Yun Yu

A familial cancer is like a curse to individuals in the family. In Dutch German Longhaired Pointer (GLP) dogs, a familial thyroid cancer (TC) was identified. Many GLPs were affected by, and died of, this familial TC. The aim of the research project described in this thesis was to eradicate this familial TC from the GLP population, and also to further our knowledge on causes of familial cancers in dogs.

In Chapter 2, we described the clinical details and histological diagnoses of all the affected GLPs diagnosed in the past ~20 years. The TCs identified in those 54 histologically diagnosed GLPs belong to thyroid follicular cell carcinoma (FCC) and manifested five sub-types: 1) follicular thyroid carcinoma (FTC), 2) compact thyroid carcinoma (CTC), 3) follicular-compact thyroid carcinoma (FCTC), 4) papillary thyroid carcinoma (PTC), and 5) carcinosarcoma. Most of the affected GLPs are closely related where 45 of them can be traced to a pair of first-half cousins. With the pedigree we estimated the heritability to be 0.62. According to the pedigree, this familial FCC has most likely an autosomal recessive inheritance pattern. Pedigree-based inbreeding was estimated for each dog, and it turned out that affected GLPs had higher inbreeding levels, suggesting that inbreeding contributed to incidence of the familial FCC.

In Chapter 3, I identified the germline risk mutations of the familial FCC in the Dutch GLPs. I performed a genome-wide association study and homozygosity mapping based on a combination of SNP array genotype and whole-genome sequencing (WGS) data to identify the target genomic region. Subsequently, I used WGS data from 11 affected and 11 unaffected GLPs to fine-map the potential causal variant(s) for the FCC. This yielded two deleterious mutations (chr17:800788G>A (S55F>V) and chr17:805275C>T (R45T>M)) in the TPO gene to be the germline risk mutations. We genotyped these two variants in 186 GLPs (59 affected and 127 unaffected) and revealed that homozygous recessive genotypes of these two SNPs confer a relative risk of 16.94 and 16.64 respectively, in comparison to homozygous wild-type genotypes.

In Chapter 4, I investigated the somatic mutations of the familial FCCs with the aim to identify driver mutations. I identified somatic single nucleotide variants (SNVs), insertions and deletions (InDels), structural variants (SVs), and copy number alternations (CNAs) using WGS data from FCC tissue and matched blood samples from 7 affected GLPs. Among somatic SNVs, I identified a recurrent deleterious mutation in the GNAS gene (chr24:41557087C>A, GNAS A204D) in 4 of 7 sequenced FCC tissues. Through Sanger sequencing, this somatic mutation was further identified in FCC tissues of 20 out of 32 GLPs. The high prevalence of the GNAS A204D mutation indicates that it is a promising driver mutation. I also found that this GNAS A204D somatic mutation is associated with lower somatic mutation burden. Meanwhile, to reveal potential mutational processes during tumorigenesis, I constructed mutational signatures of these FCCs and identified signatures that are similar to human SBS5 and SBS40, suggesting an endogenous mutagenesis factor in tumorigenesis.

In Chapter 5, I investigated the genetic diversity of Dutch GLPs and compared inbreeding levels between GLP and 11 other pointer setter breeds. I revealed that Dutch GLPs have relatively low inbreeding in comparison to the 11 other pointer setter breeds. Furthermore, I investigated the genetic relationship between GLP and those 11 pointer setter breeds and revealed good consistence between identified genetic relationship and breeding history of these breeds. Lastly, I identified the genomic selection signatures in GLPs using a runs of homozygosity (ROH) islands approach. I showed that a ROH segment identified on chromosome 30, harboring the RYRP, FMNN, and GREMN genes, might be selected for athletic performance.

In Chapter 6, I tested a new approach to use prior knowledge on signaling pathways to predict driver mutations. I calculated a cancer pathway score for each signaling pathway and then computed a cancer gene score for each gene. I showed that driver genes have higher cancer gene scores than passenger genes, implying that this cancer gene score is useful in distinguishing driver and passenger genes. I then trained Random Forest Classifier models using the cancer gene score as a feature, along with SIFT score, PolyPhen2 score, and recurrence of the mutation. On average, I observed a prediction accuracy of those trained Random Forest Classifiers, measured by F1 score (harmonic mean of precision and recall), of 0.90 (ranging between 0.85 - 0.94), demonstrating that these features, including the cancer gene score, can contribute to driver mutation prediction.

Finally, in Chapter 7, I brought major findings in chapter 2-6 together and discussed them in context. I discussed the utility of genetic test based on the PCR-RFLP experiment described in chapter 2 for the identification of GLPs predisposed to the FCC and how to use this genetic test to assist selective breeding for healthy GLPs. I connected the germline risks (chapter 3) and somatic driver mutation (chapter 4) and discussed the mechanisms underlying FCC development. Lastly, I discussed the advantage of using dogs as disease models and the specific value of GLPs studied in this thesis as cancer models.