Analysis of Copy-Number Variations and Feline Mammary Carcinoma Survival

Feline mammary carcinomas (FMCs) are highly malignant. As the disease-free survival (DFS) and overall survival (OS) are short, prognostication is crucial. Copy-number variations (CNVs) analysis by next-generation sequencing serves to identify critical cancer-related genomic regions. Thirty-three female cats with FMCs were followed during two years after surgery. Tumours represented tubulopapillary and solid carcinomas encompassing six molecular subtypes. Regardless of the histopathological diagnosis, molecular subtypes showed important differences in survival. Luminal A tumours exhibited the highest DFS (p = 0.002) and cancer-specific OS (p = 0.001), and the lowest amount of CNVs (p = 0.0001). In contrast, basal-like triple-negative FMCs had the worst outcome (DFS, p < 0.0001; and OS, p < 0.00001) and were the most aberrant (p = 0.05). In the multivariate analysis, copy-number losses (CNLs) in chromosome B1 (1–23 Mb) harbouring several tumour-repressors (e.g. CSMD1, MTUS1, MSR1, DBC2, and TUSC3) negatively influenced DFS. Whereas, copy-number gains (CNGs) in B4 (1–29 Mb) and F2 (64–82.3 Mb) comprising epithelial to mesenchymal transition genes and metastasis-promoting transcription factors (e.g. GATA3, VIM, ZEB1, and MYC) negatively influenced DFS and cancer-specific OS. These data evidence an association between specific CNVs in chromosomes B1, B4 and F2, and poor prognosis in FMCs.

In group TC, the Ki-67 proliferation index ranged from 2% to 54.2% (mean [SD]: 19.1 [14.4] %); 14 tumours displayed high Ki-67 indexes (≥14%, cut-off value according to Soares et al. 34 ), while the remaining 11 showed low Ki-67 scores (<14%). On the other hand, all cases included in the group SC were characterised by high Ki-67 indexes (≥14%), ranging from 14.4% to 44 37,38 , and the novel histological malignancy grading system for evaluation of FMCs [Mills-2015] 37,38 ), Supplementary Table 3. There was a minimal level of agreement when the histological grade assigned to each tumour using the three different systems was compared using Cohen's Kappa test: EE and MMEE (K = 0.25), EE and Mills-2015 (K = 0.27), MMEE and Mills-2015 (K = 0.24). Moreover, in only ten out of the 33 cases evaluated (30.3%) tumours obtained the same grade (five, grade I; two, grade II; and three, grade III) when using any of the three grading systems evaluated.
Molecular subtyping. The LA molecular subtype (Fig. 1a-c) was the most common, with a frequency of 30.3% (10/33), followed by the normal-like TN subtype 24.2% (8/33). Two cases (fHER2 IHC-3+) were classified as LB fHER2+. After confirmation of CNGs encompassing HER2, one case (fHER2 IHC-2+) was confirmed as fHER2+ subtype (Fig. 1d-f). LB fHER2−, and basal-like TN (Fig. 1g-i) tumours were detected in ten cases, five cases (15.1%) each subtype. Moreover, in two cases included in the TC group subtyping was not performed due to loss of the neoplastic lesion from the paraffin block after consecutive slicing, in those cases, it was not possible to evaluate all required markers to assess the St. Gallen classification.
Tumours in the group TC (n = 25) included multiple molecular subtypes. In this group, ten cats (40%) had LA subtype tumours. Whereas four cats (16%) had normal-like TN tumours, five cats (20%) had LB www.nature.com/scientificreports www.nature.com/scientificreports/ fHER2− tumours, two cats had LB fHER2+ tumours, two cats had basal-like TN tumours, and two cats were not classified, Supplementary Table 2. Interestingly, LA subtype tumours (all included in group TC) had a significantly lower Ki-67 index (p < 0.001) than any other molecular subtype. The group SC (n = 8) was less heterogeneous in terms of molecular diversity. In this group, seven cats (87.5%) had triple-negative FMCs including four normal-like TN and three basal-like TN tumours. Moreover, one cat included in this group was classified as fHER2+, Supplementary Table 2.
