Circulating microRNA as biomarkers of canine mammary carcinoma in dogs.

Abstract Background Differentiating benign from canine malignant mammary tumors requires invasive surgical biopsy. Circulating microRNAs (miRNA) may represent promising minimally invasive cancer biomarkers in people and animals. Objectives To evaluate the serum mRNA profile between dogs with and without mammary carcinoma, and to determine if any of these markers have prognostic significance. Animals Ten healthy client‐owned female dogs (5 intact, 5 spayed) and 10 dogs with histologically confirmed mammary carcinoma were included; 9 were client‐owned, whereas 1 was a research colony dog. Methods Retrospective study. Serum miRNA was evaluated by RNA deep‐sequencing (RNAseq) and digital droplet PCR (dPCR).Expression of candidate biomarkers miR‐18a, miR‐19b, miR‐29b, miR‐34c, miR‐122, miR‐125a, and miR‐181a was compared with clinical characteristics, including grade, metastasis, and survival. Results 452 unique serum miRNAs were detected by RNAseq. Sixty‐five individual miRNAs were differentially expressed (>±1.5‐fold) and statistically significant between groups. Serum miR‐19b (P = .003) and miR‐125a (P < .001) were significantly higher in the mammary carcinoma group by dPCR. Both had high accuracy based on receiver operator characteristic area under the curve (0.930 for miR‐125a; 0.880 for miR‐19b). Circulating miR‐18a by RNAseq was significantly higher in mammary carcinoma dogs with histologic evidence of lymphatic invasion (P = 0.03). There was no significant association with any miRNA and survival or inflammatory status. Conclusions and Clinical Importance Circulating miRNAs are differentially expressed in dogs with mammary carcinoma. Serum miR‐19b and miR‐18a represent candidate biomarkers for diagnosis and prognosis, respectively.


| INTRODUCTION
Mammary tumors in dogs (canine mammary tumors, CMT) are 1 of the most common neoplasms in sexually intact female dogs, with a variable prognosis reflective of the fact that approximately half are benign and half are malignant. 1 Key histologic prognostic factors include subtype, grade, stage, and evidence of lymphatic invasion. Dogs with Grade III tumors have significantly shortened survival compared with dogs that have Grade I or II CMT. 2 Dogs with tumor lymphatic invasion identified in biopsies have a 3-fold higher rate of tumor recurrence, distant metastasis, and death. 2 Finally, dogs with inflammatory mammary carcinoma have widespread metastasis and a grave prognosis with average survival of 25 days from diagnosis. 3,4 Obtaining this prognostic information currently requires an invasive tissue biopsy. A minimally invasive biomarker for CMT could improve detection and clinical decision-making. Biomarkers are any quantitative measure of a disease or physiological state, and commonly include a variety of traditional or novel biochemical analytes. 5 Circulating microRNAs (miRNA) are small noncoding RNA molecules present in blood that show promise as powerful noninvasive biomarkers in human oncology. Unlike most RNA, serum miRNA concentrations are stable over time, temperature, and multiple freezethaw cycles, making them practical to assay. 6,7 One prospective study of women with breast cancer identified a miRNA signature (miR-21, miR-23b, miR-190, miR-200b, and miR-200c) that predicted tumor recurrence and shorter survival. 8 A panel of serum miRNAs, including miR-19a, miR-15, and miR-181a, correlated with patient tumor burden and decreased after surgical resection. 9 miR-331 and miR-195 accurately discriminate patients with metastatic breast cancer from those with only local disease. 10 miR-19a and miR-205 are higher in patients with Luminal A breast cancer that was chemoresistant to epirubicin and paclitaxel. 11 There are fewer published studies for miRNA in CMT, particularly as biomarkers in serum or plasma. miR-126 and miR-214 both are significantly increased in dogs with mammary carcinoma (along with a number of other malignancies) relative to healthy controls. 12 Malignant CMT tissues show differential miRNA expression by grade and metastasis, but the proposed miRNAs of interest did not significantly differ in plasma. 13 A recent in vitro study demonstrated that CMT cells secrete exosomes enriched in miRNAs which could be released into blood, and that the exosomal miRNA pattern is predicted to regulate the estrogen receptor (ESR1), key tumor suppressor PTEN, and other genes relevant to human and canine mammary cancer. 14 Other in vitro studies have suggested miR-143 and miR-138a are dysregulated in some CMT cells, and that miR-141 plays a role in CMT development by inhibiting tumor suppressor INK4A. [15,16] There is currently a lack of consensus on the most appropriate normalization strategy for circulating miRNAs. 17 Normal "reference" genes that are abundant in cells and tissues, such as snoRNAs, are generally absent to minimally detectable in serum. As an alternative approach, some authors recommend absolute quantification through a standard-curve calibrated to an exogenous spike-in miRNAs such as cel-miR-39. 18 Others normalize quantitative reverse-transcription PCR (qPCR) to plasma input volume. 19 RNA deep-sequencing (RNAseq) allows relative and absolute quantification based on normalization across millions of all mapped reads. 20 Digital droplet PCR (dPCR) provides absolute quantification without a normalization gene by measuring tens of thousands of PCR reactions in parallel and assaying against a standard curve for 6-carboxyfluorescein (FAM) fluorescence. 17 qPCR and dPCR for miRNAs in lung cancer have high correlation between the assays, with dPCR having lower coefficient of variation and greater reliability. 21 We determined that the optimal combination of sensitivity and robust results for profiling circulating miRNAs in this cohort was initial target identification by RNAseq validated by dPCR absolute quantification.
Our hypotheses were that (1) the serum miRs would be differentially expressed between healthy dogs and those with CMT with good diagnostic performance, (2) that multiple assay methods (RNAseq and dPCR) would provide similar results, and (3) that these miRNAs would be significantly correlated with tumor grade, lymphatic invasion, inflammatory carcinoma status, and survival time.

