Comparison of buccal and blood-derived canine DNA, either native or whole genome amplified, for array-based genome-wide association studies

A cautionary tale of low-pass sequencing and imputation with respect to haplotype accuracy

BACKGROUND

Low-pass whole-genome sequencing and imputation offer significant cost savings, enabling substantial increases in sample size and statistical power. This approach is particularly promising in livestock breeding, providing an affordable means of screening individuals for deleterious alleles or calculating genomic breeding values. Consequently, it may also be of value in companion animal genomics to support pedigree breeding. We sought to evaluate in dogs the impact of low coverage sequencing and reference-guided imputation on genotype concordance and association analyses.

RESULTS

DNA isolated from saliva of 30 Labrador retrievers was sequenced at low (0.9X and 3.8X) and high (43.5X) coverage, and down-sampled from 43.5X to 9.6X and 17.4X. Genotype imputation was performed using a diverse reference panel (1021 dogs), and two subsets of the former panel (256 dogs each) where one had an excess of Labrador retrievers relative to other breeds. We observed little difference in imputed genotype concordance between reference panels. Association analyses for a locus acting as a disease proxy were performed using single-marker (GEMMA) and haplotype-based (XP-EHH) tests. GEMMA results were highly correlated (r ≥ 0.97) between 43.5X and ≥ 3.8X depths of coverage, while for 0.9X the correlation was lower (r ≤ 0.8). XP-EHH results were less well correlated, with r ranging from 0.58 (0.9X) to 0.88 (17.4X). Across a random sample of 10,000 genomic regions averaging 17 kb in size, we observed a median of three haplotypes per dog across the sequencing depths, with 5% of the regions returning more than eight haplotypes. Inspection of one such region revealed genotype and phasing inconsistencies across sequencing depths.

CONCLUSIONS

We demonstrate that saliva-derived canine DNA is suitable for whole-genome sequencing, highlighting the feasibility of client-based sampling. Low-pass sequencing and imputation require caution as incorrect allele assignments result when the subject possesses alleles that are absent in the reference panel. Larger panels have the capacity for greater allelic diversity, which should reduce the potential for imputation error. Although low-pass sequencing can accurately impute allele dosage, we highlight issues with phasing accuracy that impact haplotype-based analyses. Consequently, if accurately phased genotypes are required for analyses, we advocate sequencing at high depth (> 20X).

Validation of a Chromosome 14 Risk Haplotype for Idiopathic Epilepsy in the Belgian Shepherd Dog Found to Be Associated with an Insertion in the RAPGEF5 Gene

An idiopathic epilepsy (IE) risk haplotype on canine chromosome (CFA) 14 has been reported to interact with the CFA37 common risk haplotype in the Belgian shepherd (BS). Additional IE cases and control dogs were genotyped for the risk haplotypes to validate these previous findings. In the new cohort, the interaction between the two regions significantly elevated IE risk. When the haplotypes were analyzed individually, particular haplotypes on both CFA14 (ACTG) and 37 (GG) were associated with elevated IE risk, though only the CFA37 AA was significantly associated (p < 0.003) with reduced risk in the new cohort. However, the CFA14 ACTG risk was statistically significant when the new and previous cohort data were combined. The frequency of the ACTG haplotype was four-fold higher in BS dogs than in other breeds. Whole genome sequence analysis revealed that a 3-base pair predicted disruptive insertion in the RAPGEF5 gene, which is adjacent to the CFA14 risk haplotype. RAPGEF5 is involved in the Wnt-β-catenin signaling pathway that is crucial for normal brain function. Although this risk variant does not fully predict the likelihood of a BS developing IE, the association with a variant in a candidate gene may provide insight into the genetic control of canine IE.