Tissue samples DNA quantification. Thirty-three tumours were included: 23 formalin-fixed paraffin-embedded (FFPE) samples (group TC = 16, and group SC = 7), and ten fresh-frozen (FT) samples (group TC = 9; and group SC = 1). The DNA yields from FT samples were more consistent when compared to those of FFPE samples, ranging from 1466. 2  In the group TC, four FFPE samples stored for over 15 years (LA = 1, LB fHER2− = 1, normal-like TN = 1, and basal-like TN = 1) failed library preparation due to low DNA quality. In three (two FFPE, and one FF) samples (all LA subtype) of the same group, no CNVs were detected, these seven samples were later excluded from CNVs analysis. Twenty-six samples were suitable for CNVs analysis (FFPE = 17, and FT = 9) as follows: LA = 6, LB fHER2+ = 2, LB fHER2− = 4, fHER2+ = 1, normal-like TN = 7, basal-like TN = 4, and not-classified = 2; distribution of molecular subtypes across histological groups (TC and SC) is detailed in Supplementary Table 2. follow-up and survival analysis censoring. At the end of the study period (24 months), 12 cats (48%) in the group TC (n = 25) had developed local recurrence and eight of them (32%) subsequent distant metastasis (eight, pulmonary/pleural; and one, intestinal). One cat (normal-like TN) in this group developed pulmonary/ pleural metastasis without previous local recurrence, and 12 cats (48%) did not develop any kind of recurrence (Supplementary Table 4); cats without recurrence (n = 12) at the end of the study period were censored from DFS analysis. Eight cats (32%) in the group TC (LA = 7, and not-classified = 1) were still alive at the time of censorship; however, two of them developed local recurrence (21, and 24 months after surgery). In 11 cases (44%) included in group TC (44%) the cause of death was related to the progression of the neoplastic disease, while six cats (24%) died due to non-tumour related causes as follows: two, anaesthetic complications during mastectomy; two, renal failure; one, euthanised during mastectomy due to concomitant intestinal primary tumour and liver metastasis; and one, severe cardiac disease (Supplementary Table 4). All animals alive at the end of the study period (n = 8) and those who died due to non-tumour related causes (n = 6) were censored from cancer-specific OS analysis.
In the group SC (n = 8), five cats (63%) developed local recurrence and pulmonary/pleural metastasis, and three cats (all basal-like TN) developed pulmonary/pleural metastasis without previous local recurrence, Supplementary Table 4. Additionally, all animals in group SC died due to the progression of the neoplastic disease. Accordingly, none of the animals in the group SC was censored from DFS and cancer-specific OS analysis. After censorship, 21 cats (TC = 13, and SC = 8) were included in the DFS analysis and 19 cats (TC = 11, and SC = 8) were included in the cancer-specific OS analysis. cnVs detection and frequency analysis. 140 CNVs were detected using CNV-seq 39 [15.03] %). Accordingly, eleven samples in the group TC (n = 18) were scored as low-CNVs (<median overlapped aberrant genomic windows percentage), while the remaining seven as high-CNVs (≥median overlapped aberrant genomic windows). In contrast, group SC included six samples with high-CNVs and two with low-CNVs, Supplementary Table 5.
In both groups (TC and SC) CNVs affected all chromosomes. However, the frequency of CNVs in the group TC was very low in comparison to the group SC. In group TC (n = 18), 81 among the 97 validated CNVs were observed in less than 20% (4/18) of the cases, 13 were not detected in any of the patients allocated in the group, and only one (CNG in FCA E3 37-42.5 Mb) was observed in at least 30% (6/18) of the cats included in the group, Supplementary Table 6. In contrast, group SC displayed a higher frequency of CNVs, 71 among the 97 validated CNVs were detected in more than 20% (2/8) of the patients allocated in group SC, 24 CNVs were observed in at least 50% (4/8), and one (CNG in FCA B4 1-29 Mb) was observed in 75% (6/8) of the cases, Supplementary Table 6. On the other hand, poorly differentiated tumours (grade III, according to the EE system), and tumours with high Ki-67 index showed a significantly higher amount of aberrant genomic windows affected by CNVs (p = 0.01, and p = 0.0002; respectively).
Survival differences among histological groups and molecular subtypes. The DFS and cancer-specific OS of cats in the group SC were significantly lower (p = 0.003 and p = 0.001; respectively) than those of cats allocated in the group TC (Table 1). The lower frequency of CNVs in the group TC ( Fig. 2a) compared with the group SC ( Fig. 2b) corresponded with the higher survival rates observed in the group TC (Fig. 2c,d). However, a subset of patients (n = 7) in the group TC displayed shorter survival intervals (DFS; p < 0.00001, and OS; p = 0.0004) in comparison to the rest of the cats allocated in the group (Fig. 2c,d). Moreover, those patients mainly represented molecular subtypes with worst prognosis (e.g. LB fHER2−, and triple-negative) and were characterised by high-CNVs scores, high Ki-67 indexes, and a significantly (p = 0.001) higher percentage of overlapped aberrant genomic windows (mean [SD]: 27.02 [14.73] %). On the other hand, group TC included all LA subtype tumours characterised by lower histological malignancy grades (EE system), low Ki-67 indexes and higher survival intervals. In contrast, all patients included in the group SC displayed reduced survival intervals. Moreover, this group predominantly included molecular subtypes with worst prognosis (i.e. fHER2+, basal-like TN, and normal-like TN), high Ki-67 indexes, and high histological malignancy grades when applying the EE system.
To explore survival differences observed within the patients allocated in group TC, we independently evaluated the survival intervals of patients distributed according to molecular subtype, amount of CNVs, and additional clinical and histopathological variables evaluated. Regardless of the histopathological diagnosis, patients showed important differences in survival times according to molecular subtype and CNVs score (Table 1).