| Sample groups and tumor pathology
Ten healthy female dogs (5 spayed and 5 intact) were prospectively recruited for the control group and 10 dogs with mammary carcinoma were included in the CMT group. Exclusion criteria for healthy females was any evidence of disease by a veterinarian's physical examination, or abnormalities on CBC or serum biochemistry tests. The 5 healthy intact females varied by stage of estrus at the time of blood collection, and included 3 in estrus, 1 in diestrus, and 1 in anestrus.
Nine of the 10 dogs with mammary carcinoma were enrolled in a previous study on dendritic cell fusion vaccines for CMT; the tumor tissue and serum from all of these dogs were collected before any treatments or interventions. 22 One of the 10 CMT dogs (MC10) was part of a breeding colony for research dogs and was scheduled for euthanasia because of age and quality of life concerns; a large mammary tumor was discovered before euthanasia, and fresh whole blood, serum, and tumor tissue were collected from this dog immediately postmortem.
Two board-certified anatomic pathologists blinded to dog identity confirmed the malignant status of the CMT biopsy specimens. Tumors were subtyped histologically, graded, and assessed for the presence or absence of lymphatic/vascular invasion by blinded pathologist blinded to dog identity as previously described. 2  After processing the sequencing reads from RNA-seq experiments from each sample was performed as per the HudsonAlpha Discovery unique in-house pipeline. Briefly, quality control checks on raw sequence data from each sample was performed using FastQC (Babraham Bioinformatics, London, United Kingdom). Raw reads were imported on a commercial data analysis platform (Avadis NGS, Strand Scientifics, California).