Whole Genome Sequencing Reveals Multiple Linked Genetic Variants on Canine Chromosome 12 Associated with Risk for Symmetrical Lupoid Onychodystrophy (SLO) in the Bearded Collie

In dogs, symmetrical lupoid onychodystrophy (SLO) results in nail loss and an abnormal regrowth of the claws. In Bearded Collies, an autoimmune nature has been suggested because certain dog leukocyte antigen (DLA) class II haplotypes are associated with the condition. A genome-wide association study of the Bearded Collie revealed two regions of association that conferred risk for disease: one on canine chromosome (CFA) 12 that encompasses the DLA genes, and one on CFA17. Case-control association was employed on whole genome sequencing data to uncover putative causative variants in SLO within the CFA12 and CFA17 associated regions. Genotype imputation was then employed to refine variants of interest. Although no SLO-associated protein-coding variants were identified on CFA17, multiple variants, many with predicted damaging effects, were identified within potential candidate genes on CFA12. Furthermore, many potentially damaging alleles were fully correlated with the presence of DLA class II risk haplotypes for SLO, suggesting that the variants may reflect DLA class II haplotype association with disease or vice versa. Strong linkage disequilibrium in the region precluded the ability to isolate and assess the individual or combined effect of variants on disease development. Nonetheless, all were predictive of risk for SLO and, with judicious assessment, their application in selective breeding may prove useful to reduce the incidence of SLO in the breed.

Genetic characterization of Addison's disease in Bearded Collies

BACKGROUND

Primary hypoadrenocorticism (or Addison's disease, AD) is an autoimmune disease that results in destruction of the adrenal cortex and consequent adrenal insufficiency. The disease has been described in purebred and mixed breed dogs, although some breeds, including the Bearded Collie, are at increased risk for AD. Candidate gene approaches have yielded few associations that appear to be breed-specific. A single other genome-wide association study reported no significant regions of association for AD in Standard Poodles. The present study aimed to identify genomic regions of association for canine AD in Bearded Collies.

RESULTS

Our study consists of the first genome-wide association analysis to identify a genome-wide significant region of association with canine AD (CFA18). Peaks of suggestive association were also noted on chromosomes 11, 16 and 29. Logistic regression analysis supported an additive effect of risk genotypes at these smaller effect loci on the probability of disease associated with carrying a risk genotype on CFA18. Potential candidate genes involved in adrenal steroidogenesis, regulation of immune responses and/or inflammation were identified within the associated regions of chromosomes 11 and 16. The gene-poor regions of chromosomes 18 and 29 may, however, harbor regulatory sequences that can modulate gene expression and contribute to disease susceptibility.

CONCLUSION

Our findings support the polygenic and complex nature of canine AD and identified a strongly associated locus on CFA18 that, when combined with three other smaller effect loci, was predictive of disease. The results offer progress in the identification of susceptibility loci for canine AD in the Bearded Collie. Further studies are needed to confirm association with the suggested candidate genes and identify actual causative mutations involved with AD susceptibility in this breed.

Genome-wide association analysis of idiopathic epilepsy in the Belgian shepherd

Background

Idiopathic epilepsy (IE) is a common neurological disorder in the domestic dog, and is defined as repeated seizure activity having no identifiable underlying cause. Some breeds, such as the Belgian shepherd dog, have a greater prevalence of the disorder. Previous studies in this and other breeds have identified ADAM23 as a gene that confers risk of IE, although additional loci are known to exist. The present study sought to identify additional loci that influence IE in the Belgian shepherd dog.

Results

Genome-wide association studies (GWAS) revealed a significant association between IE and CFA 14 (p < 1.03 E− 08) and a suggestive association on CFA 37 (p < 2.91 E− 06) in a region in linkage disequilibrium with ADAM23. Logistic regression identified a 2-loci model that demonstrated interaction between the two chromosomal regions that when combined predicted IE risk with high sensitivity.

Conclusions

Two interacting loci, one each on CFAs 14 and 37, predictive of IE in the Belgian shepherd were identified. The loci are adjacent to potential candidate genes associated with neurological function. Further exploration of the region is warranted to identify causal variants underlying the association. Additionally, although the two loci were very good at predicting IE, they failed to capture all the risk, indicating additional loci or incomplete penetrance are also likely contributing to IE expression in the Belgian shepherd dog.

Impact of DNA source on genetic variant detection from human whole-genome sequencing data

BACKGROUND

Whole blood is currently the most common DNA source for whole-genome sequencing (WGS), but for studies requiring non-invasive collection, self-collection, greater sample stability or additional tissue references, saliva or buccal samples may be preferred. However, the relative quality of sequencing data and accuracy of genetic variant detection from blood-derived, saliva-derived and buccal-derived DNA need to be thoroughly investigated.

METHODS

Matched blood, saliva and buccal samples from four unrelated individuals were used to compare sequencing metrics and variant-detection accuracy among these DNA sources.