Patients showed significant differences in survival intervals when applying the EE grading (DFS; p = 0.05 and cancer-specific OS; p = 0.02) and the MMEE grading (cancer-specific OS; p = 0.02). Survival intervals were significantly different between patients with well-differentiated (grade I) and poorly-differentiated (grade III) tumours using the EE (DFS, p = 0.04; and cancer-specific OS, p = 0.005) and the MMEE (cancer-specific OS, p = 0.001) grading systems. However, survival intervals between grades I and II, and grades II and III were not significantly different when applying any of the histological grading systems employed (EE, MMEE, and Mills-2015). All additional independent variables considered including breed, age, and reproductive status did not influence DFS and cancer-specific OS (p > 0.05).
F2 64-82.3 Mb (human homologue region located in HSA 8q) was significant in multivariate analysis and harboured several breast cancer-related genes affected by CNGs in HBCs 19,28,41 including EMT-related genes (i.e. MYC, SCRIB, NDRG1, and PTK2) commonly affected by CNGs in multiple human cancer types 21 (Tables 3 and 4), basal-like TN 2.8 ± 0.6 (n = 5) 3.9 ± 0.7 (n = 5) among them, MYC amplification is validated as somatic-CNG by the Cancer Gene Census (CGC) 59 . Moreover, this genomic region harboured ASAP1 which encodes an oncoprotein correlated with enhancing cell motility, invasiveness, and poor survival in human cancers including HBCs 60-63 . Besides CNGs influencing both DFS and cancer-specific OS, FCA E3 1-34.5 Mb (human homologue region located in HSA 7p) was only correlated with poor DFS and was the most common CNG detected (Supplementary Table 6). This CNG harboured multiple members of the leukocyte transendothelial migration pathway, including cell junction components (e.g. CLDN3, CLDN4, and CLDN15), and serine-and threonine-specific protein kinases (e.g. PRKCB). Moreover, CNGs in FCA E3 1-34.5 Mb included genes (e.g. MYL10, ACTB, RAC1, and MYLPF) involved in cell movement, actin cytoskeleton remodelling, and cellular proliferation and motion such as SBDS, ITGAM, PDGFA, and RAC1. Among them, RAC1 actively promotes cell motility in breast cancer cells 64,65 and PDGFA has been reported as highly expressed in FMCs and derived cell lines 66 . Furthermore, E3 1.1-34.5 Mb harboured SERPINE1; a gene encoding a cell adhesion protein involved in invasion and metastasis promotion in different human malignancies including HBCs 67-69
CNLs included a considerable amount of genes involved in specific biological processes (i.e. cell adhesion, biological adhesion, and cell-cell adhesion) that would normally prevent cellular migration, Table 6. Among deleted regions correlated with poor outcome, CNLs affecting the proximal region of FCA A2 (23-37.3 Mb, and 37.3-48 Mb; human homologue regions located in HSA 3p) comprised genes implicated in cell and biological adhesion, including several contactins (e.g. CNTN3, CNTN4, and CNTN4) active during embryonic development 49 , Tables 5 and 6. On the other hand, CNLs in FCA B1 1-23 Mb (human homologue region located in HSA 8p) were significant in multivariate analysis and harboured multiple tumour suppressors (e.g. CSMD1, MSR1, MTUS1, and TUSC3) previously correlated with poor prognosis in HBCs 15,19,23,27,70 . Deleted genes predominantly detected in proximal FCA D1 (1-20 Mb, 21-39.4 Mb, and 41.2-97 Mb; human homologue regions located in HSA 11q) encoded extracellular matrix proteins and membrane-associated guanylate kinases (e.g. TECTA, and MAGI1), and cancer-associated CAMs (e.g. CADM1, OPCML, MCAM, HEPACAM, and ESAM) 49 , Table 6. Moreover, deleted regions affecting FCA D1 included genes validated as cancer variants by the CGC 59 including ATM validated as somatic-CNL, and tumour suppressor WT1 validated somatic-and germline-CNL. cnVs distribution across molecular subtypes. We observed a negative correlation between the percentage of aberrant genomic windows (Fig. 3a) and survival intervals (Fig. 3b,c). Among all subtypes studied, LA subtype tumours displayed the highest survival intervals (Fig. 3b,c), and the significantly (p = 0.001) lowest percentage of aberrant genomic windows affected by CNVs (n = 6, mean [SD]: 3.05 [3.6] %, Fig. 3a). This subtype exhibited a very low occurrence of CNVs (Fig. 3d) in comparison with the other molecular subtypes. Moreover, all patients within the group had a low-CNVs score, Supplementary    www.nature.com/scientificreports www.nature.com/scientificreports/ (Fig. 3d). CNVs affecting LA subtype tumours included two out of the ten CNVs associated with poor survival intervals (CNLs in B1 1-23 Mb and CNGs in D4 1-16.7 Mb), however, both of them were detected in only one case. Interestingly, none of the patients within the group harboured the most common CNV observed in this study (CNGs in FCA E3 1-34.5 Mb).
