Treg function [58]. Taking into consideration previous findings in other species, the association of SNP16G with canine diabetes mellitus and hypoadrenocorticism and the association of the C insertion with hypoadrenocorticism in cocker spaniels and springer spaniels, this small region of the canine CTLA4 promoter is very interesting for further investigation.
SNP7, associated with hypoadrenocorticism in springer spaniels, is adjacent to a predicted binding site for T cell factor 1 (TCF-1), one of a family of TCF/ Lymphoid enhancer-binding factor (LEF) transcription factors forming part of the wnt signalling pathway [59]. TCF is particularly important during thymic development [60], which might be mediated by wnt3a [61]. The wnt signalling pathway, including TCF-1 and LEF-1, can also modulate mature T cell function, including prolonging regulatory T cell (Treg) survival and therefore, potentially impacting on autoimmunity [61]. Four potential TCF/LEF sites have been identified within the 800 bp upstream of the start codon in human CTLA4 [62], one of which contains the C(− 318) T SNP [63] associated with altered promoter activity [63,64] and with susceptibility to Grave's disease [65], autoimmune pancreatitis [66] but not diabetes [67,68]. In one study, when a melanoma cell line was treated with wnt3a, CTLA4 was the most up-regulated gene, which was proposed to represent a mechanism of immune evasion [62]. In Jurkat and HeLa cell lines, addition of LEF-1 has been shown to increased CTLA4 promoter activity, with the -318 T variant consistently more active than the -318C variant [63]. It is highly likely that some of the promoter variants identified do not confer any functional consequences for expression of canine CTLA4. In mice, seven intronic SNPs and three intronic indels have been found, but these were not associated with any functional outcomes [69]. Also, any polymorphisms that segregate with hypoadrenocorticism could be merely in linkage disequilibrium with enhancers, silencers, insulators or other locus control regions [70]. In a large study of the human CTLA4 region, 108 SNPs were typed, 78 of which were in a 100 kb CTLA4 LD block, only 23 of these were not significantly associated with Graves' disease at p < 0.05 and mathematical modelling was used to determine which variation was most likely functionally associated with disease [71]. Given that LD blocks in dogs are larger than in human beings [72] the possibility of genes in this region (especially ICOS or CD28) being involved remains, although a microsatellite marker flanking canine CD28 was not associated with hypoadrenocorticism in NSDTRs in one study [33].

Conclusions
This study of canine CTLA4 promoter variants in three dog breeds adds further detail to previous research and provides some evidence that CTLA4 may from part of the susceptibility aetiology of canine hypoadrenocorticism in some breeds. This work supports the belief that the pathophysiology of hypoadrenocorticism is immunemediated. This study represents a further step in unravelling the complex genetics of hypoadrenocorticism and in establishing a potential role for CTLA4.