RESULTS

We observed significant differences among DNA sources for sequencing quality metrics such as percentage of reads aligned and mean read depth (p<0.05). Differences were negligible in the accuracy of detecting short insertions and deletions; however, the false positive rate for single nucleotide variation detection was slightly higher in some saliva and buccal samples. The sensitivity of copy number variant (CNV) detection was up to 25% higher in blood samples, depending on CNV size and type, and appeared to be worse in saliva and buccal samples with high bacterial concentration. We also show that methylation-based enrichment for eukaryotic DNA in saliva and buccal samples increased alignment rates but also reduced read-depth uniformity, hampering CNV detection.

CONCLUSION

For WGS, we recommend using DNA extracted from blood rather than saliva or buccal swabs; if saliva or buccal samples are used, we recommend against using methylation-based eukaryotic DNA enrichment. All data used in this study are available for further open-science investigation.

Novel Locus Associated with Symmetrical Lupoid Onychodystrophy in the Bearded Collie

Symmetrical lupoid onychodystrophy (SLO) is characterized by inflammation of the nail bed and nail sloughing that causes affected dogs considerable pain. Disease etiology remains unclear, although an autoimmune component is suspected. A genome-wide association study on Bearded Collies revealed regions of association on canine chromosomes (CFA) 12 and 17. The large region of association on CFA12 likely consists of two smaller linked regions, both of which are also linked to the dog leukocyte antigen (DLA) class II genes. Dogs homozygous for the alternate allele at the top CFA12 SNP also carried two DLA class II risk haplotypes for SLO, and this locus explained most of the increased risk for disease seen throughout the CFA12 region of association. A stronger peak was seen on CFA17 when analysis was done solely on dogs that carried DLA class II risk haplotypes for SLO. The majority of SLO dogs carried a homozygous alternate genotype on CFA12 and at least one CFA17 risk haplotype. Our findings offer progress toward uncovering the genetic basis of SLO. While the contribution of the CFA17 region remains unclear, both CFA12 and CFA17 regions are significantly associated with SLO disease expression in the Bearded Collie and contain potential candidate genes for this disease.

DLA class II risk haplotypes for autoimmune diseases in the bearded collie offer insight to autoimmunity signatures across dog breeds

BACKGROUND

Primary hypoadrenocorticism (Addison's disease, AD) and symmetrical lupoid onychodystrophy (SLO) are two clinical conditions with an autoimmune etiology that occur in multiple dog breeds. In man, autoimmunity is associated with polymorphisms in immune-related genes that result in a reduced threshold for, or defective regulation of, T cell activation. The major histocompatibility complex (MHC) class II genes encode molecules that participate in these functions, and polymorphisms within these genes have been associated with autoimmune conditions in dogs and humans. Bearded collies have a relatively high prevalence of autoimmune diseases, particularly AD and SLO. Our study assessed the relationship between particular MHC (dog leukocyte antigen, DLA) class II haplotypes and the two autoimmune diseases most common in this breed. Moreover, five unrelated breeds at increased risk for AD were studied for comparative purposes and analyzed in the context of extant literature.

RESULTS

A single DLA class II three-locus haplotype, determined by sequence-based typing, was associated with increased risk for AD (DLA-DRB1*009:01/DQA1*001:01/DQB1*008:02) in bearded collies. Comparative analysis with the five additional breeds showed limited allele sharing, with DQA1*001:01 and DQB1*002:01 being the only alleles observed in all breeds. A distinct three-locus risk haplotype (DLA-DRB1*001:01/DQA1*001:01/DQB1*002:01) was associated with AD in the West Highland white terrier and Leonberger. Two different risk haplotypes were associated with increased risk for SLO in the bearded collie (DLA-DRB1*018:01/DQA1*001:01/DQB1*002:01 and DLA-DRB1*018:01/DQA1*001:01/ DQB1*008:02).

CONCLUSION

Two-locus DQ haplotypes composed of DLA-DQA1*001:01 in association with DLA-DQB1*002:01 or DLA-DQB1*008:02 make up the four risk haplotypes identified in the present study and are also found in other risk haplotypes previously associated with diabetes mellitus and hypothyroidism across different dog breeds. Our findings build upon previously published data to suggest that this two-locus (DQ) model serves as a good indicator for susceptibility to multiple organ-specific autoimmune diseases in the canine population. However, it is also clear that additional loci are necessary for actual disease expression. Investigation of affected and unaffected dogs carrying these predisposing DQ haplotype signatures may allow for the identification of those additional genetic components that determine autoimmune disease expression and organ specificity.