The two cases classified as LB fHER2+ (Fig. 3e) displayed a different pattern of aberration, one case was classified as low-CNVs (Supplementary Table 5) score and carried CNLs affecting FCAs A1 and A2. In contrast, the other case was classified as high-CNVs (Supplementary Table 5) and carried CNLs in FCAs A1, A2, B2, C1 and C2, and CNGs affecting FCAs B1, B4, D2, and E3. Interestingly, this case was the only one among all cases included in this study carrying CNGs affecting B1 (1-23 Mb). Only CNLs in FCA A1 (179-194 Mb and 198-203.7 Mb) affected both LB fHER2+ cases (Fig. 3e), however, none of them was correlated with reduced survival intervals.
Among LB fHER2− cases (n = 4), two cats displayed low-CNVs scores and two high-CNVs scores, Supplementary Table 5. CNVs affected all FCAs except A1 and E2, nevertheless, most of the structural rearrangements affected patients individually and none of them affected more than two patients simultaneously (Fig. 3e). Aberrations affecting two patients within this subtype included the ten CNVs correlated with reduced survival intervals listed in Tables 3 and 5.
The only case classified as fHER2+ exhibited nine CNGs affecting FCAs B4, E1, and E3, and two CNLs in FCAs A1 and D2 (Fig. 3g). Eight out of the nine CNGs were correlated with reduced DFS or cancer-specific OS and four of them (B4 1-29 Mb and E1 [19.3-37 Mb, 38-44 Mb, and 44-62 Mb]) affected both. On the other hand, none of the CNLs detected was correlated with poor outcome.
Normal-like TN-FMCs (n = 7) included three cases scored as low-and four as high-CNVs, Supplementary Table 5. Despite the poor survival observed and the high amount of CNVs detected (n = 77) in this subtype, almost half of them (n = 32) affected patients individually (Fig. 3h). Nineteen CNVs affected more than 40% (3/7) of the patients. Among them, ten negatively influenced DFS, and three (CNGs in FCAs A3 1-31 Mb, and B4 1-29 Mb; and CNLs in FCA D1 1.1-20 Mb) affected both DFS and cancer-specific OS. In this subtype, CNLs in FCA X (1-26 Mb and 108-126.3 Mb) were the most common (five and four patients; respectively). However, none of them was correlated with reduced survival intervals.
Finally, all basal-like TN-FMCs (n = 4) were classified as high-CNVs score, Supplementary Table 5. Moreover, this subtype exhibited the significantly (p = 0.05) highest amount of aberrant genomic windows (Fig. 3a) and the lowest survival intervals among all molecular subtypes studied (Fig. 3b,c). Basal-like TN-FMCs displayed 78 structural rearrangements affecting all chromosomes but D3 (Fig. 3i). CNVs affecting this subtype included all rearrangements correlated with reduced survival intervals listed in Tables 3 and 5,  and Supplementary Table 7 [54-88.9 Mb] were assessed using functional clustering and KEGG pathway analyses to detect cancer-related genes and identify deregulated biological processes characterising the poor prognosis observed in this subtype.

Discussion
CNVs serve to identify cancer-related genomic regions 16,22,71 . To the best of our knowledge, this study represents the first analysis of CNVs influence on FMCs survival. Cats represented tubulopapillary carcinomas (group TC), and solid carcinomas and comedocarcinomas (group SC). As reported 37,72 , the group SC displayed lower survival intervals (DFS and cancer-specific OS) than the group TC. However, a subset of patients in group TC exhibited worse prognosis than others allocated in the same group. Considering this, an independent classification (St. Gallen) was applied to identify molecular subsets with distinctive biological behaviour (i.e. LA, LB fHER2+, LB fHER2−, fHER2+, and normal-like TN, and basal-like TN) 9,10 . Moreover, to identify factors underlying survival differences, we performed univariate and multivariate analyses including genomic characteristics explored (i.e. percentage of aberrant genomic windows, CNVs score, and presence of specific CNVs) and known malignancy indicators (e.g. histological grading, Ki-67 index, etc.).
The group TC was molecularly heterogeneous including subtypes with best and worst prognoses, which explains the outcome differences observed within the group. Conversely, the group SC mainly included highly-aberrant TN-FMCs. These results highlight the value of the St. Gallen classification to predict prognosis, especially in tubulopapillary carcinomas, in which the histological diagnosis fails to differentiate patients according to possible outcome. In line with previous studies 9,10 , molecular subtype was correlated with prognosis. We  www.nature.com/scientificreports www.nature.com/scientificreports/ observed a negative correlation between the percentage of aberrant genomic windows and survival intervals. LA subtype tumours were the least aberrant and displayed the longest survival rates. In contrast, basal-like TN-FMCs shown the poorest survival and were the most aberrant, which provide evidence for a higher genomic instability.