| RNA extraction and miRNA RNAseq
Adapter trimming was done to remove ligated adapter from 3 0 ends of the sequenced reads with only 1 mismatch allowed, poorly aligned 3 0 ends were also trimmed. Sequences shorter than 15 nucleotides length were excluded from further analysis. Trimmed reads with low qualities (base quality score less than 30, alignment score less than 95, mapping quality less than 40) were also removed. Filtered reads were then used to extract and count the small RNAs which were annotated using miRNAs from the miRBase release 20 database (http://www.mirbase.org/). Samples were subjected to quantification and active region quantification (Avadis NGS, Strand Scientifics). The quantification operation carries out measurement at both the gene level and at the active region level. Active region quantification considers only reads whose 5 0 end matches the 5 0 end of the mature miRNA annotation. Samples were then grouped by identifiers and the differential expression of each miRNA was calculated based on the fold change observed between different groups.
Serum RNA was converted to cDNA using the TaqMan Advanced   Sixty-five individual miRs were differentially expressed (>±1. 5-fold) and statistically significant between healthy females and those with CMT. The volcano plot in Figure 2 graphically illustrates this differential miRNA expression between groups. Table SS2 shows all significantly differentially expressed miRNAs in CMT samples compared to controls. Some of these upregulated miRs have been previously identified as upregulated in CMT exosomal RNA shed by cultured CMT cells, including miR-18a, miR-19b, miR-29b/c, miR-34c, miR-181c, miR-215, and miR-345. 14
miR-18a, miR-19b, and miR-181a were the most abundantly expressed F I G U R E 1 Principal component analysis (PCA) plot for circulating microRNAs. Mammary carcinoma dogs are plotted in red, healthy control dogs are plotted in blue. Square data points represent sexually intact healthy females, whereas circles represent spayed healthy females F I G U R E 2 Volcano plot for serum microRNA expression by RNA deep-sequencing. Red dots in the upper right are significantly upregulated in the canine mammary tumor group by >1.5-fold, whereas green dots in the upper left are significantly downregulated >1.5-fold. Black dots were miRs that were expressed with either a <±1.5-fold difference or not statistically significantly different in expression between groups miRs in the set tested by dPCR, with others having lower absolute expression. miR-34c and miR-125a had the largest magnitude relative fold-change between the mammary carcinoma group and healthy control group (Figure 3). Both miR-19b and miR-125a were significantly higher in the mammary carcinoma group than among healthy control dogs by dPCR ( Figure 4A,C). miR-34c was substantially higher among dogs with mammary carcinoma than healthy subjects, although this difference narrowly missed statistical significance ( Figure 4B, Table 2).
One dog in the healthy control group (subject HS3) was an outlier with extremely high miR-19b expression (32 364 copies/μL). Clinical followup on this dog revealed that within 1 year of this sample collection it developed widespread pulmonary metastasis from an unknown primary cancer and died shortly thereafter.
RNA deep-sequencing and dPCR assays were compared by assessing miRNA fold-change and statistical significance between the mammary carcinoma and healthy control groups, and this data is summarized in Table 2. Two of 7 miRNAs were significantly different by both methodologies (miR-19b and miR-125a). Results between the assays were largely similar in direction of fold-change, with the notable exception of miR-125a and miR-122, which were both increased by dPCR despite being downregulated according to RNAseq. Six of 7 miRNAs had less-extreme fold-change by dPCR than RNAseq (with the exception of miR-125a). miR-181a was abundantly expressed in both carcinoma and control cohorts, but did not differ statistically between groups by either RNAseq or dPCR.

| Diagnostic performance
To evaluate the ability of these miRNAs to discriminate clinical cases from control subjects, ROC plots were generated. The highest ROC area under the curve (AUC) was miR-125a at 0.930 ( Figure 5A), indicating excellent ability to discriminate between dogs with mammary carcinoma and healthy controls in this population. miR-19b also had a high ROC-AUC at 0.880 ( Figure 5B). When excluding the outlier healthy control HS3 because of the possibility of occult neoplasia, the AUC-ROC for miR-19b increased to 0.978, which would indicate near-perfect ability to discriminate mammary tumor-bearing dogs from dogs without neoplasia.
All other miRNAs had fair to poor ROC-AUC ( Table 3).
Because of suitable biomarker characteristics for miR-19b (high absolute expression, strong upregulation by the CMT group by RNAseq and dPCR, robust ROC-AUC), additional diagnostic test parameters were calculated for this miRNA. The diagnostic sensitivity and specificity of miR-19b varied by the selected cutoff, and whether dog HS3 was included or not. At 11 600 copies/μL and including HS3, miR-19b had a sensitivity of 100%, a specificity of 80%, and a positive likelihood ratio of 5.0 (95% CI: 1.96-17.64). At a cutoff of 13 000 copies/μL, the sensitivity decreased to 80% whereas specificity increased to 90%; the positive likelihood ratio increased to 8