Protein expression and genetic variability of canine Can f 1 in golden and Labrador retriever service dogs

BACKGROUND

Valued for trainability in diverse tasks, dogs are the primary service animal used to assist individuals with disabilities. Despite their utility, many people in need of service dogs are sensitive to the primary dog allergen, Can f 1, encoded by the Lipocalin 1 gene (LCN1). Several organizations specifically breed service dogs to meet special needs and would like to reduce allergenic potential if possible. In this study, we evaluated the expression of Can f 1 protein and the inherent variability of LCN1 in two breeds used extensively as service dogs. Saliva samples from equal numbers of male and female Labrador retrievers (n = 12), golden retrievers (n = 12), and Labrador-golden crosses (n = 12) were collected 1 h after the morning meal. Can f 1 protein concentrations in the saliva were measured by ELISA, and the LCN1 5' and 3' UTRs and exons sequenced.

RESULTS

There was no sex effect (p > 0.2) nor time-of-day effect; however, Can f 1 protein levels varied by breed with Labrador retrievers being lower than golden retrievers (3.18 ± 0.51 and 5.35 ± 0.52 μg/ml, respectively, p < 0.0075), and the Labrador-golden crosses having intermediate levels (3.77 ± 0.48 μg/ml). Although several novel SNPs were identified in LCN1, there were no significant breed-specific sequence differences in the gene and no association of LCN1 genotypes with Can f 1 expression.

CONCLUSIONS

As service dogs, Labrador retrievers likely have lower allergenic potential and, though there were no DNA sequence differences identified, classical genetic selection on the estimated breeding values associated with salivary Can f 1 expression may further reduce that potential.

Overview of genetic research in anorexia nervosa: The past, the present and the future

BACKGROUND

Even though the evidence supporting the presence of a heritable component in the aetiology of anorexia nervosa (AN) is strong, the underlying genetic mechanisms remain poorly understood. The recent publication of a genome-wide association study (GWAS) of AN (Boraska, Mol Psychiatry, 2014) was an important step in genetic research in AN.

OBJECTIVE

To briefly sum up strengths and weaknesses of candidate-gene and genome-wide approaches, to discuss the genome-wide association studies of AN and to make predictions about the genetic architecture of AN by comparing it to that of schizophrenia (since the diseases share some similarities and genetic research in schizophrenia is more advanced).

METHOD

Descriptive literature review.

RESULTS

Despite remarkable efforts, the gene-association studies in AN did not advance our knowledge as much as had been hoped, although some results still await replication.

DISCUSSION

Continuous effort of participants, clinicians and researchers remains necessary to ensure that genetic research in AN follows a similarly successful path as in schizophrenia. Identification of genetic susceptibility loci provides a basis for follow-up studies.

Identification of a common risk haplotype for canine idiopathic epilepsy in the ADAM23 gene

Background

Idiopathic epilepsy is a common neurological disease in human and domestic dogs but relatively few risk genes have been identified to date. The seizure characteristics, including focal and generalised seizures, are similar between the two species, with gene discovery facilitated by the reduced genetic heterogeneity of purebred dogs. We have recently identified a risk locus for idiopathic epilepsy in the Belgian Shepherd breed on a 4.4 megabase region on CFA37.

Results

We have expanded a previous study replicating the association with a combined analysis of 157 cases and 179 controls in three additional breeds: Schipperke, Finnish Spitz and Beagle (p_c = 2.9e–07, p_GWAS = 1.74E-02). A targeted resequencing of the 4.4 megabase region in twelve Belgian Shepherd cases and twelve controls with opposite haplotypes identified 37 case-specific variants within the ADAM23 gene. Twenty-seven variants were validated in 285 cases and 355 controls from four breeds, resulting in a strong replication of the ADAM23 locus (p_raw = 2.76e–15) and the identification of a common 28 kb-risk haplotype in all four breeds. Risk haplotype was present in frequencies of 0.49–0.7 in the breeds, suggesting that ADAM23 is a low penetrance risk gene for canine epilepsy.