To evaluate the influence of genomic instability on survival, tumours were classified into high-or low-CNVs (CNVs score). A high-CNVs score negatively influenced survival intervals in the univariate analysis. Since the recurrence of specific CNVs indicates that such regions are likely to harbour cancer-related genes 73  Thirdly, D4 1-16.7 Mb (human homologue in HSA 9p) gain is associated with a distinctive molecular subtype detected across major human cancer types, including triple-negative breast cancer (hTNBC) 57,78,79 . In this study, this aberration was observed in six patients including two normal-like TN and three basal-like TN tumours highlighting the genomic similarities between triple-negative molecular phenotypes across FMCs and HBCs. Tumours carrying this mutation are characterised by the co-gain of PD-L1, PD-L2, and JAK2, and might benefit from immune checkpoint targeted therapies 57,78,79 or adjuvant therapy with JAK2-specific inhibitors 80,81 . Interestingly, the Jak-STAT signalling pathway was the only KEGG pathway enriched with amplified genes in the basal-like TN subtype. Finally, CNGs in F2 64-82.3 Mb (human homologue in HSA 8q) resemble a CNG reported in HBCs 18,41 , canine mammary tumours 82 , and HBC-and FMC-derived cell lines 19,83,84 . This region harbours more than 30 HBC-related genes 19,28,41 , among those MYC, SCRIB, NDRG1, and PTK2 are EMT-related genes frequently amplified in human cancer 21 . Furthermore, proto-oncogene MYC amplification is a somatic-CNG validated by the CGC 59 .
Among CNGs detected, survival intervals remained negatively influenced by CNGs in B4 1-29 Mb and F2 64-82.3 Mb in the multivariate analysis. The influence of these aberrations on EMT-elicitation in an FMC-derived cell line was described for our group 84 . In this study, these CNGs were associated with poor outcomes and were commonly observed across the population except for the LA subtype. These results now provide evidence about their influence on FMC survival and also highlight the importance of detecting EMT-associated aberrations to predict early recurrence and reduced survival.
Besides CNGs influencing survival intervals, a CNG affecting E3 1.1-34.5 Mb (human homologue in HSA 7p) was the most common and affected all molecular subgroups except the LA subtype. This aberration is also reported in human 20 and canine mammary tumours 82 . Genes in this region enriched the leukocyte transendothelial migration pathway. The similarities in the molecular mechanisms behind the first steps of leukocytes and neoplastic cells extravasation are well known 85,86 . Moreover, this region encompasses up to 50 genes implicated in actin cytoskeleton organization and cellular motion including SBDS, ITGAM, PDGFA, and RAC1 49 . Our findings suggest that FMC cells might use part of the molecular machinery orchestrating leukocyte extravasation www.nature.com/scientificreports www.nature.com/scientificreports/ to facilitate dissemination and invasion through the endothelium. Moreover, cats with tumours carrying this aberration might be a suitable model for the identification of molecular targets which suppression might prevent or reduce cellular migration and tumour dissemination. Among possible molecular targets in the region, CLDN3 and CLDN4 overexpression characterises invasive HBCs 87 and CLDN4 might be necessary for vasculogenic mimicry 88 .
On the other hand, common CNLs negatively influencing survival intervals harboured important tumour suppressors (e.g. CSMD1, MTUS1, MSR1, DBC2, and TUSC3) and genes enriching biological processes that would normally prevent cellular movement. Moreover, aditional genes in deleted regions enriched common deregulated metabolic pathways in cancer such as Pyruvate metabolism, PPAR signalling pathway, and N-Glycan biosynthesis [89][90][91][92] 95,96 . The deletion or down-regulation of genes BARX2, HEPACAM, and OPCML included in those regions is a potential biomarker and therapeutic target in different types of human malignancies [97][98][99][100][101] including HBCs [102][103][104] . OPCML and BARX2 downregulation is associated with EMT-elicitation and poor prognosis in different types of human cancer 97,99,105 . This findings highlight the importance of this deletion as a useful biomarker for poor prognosis and its potential for the identification of therapeutic targets.
As observed in other molecular subtypes, CNLs affecting basal-like TN-FMCs harboured tumour suppressors or genes enriching biological processes normally preventing tumour dissemination. Moreover, basal-like TN-FMCs displayed additional CNLs associated with poor outcome (i.e. CNLs in A1 105-124 Mb, and D2 54-88.9 Mb) also detected in normal-like TN-FMCs. Among genes deleted, tumour suppressor CTNNA1 (A1 105-124 Mb, human homologue in HSA 5q) is implicated in cell-cell adhesion maintainance and regulation of multiple signalling pathways in human cancer 106 . Furthermore, CNLs affecting the basal-like TN-FMCs enriched the basal cell carcinoma pathway, and the Wnt signalling pathway commonly deregulated in hTNBCs 28,77,106 . Moreover, multiple enzyme-coding genes (e.g. LDHC, LDHA, ALDH7A1, and LDHAL6A) participating in the reprogramming of metabolic pathways were deleted in this subtype.