| Circulating miRNA association with survival
Survival times from time of surgical resection to spontaneous death or euthanasia were available for 8 of 10 dogs in the CMT group. There were no statistically significant correlations between any of the 7 circulating miRNAs by RNAseq or dPCR and survival time in days (Table 4).

| DISCUSSION
Our study demonstrates that serum from dogs with mammary carcinoma is enriched with hundreds of circulating miRNAs. While the overall expression pattern between dogs with malignant CMT and healthy controls had substantial overlap based on PCA, a number of individual miRNAs were significantly upregulated or downregulated in the CMT group. Previous in vitro research on CMT exosomes reveals a number of these, including miR-18a, miR-19b, miR-29b/c, miR-34c, miR-181c, miR-215, and miR-345 that are predicted to target a number of important genes and pathways relevant to mammary tumorigenesis, such as ESR1 and the tumor suppressor PTEN. 14 Of these miRNAs, miR-29b is upregulated in malignant CMT versus normal mammary tissue. 23 Circulating miR-126 and miR-214 are both increased in dogs with a variety of tumors, including mammary malignancy, however neither was between those significantly or differentially expressed between CMT and healthy dogs in this dataset. 12 miR-19b was strongly and significantly upregulated in the CMT group by both RNAseq and dPCR. Furthermore, ROC-AUC and sensitivity/specificity analysis indicated this particular miRNA had good ability to differentiate between the 2 groups. Although this is a small cohort, and neither animals with nonneoplastic mammary disease (ie, mastitis) nor subjects with nonmammary neoplasms were included, this suggests miR-19b is worth further investigation as a potential biomarker for mammary carcinoma in dogs. This agrees with prior studies that show circulating miR-19a (closely related to miR-19b) has prognostic significance in women with breast cancer. 9,11 Interestingly, 1 dog in the healthy spayed female control cohort (HS3) had extremely high miR-19b expression and went on to develop widespread metastasis from an unidentified primary cancer within 12 months of sample collection. While this dog did not show any outward evidence of occult malignancy on physical examination or Our study has a number of limitations. First, the small sample size might have been underpowered to detect modest but real group differences, especially for dPCR. Notably, absolute expression for miR-34c was prominently upregulated in the mammary carcinoma group, but the P value was slightly above the alpha .05 boundary of statistical significance for dPCR. However, despite the modest number of biological replicates, RNAseq identified millions of small RNA reads, many of which were differentially expressed. The risk of false positives detected by RNAseq simultaneously analyzing hundreds of miRNAs was mitigated statistically through the Benjamini-Hochberg correction procedure and technically by validation with a different quantification method (dPCR). 25 In addition, both the mammary carcinoma and healthy control group subjects were robust, with the former having a variety of tumors of different histologic type and grade, and the latter including both OHE dogs and bitches in various stages of estrus. The diversity in histopathologic characteristics is especially important, and 2 important prognostic factors included high-grade tumors (Grade III) and tumors with lymphatic invasion. 2 Second, this population did not include subjects with nonmalignant mammary pathology such as mastitis or benign mammary tumors. This could be relevant as research evaluating miRNA in cow and porcine milk has identified particular miRNA signature that increase with mastitis, including miR-21, miR-146a, miR-155, miR-222, and miR-383. 26 Fortunately, these miRNAs are not among the most relevant potential biomarkers identified in this dataset.