Conclusions

These results implicate ADAM23 in common canine idiopathic epilepsy, although the causative variant remains yet to be identified. ADAM23 plays a role in synaptic transmission and interacts with known epilepsy genes, LGI1 and LGI2, and should be considered as a candidate gene for human epilepsies.

Electronic supplementary material

The online version of this article (doi:10.1186/s12864-015-1651-9) contains supplementary material, which is available to authorized users.

Effectiveness of single-nucleotide polymorphisms to investigate cattle rustling

Short tandem repeats (STR)s have been the eligible markers for forensic animal genetics, despite single-nucleotide polymorphisms (SNP)s became acceptable. The technology, the type, and amount of markers could limit the investigation in degraded forensic samples. The performance of a 32-SNP panel genotyped through OpenArrays(TM) (real-time PCR based) was evaluated to resolve cattle-specific forensic cases. DNA from different biological sources was used, including samples from an alleged instance of cattle rustling. SNPs and STRs performance and repeatability were compared. SNP call rate was variable among sample type (average = 80.18%), while forensic samples showed the lowest value (70.94%). The repeatability obtained (98.7%) supports the used technology. SNPs had better call rates than STRs in 12 of 20 casework samples, while forensic index values were similar for both panels. In conclusion, the 32-SNPs used are as informative as the standard bovine STR battery and hence are suitable to resolve cattle rustling investigations.

Who is who? Identification of livestock predators using forensic genetic approaches

Molecular identifications of salivary DNA are increasingly applied in wildlife forensic investigations, and are successfully used to identify unknown livestock predators, or elucidate cases of large carnivore attacks to humans. In Europe most of livestock predations are attributed to wolves (Canis lupus), thought free-ranging dogs are sometime the responsible, and false predations are declared by breeders to obtain compensations. In this study we analyzed 33 salivary DNA samples collected from the carcasses of 13 sheep and a horse presumably predated by wolves in seven farms in central Italy. Reliable individual genotypes were determined in 18 samples (corresponding to samples from nine sheep and the horse) using 12 unlinked autosomal microsatellites, mtDNA control-region sequences, a male-specific ZFX/ZFY restriction-site and four Y-linked microsatellites. Results indicate that eight animals were killed by five wolves (a male and four different females), the ninth by a female dog while the horse was post-mortem consumed by a male dog. The genotype of one female wolf matched with the genetic profile of a female wolf that was non-invasively sampled 4 years before in the same area near livestock predation remains. Genetic identifications always supported the results of veterinary reports. These findings show that salivary DNA genotyping, together with detailed veterinary field and necropsy reports, provides evidence which helps to correctly identify species, gender and individual genetic profiles of livestock predators, thus contributing to clarify attack dynamics and to evaluate the impact of wolf predations on husbandry.

Twins, tissue, and time: an assessment of SNPs and CNVs

With the desire to assess genetic variation across the lifespan in large-scale collaborative projects, one question is whether inference of copy number (CN) is sensitive to the source of material for deoxyribonucleic acid (DNA) analysis (e.g., blood and buccal) and another question is whether CN is stable as individual sage. Here, we address these questions by applying Affymetrix 6.0 single nucleotide polymorphism (SNP)micro-arrays to 1,472 DNA samples from 710 individuals from the Netherlands Twin Register, including twin and non-twin individuals (372 with buccal and blood derived DNA and 388 with longitudinal data).Similar concordance for CN and genotype inference between samples from the same individual [or from the monozygotic (MZ) co-twins] was found for blood and buccal tissues. There was a small but statistically significant decrease in across-tissue concordance compared with concordance of samples from the same tissue type. No temporal effect was seen on CN variation from the 388 individuals sampled at two time points ranging from 1 to 12 years apart. The majority of our individuals were sampled at age younger than 20 years. Genotype concordance was very high (~ > 99%) between co-twins from 43 MZ pairs. For75 dizygotic (DZ) pairs, ~was ~65%. CN estimates were highly consistent between co-twins from MZ pairs for both deletions (f?2 ~ 90%) and duplications (~ ~ 86%). For DZ, these were similar for within-individual comparisons, but naturally lower between co-twins (~ ~ 50-60%). These results suggest that DNA from buccal samples perform as well as DNA from blood samples on the current generation of micro-array technologies.