Consistent with other studies 1,34,107 , a high-Ki-67 index remained associated with reduced cancer-specific OS in the multivariate analysis. As previously reported 1,36,38,72,108 the EE, and the MMEE grading systems were correlated with poor outcome in the univariate analysis. Nonetheless, among the grading systems employed, the Mills-2015 system showed the highest level of agreement (66.7%) between well-differentiated tumours and low Ki-67 indexes. All systems evaluated (EE, MMEE, and Mills-2015) detected poorly-differentiated tumours efficiently, nonetheless, a minimal level of agreement among the three systems was observed (K<0.30). Furthermore, none of them revealed significant differences in survival times when comparing moderately-differentiated tumours with poorly-differentiated or with well-differentiated tumours independently.
Considering possible differences in HER2 expression regulation between HBCs and FMCs 109-112 and the most recent recommendations for IHC-2+ validation in HBCs 31 , we correlated fHER2 IHC-expression with CNGs affecting HER2. We identified a CNG (E1 38-44 Mb) harbouring HER2 in seven patients (log2-ratio >0.2), four of them with a log2-ratio >0.5 including three IHC-negative (0 or 1+) and one IHC-2+ cases. In this study, two IHC-3+ tumours did not carry HER2-associated CNGs. In line with reports in HBCs 113,114 and as recently reported in FMCs 110 , some IHC-3+ tumours do not display HER2 amplifications questioning the role of this aberration as a driver mutation in FMCs. Considering the higher proportion of fHER2 IHC-negative cases carrying this mutation, it is possible that it might be associated with the amplification of additional cancer-related genes in the region (e.g.CDK12, TOP2A, EIF1, and STAT3) rather than inducing fHER2 overexpression as typically observed in HBCs.
Findings of this study in combination with reports on FMCs 109,111,115,116 question the utility of HER2 amplification to validate fHER2 IHC-overexpression. In line with previous studies 109,111,115,116 , FMCs with fHER2 IHC-overexpression and normal HER2 copy-number might be a suitable model for the homologue HBC subgroup (HER2; IHC-3+/FISH-) 117 . Further studies using larger cohorts should apply the new interpretation system 31 to evaluate the suitability of HER2 amplifications detection to validate fHER2-IHC overexpression in FMCs. Besides HER2 amplification, a higher fHER2 serum concentration (≥10 ng/ml) was detected in cats carrying IHC-2+ and −3+ tumours 112 . However, serum fHER2 levels between IHC-2+ and IHC-3+ were not statistically compared.
As HER2-mRNA levels positively correlate with a stronger IHC-expression on FMCs and FMC-derived cell lines without HER2 amplification 115,116 , it is possible that as documented in HBCs 118,119 and suggested in

Functional clustering of genes in CNGs
Besides somatic-CNVs affecting neoplastic cells, normal cells also have variations in the number of copies, representing a source of genomic diversification during evolution 120,121 . Those mutations usually affect smaller genomic regions than somatic-CNVs and some of them-usually referred to as germline-CNVs-are implicated in cancer inheritance 15,16,[122][123][124] . Conversely, the abundance and typically larger size of somatic-CNVs detected in neoplastic cells imply that aneuploidy might be useful to maintain cellular proliferation and sustain tumour progression 18,96,[125][126][127] . Considering the frequency and size of CNVs detected in this study, it is more likely that they mainly represent somatic-CNVs acquired and accumulated during clonal expansion. However, paired germline samples were not included, consequently the NGS analysis employed did not distinguish germline and somatic variants, and sequencing results may contain both findings. According to the last standards and guidelines for the interpretation and reporting of sequence variants in cancer 123 , literature review and database queries might be helpful to determine whether the CNV has been reported as a recurrent germline variant. Consequently, we compared our data with somatic-CNVs detected in HBCs 19,27,28,[41][42][43] , somatic-and germline-CNVs validated by the CGC 59 , EMT-related genes affected by somatic-CNGs in multiple human cancer types 21 , and CNVs affecting FMC-and HBC-derived cell lines 19,84 .
As far as we know, there is only one publication characterising CNVs in cats according to different breeds (Genova et al. 121 ), and none characterising germline-CNVs associated with FMCs predisposition. When comparing our data with those of Genova et al., we observed that none of the animals in that study carried CNGs affecting B4 1-29 Mb and F2 64-82.3 Mb-enriched with cancer-related genes and significantly associated with poor outcome. Interestingly, in that study, three individuals carried two small (<0.02 Mb) CNLs comprised within B1 1-23 Mb (i.e. B1 14.5 Mb, and B1 20.7 Mb). Among them, B1 20.7 Mb harbours tumour suppressor PCM1 reported as deleted in this study and strongly associated with poor prognosis. Further studies comparing paired neoplastic-normal samples are necessary to disclose possible cancer-related germline-CNVs.
