MicroRNA表达可在诊断前采集的血清样本中检测金毛寻回犬的犬多中心性淋巴瘤

Canine lymphoma (cL) is one of the most common cancers in dogs, accounting for an estimated 24% of all canine malignancies. Multicentric cL (m-cL) represents 85% of all cL cases and shares many characteristics with human non-Hodgkin's lymphoma (hNHL). Due to these similarities as well as shorter overall lifespans, shared environments, and more homogenous genetics found in purebred dogs, m-cL is a valuable spontaneous animal model for hNHL. Prognosis for m-cL is quite variable, primarily depending on immunophenotype (T or B cell origin) and stage, and has median survival times ranging from 6-12 months after CHOP chemotherapy. Unfortunately, dogs aren't often diagnosed until they have late-stage disease, and relapse following chemotherapy is very common. A biomarker for early diagnosis could help identify patients when their disease is at an earlier stage, leading to better treatment responses and longer survival times. Potential biomarkers for early detection include microRNAs (miRNAs), which are small, non-coding pieces of RNA that control gene expression and are readily isolated from blood. These molecules are dysregulated in several types of cancer, and their sequences are nearly or entirely identical between dogs and humans. This makes them good non-invasive biomarkers with cross-species applications. Given the growing body of literature showing differential expression of miRNAs at the time of diagnosis, we sought to investigate whether any of these miRNAs could detect m-cL patients in samples taken prior to the onset of clinical symptoms and diagnosis. MiRNA expression was measured by real-time quantitative PCR in serum samples from 46 m-cL dogs and 40 control dogs enrolled in the Golden Retriever Lifetime Study. Each dog had a minimum of 3 samples, taken at roughly 1-year intervals for their whole lives. For the m-cL dogs, the samples were taken anywhere from 3 weeks to 5 years prior to diagnosis, with some having a 4 th sample taken at the time of diagnosis. We trained 5-fold cross-validated logistic regression models to classify samples as either m-cL or controls on 2 different training data splits. The first used only the samples from each m-cL dog taken between 3 weeks and 1-year pre-diagnosis, as we anticipated this range to be the most predictive of early disease. The second model included a random selection of 2 samples from each m-cL dog, resulting in a mixed-time population. We did this to examine the differences in miRNA selection and performance when the training populations were more or less homogenous, with respect to sampling time. As m-cL has a variety of subtypes, we also repeated our analyses using only the B-cell (n=15) or T-cell (n=26) immunophenotyped patients. We achieved accuracies between 65-88% across the training sets with similar miRNAs included in all models. With further testing, miRNAs offer potential for early diagnosis in both veterinary and human patients.