Canine hemangiosarcoma (HSA), commonly referred to as angiosarcoma in humans, is a highly aggressive cancer originating from the vascular endothelium in dogs. Canine visceral HSA is very common, comprising up to 5% of all noncutaneous primary malignant tumors. Human angiosarcoma, however, is considered uncommon or rare, limiting resources for research. For both species, an accurate and rapid diagnostic method is needed to allow proper patient management. Especially in dogs, there is a need for a diagnostic test that can be used prior to clinical decision making, such as serum biomarkers, to differentiate HSA from benign lesions (HSA-like).
Biomarkers are biomolecules in the body whose presence reflects physiological or disease-induced changes. Proteomics-based biomarker discovery studies require a protein database. The current canine databases available consist mostly of proteins predicted by evidence at the transcript level. These canine databases include redundant sequences and are functionally poorly annotated. Therefore, the aims of this thesis were to 1) generate an in-house non-redundant canine protein database to facilitate MS/MS analysis on canine samples, 2) develop a glyco-biomarker discovery pipeline, 3) use the pipeline to identify and validate biomarkers for canine HSA and 4) functionally annotate the in-house canine protein database using the well-annotated human protein database. The annotated canine protein database was then used to compare functional aspects of top biomarker candidates identified for canine hemangiosarcoma with human orthologs to speculate that these candidates could be biomarkers for human angiosarcoma.
To develop a glyco-biomarker discovery platform, a lectin-based glycoproteomics approach was taken. Lectins are natural proteins with binding affinity to specific glycan structures. Lectin Magnetic Bead Array (LeMBA) was optimized on a liquid handler for high-throughput, utilizing 20 lectins to enrich for the serum sub-glycoproteome. Direct coupling to tandem mass spectrometry (MS/MS) allowed identification of the proteins bound to each lectin fraction. LeMBA-MS/MS demonstrated linearity, nanomolar sensitivity and specificity as shown by a proof-of-concept experiment using neuraminidase.
To facilitate high-throughput analysis of LeMBA-MS/MS, a label-free quantitation method was included in the LeMBA workflow, and a database incorporated with a statistical analysis pipeline called GlycoSelect was developed. Ten canine control and 10 canine visceral HSA serum samples were processed through LeMBA-GlycoSelect to identify candidate biomarkers. GlycoSelect analysis resulted in the identification of 146 lectin-protein candidates showing differential lectin binding, demonstrating the utility of LeMBA to identify glyco-biomarkers.
Candidates identified were validated in an independent, larger cohort of samples using LeMBA coupled multiple reaction monitoring (MRM). Three groups were employed: normal, HSA and a group called HSA-like consisting of sera from dogs with benign visceral lesions. Seven lectins were used for validation and 58 proteins were measured for each lectin by MRM. The lectin-bound protein abundance of approximately 50% of the 406 lectin-protein combinations screened was significantly different between the normal and HSA groups. The best candidate, Aleuria aurantia lectin bound to complement C4-A (AAL-C4A), had an area under the receiver operating characteristic curve (AUROC) of 0.94 with a sensitivity and specificity of 93.6% and 84.0%, respectively. Between HSA-like and HSA samples, the AUROC of the best candidate, Pisum sativum agglutinin bound to alpha-1B glycoprotein (PSA-A1BG), was 0.83, and sensitivity and specificity were 48.1% and 97.1%, respectively. The runner up, Phaseolus vulgaris Leukoagglutinin bound to fibronectin1 (LPHA-FN1), had an AUROC of 0.79 with a sensitivity of 33.3% and specificity of 97.2%.
The in-house canine database used in this thesis was created by merging the currently available canine protein databases, UniProt, NCBI Protein and Ensembl. To experimentally identify proteins of the canine spleen, the most common site for canine visceral HSA, subcellular fractionation and shotgun proteomics was performed. This identified 1,985 proteins in the in-house database and provided localization information. Bioinformatics was used to functionally annotate proteins with domain architecture and localization information. Domain comparison and/or BLAST analysis was used to identify orthologs between human and dog. This information was used to compare biomarker candidates identified between HSA-like and HSA in dogs: human and canine A1BG and FN1. For A1BG, sequence and domain comparison revealed moderate similarity between these two species, while FN1 was highly similar. Comparison of interacting partners of A1BG and FN1 displayed striking similarity between the two species. This indicated that the function of these proteins in the human and dog may be very similar, suggesting these candidates may also be biomarkers for human angiosarcoma.
Collectively, this study is the first to perform large-scale characterization of the canine proteome, and to have identified candidate biomarkers that can differentiate between serum samples from dogs with HSA-like disease and HSA. Furthermore, the selected biomarkers had high sensitivity and specificity when comparing serum samples from normal dogs and dogs with visceral HSA. These biomarkers may also be useful in human angiosarcoma.