In this study, the DNA yields and 260/280 ratios of the two types of samples (FFPE and FT) were similar. However, the DNA isolated from four FFPE specimens stored for over 15 years was not suitable for library preparation. This may be associated with the effect of storage time and fixation process on DNA quality 128 . Moreover, in three samples (all LA subtype) no CNVs were detected. This might be related to the fact that DNA was isolated from all cellular subpopulations included in each sample. These samples were later excluded from CNVs analysis considering major contamination with healthy mammary tissue. This study may be biased because of a small sample size. In proportional hazards regression, the recommended number of events per variable (EPV) is 10 129 .
Due to the small sample size, several independent variables were characterised with a lower EPV value. Therefore, these results need to be validated in a larger series. Additionally, the wide confidence intervals observed may also be related to small sample bias 130 .
In summary, this study points up the role of specific CNVs as useful prognostic markers in FMCs and highlight the suitability of this tool to identify possible therapeutic targets and biomarkers within the rearranged regions. The CNVs landscape of FMCs included multiple rearranged genomic regions. Importantly, the amount and frequency of CNVs varied among molecular subtypes evaluated and were associated with prognosis.  www.nature.com/scientificreports www.nature.com/scientificreports/ Moreover, some aberrations were predominantly observed in TN-FMCs. Among rearrangements detected, CNGs in A3 (1-31 Mb), B4 (1-29 Mb), D4 , and F2 (64-82.3 Mb) were common and negatively influenced survival intervals. Moreover, those regions harbour several oncogenes and cancer-related genes enhancing cellular movement and migration-related pathways. Among CNGs, those affecting FCAs B4 and F2 negatively influenced DFS and cancer-specific OS in the multivariate analysis and included multiple EMT-associated genes. On the other hand, common CNLs in FCAs A2 (23-37.3 Mb and 37.3-48 Mb), B1 (1-23 Mb), and D1 (1.1-20 Mb,  were enriched with tumour repressors and encompassed genes involved in biological processes that would normally prevent cellular movement and migration. Among them, CNLs in B1 (1-23 Mb) remained associated with poor DFS in the multivariate analysis. Further efforts are necessary to sequence a larger series of FMC genomes, characterise somatic CNVs significantly associated with reduced survival, disclose possible germline-CNVs comprised within extensive rearrangements detected in this study, and investigate the suitability of cancer-related genes reported here as possible biomarkers and therapeutic targets on FMCs.

Methods
Animals. Thirty-three female cats surgically treated for FMC at the Small Animal Clinic of the University of Veterinary Medicine Hannover from 2000 to 2016 were enrolled in this study. Clinical history and reproductive status were obtained. Only cats, for which thoracic radiographs to investigate the presence of metastases at diagnosis and during the follow-up period were available, were included in this study. Patients were clinically staged using the modified World Health Organization (WHO) system 8 . Due to the previously reported reduced survival of cats with solid carcinomas (8 months) compared to tubulopapillary carcinomas (36 months) 1,72 , patients were grouped based on biological behaviour and morphology of the tumours into two categories (TC and SC). The group TC included tubulopapillary carcinomas, and the group SC comprised both solid carcinomas and comedocarcinomas representing six molecular subtypes. All cats underwent unilateral chain mastectomy, including resection of involved lymph nodes. The patients were followed up for a two-year post-operative period or until death. Long-term follow-up was obtained by clinical records analysis and telephone interviews with the owners. tissue samples. FFPE and/or FT tumour samples from 31 cats were retrospectively collected for histopathological examination and DNA isolation. In those cases, the FFPE specimens were retrieved from the archives of the Department of Pathology, University of Veterinary Medicine Hannover. The FT samples from patients retrospectively included were retrieved from the frozen tissue bank of the Small Animal Clinic, University of Veterinary Medicine Hannover. In two cases, FFPE and FT samples were prospectively collected. All samples were collected for diagnostic purposes during the medically necessary surgery after owner's written approval. Consequently, this study was not an animal experiment according to the German Animal Welfare Act and ethical approval was not required.
Histopathological examination. Paraffin sections (4 µm) of the tumour samples were stained with haematoxylin eosin (H&E) for histopathological evaluation. The samples were examined under light microscopy and the morphological diagnosis was performed following the WHO classification 7 . Histological grading of the tumours was performed using the EE system 35,36 . Additionally, we applied the MMEE grading system, and the feline-specific grading system proposed by Mills et al. (Mills-2015) 37 . For each case of our cohort, the cumulative number of mitoses was determined in 10 consecutive fields in the most mitotically active area with a field diameter of 0.575 mm (40x objective). The mitotic cut-offs for the EE and MMEE grading systems in our cohort were adapted according to the field diameter (0.51 in Castagnaro et al. 36 , 0.53 in Mills et al. 37 , and 0.575 in our study). For the Mills-2015 system, as suggested by Dagher et al. 38 , we determined the mitotic count cut-off by using receiver-operating characteristic (ROC) analysis for the study period OS-cancer specific survival rate of the cats included in this study. Therefore the mitotic count cut-off used in this study (≥35 mitoses in 10 high-power fields) was different from that proposed by Mills et al. 37 and used by Dagher et al. 38 (>62, and ≥33 mitoses in 10 high-power fields; respectively). Complete surgical excision and lymph node metastasis were histopathologically confirmed in all cases.