Quantitative criteria for improving performance of buccal DNA for high-throughput genetic analysis

Background

DNA from buccal brush samples is being used for high-throughput analyses in a variety of applications, but the impact of sample type on genotyping success and downstream statistical analysis remains unclear. The objective of the current study was to determine laboratory predictors of genotyping failure among buccal DNA samples, and to evaluate the successfully genotyped results with respect to analytic quality control metrics. Sample and genotyping characteristics were compared between buccal and blood samples collected in the population-based Genetic and Environmental Risk Factors for Hemorrhagic Stroke (GERFHS) study (https://gerfhs.phs.wfubmc.edu/public/index.cfm).

Results

Seven-hundred eight (708) buccal and 142 blood DNA samples were analyzed for laboratory-based and analysis metrics. Overall genotyping failure rates were not statistically different between buccal (11.3%) and blood (7.0%, p = 0.18) samples; however, both the Contrast Quality Control (cQC) rate and the dynamic model (DM) call rates were lower among buccal DNA samples (p < 0.0001). The ratio of double-stranded to total DNA (ds/total ratio) in the buccal samples was the only laboratory characteristic predicting sample success (p < 0.0001). A threshold of at least 34% ds/total DNA provided specificity of 98.7% with a 90.5% negative predictive value for eliminating probable failures. After genotyping, median sample call rates (99.1% vs. 99.4%, p < 0.0001) and heterozygosity rates (25.6% vs. 25.7%, p = 0.006) were lower for buccal versus blood DNA samples, respectively, but absolute differences were small. Minor allele frequency differences from HapMap were smaller for buccal than blood samples, and both sample types demonstrated tight genotyping clusters, even for rare alleles.

Conclusions

We identified a buccal sample characteristic, a ratio of ds/total DNA <34%, which distinguished buccal DNA samples likely to fail high-throughput genotyping. Applying this threshold, the quality of final genotyping resulting from buccal samples is somewhat lower, but compares favorably to blood. Caution is warranted if cases and controls have different sample types, but buccal samples provide comparable results to blood samples in large-scale genotyping analyses.

Cheek swabs, SNP chips, and CNVs: assessing the quality of copy number variant calls generated with subject-collected mail-in buccal brush DNA samples on a high-density genotyping microarray

Background

Multiple investigators have established the feasibility of using buccal brush samples to genotype single nucleotide polymorphisms (SNPs) with high-density genome-wide microarrays, but there is currently no consensus on the accuracy of copy number variants (CNVs) inferred from these data. Regardless of the source of DNA, it is more difficult to detect CNVs than to genotype SNPs using these microarrays, and it therefore remains an open question whether buccal brush samples provide enough high-quality DNA for this purpose.

Methods

To demonstrate the quality of CNV calls generated from DNA extracted from buccal samples, compared to calls generated from blood samples, we evaluated the concordance of calls from individuals who provided both sample types. The Illumina Human660W-Quad BeadChip was used to determine SNPs and CNVs of 39 Arkansas participants in the National Birth Defects Prevention Study (NBDPS), including 16 mother-infant dyads, who provided both whole blood and buccal brush DNA samples.

Results

We observed a 99.9% concordance rate of SNP calls in the 39 blood–buccal pairs. From the same dataset, we performed a similar analysis of CNVs. Each of the 78 samples was independently segmented into regions of like copy number using the Optimal Segmentation algorithm of Golden Helix SNP & Variation Suite 7.

Across 640,663 loci on 22 autosomal chromosomes, segment-mean log R ratios had an average correlation of 0.899 between blood-buccal pairs of samples from the same individual, while the average correlation between all possible blood-buccal pairs of samples from unrelated individuals was 0.318. An independent analysis using the QuantiSNP algorithm produced average correlations of 0.943 between blood-buccal pairs from the same individual versus 0.332 between samples from unrelated individuals.

Segment-mean log R ratios had an average correlation of 0.539 between mother-offspring dyads of buccal samples, which was not statistically significantly different than the average correlation of 0.526 between mother-offspring dyads of blood samples (p=0.302).

Conclusions

We observed performance from the subject-collected mail-in buccal brush samples comparable to that of blood. These results show that such DNA samples can be used for genome-wide scans of both SNPs and CNVs, and that high rates of CNV concordance were achieved whether using a change-point-based algorithm or one based on a hidden Markov model (HMM).