Immunohistochemical examination. The expression of ER, PR, HER2, CK5/6, and proliferation marker Ki-67 was evaluated by the avidin-biotin complex technique as previously described 131 . No animal was killed for the generation of positive controls, all tissue specimens used as positive controls for this study were sent to the Department of Pathology for diagnosis. Negative controls included normal serum and isotype control antibodies (details in Supplementary Table 8).
Tumours were classified into six subtypes (LA, LB fHER2+, LB fHER2−, fHER2+, normal-like TN and basal-like TN) on the basis of their ER, PR, fHER2, CK5/6, and Ki-67 expression as described elsewhere 9,10,132 , details in Supplementary Table 9. Tumours with an Allred score (Supplementary Table 10) equal to or higher than three were considered positive for ER or PR 10,133,134 . fHER2 expression was evaluated as described elsewhere 10,11,31,109,135 , Supplementary Table 11. Tumours scored as positive (3+) by IHC were considered as fHER2-positive for the St. Gallen classification. According to the last recommendations of the American Society of Clinical Oncology (ASCO) and the College of American Pathologists (CAP) for the validation of cases scored as HER2-2+ by IHC in HBC 31 , and in agreement with the high concordance reported between HER2-amplifications detected using FISH/ISH and CNGs detected by different high-resolution sequencing methods to validate HER2 status in human breast cancer (HBC) [136][137][138] , we correlated fHER2 IHC-based expression with CNGs affecting HER2 gene coding region (FCA E1 40,780,250-40,804,241 Mb). As reported elsewhere [136][137][138][139] , log2-ratio threshold for HER2 amplification was set at a higher amplitude (>0.5) for fHER2+ tumours detection to better show gene-level CNV events 140 . Accordingly, cases scored as 2+ with CNGs (log2-copy-number-ratio >0.5 141 ) in FCA E1 harbouring HER2 were considered as fHER2-positive. (2020) 10:1003 | https://doi.org/10.1038/s41598-020-57942-7 www.nature.com/scientificreports www.nature.com/scientificreports/ To identify tumours expressing basal markers, we evaluated the expression of CK5/6. Tumours were considered positive if more than 1% of the tumour cells showed positive cytoplasmic labelling 142 . The Ki-67 proliferation index was calculated as the percentage of positively stained tumour cell nuclei in 1000 tumour cells. Five to six microphotographs (40X) were randomly taken and analysed using Image J (Open Source Software, version 1.8.0; National Institutes of Health, Bethesda, MD, USA). Moreover, the Ki-67 index was considered as high when the number of positive nuclei was higher or equal than 14% 34 . DNA isolation. FFPE samples: four 10-µm-thick sections were sliced using a microtome (pfm Slide 2003, pfm medical ag). Afterwards, the sections were deparaffinised and the nucleic acids were isolated with the AllPrep DNA/RNA FFPE kit (QIAGEN) following the manufacturer's instructions. FT samples: the frozen tissue was previously homogenised using a TissueLyser (II-5 mm stainless steel bead, QIAGEN GmbH, Hilden, Germany). DNA was isolated using the AllPrep DNA/RNA Mini Kit (QIAGEN) following the manufacturer's instructions. The DNA yields were determined with the Synergy 2 microplate reader (BioTek) and the purity of each sample was determined by calculating the 260/280 ratio. Samples were stored at −80 °C until use. cnV analysis. DNA sequencing and CNVs analysis were performed according to the methodology described by Granados-Soler et al. 84 . 100-500 ng DNA was sheared to a size of 200-500 bp by ultrasound using a Covaris S2 focused ultrasonicator and sequencing libraries were prepared using the NEBNext UltraTM II DNA Library Preparation Kit for Illumina ® (New England Biolabs). Paired-end sequencing (38/37bp) was conducted on an Illumina NextSeq500 according to manufacturer's instructions. On average 36.3 Mb (SD: 5.5 Mb) reads were generated (Supplementary Table 5). Of those an average of 29.6 Mb (SD: 6 Mb) reads was mapped to the feline reference genome FelCat6.2 using BWA.
As recommended by Genova et al. 121 , two different programs were used to call large size aneuploidies by depth of coverage analysis and results were compared to reduce false-positive calls. First, we applied two different programs to generate log2-copy-number-ratios for each cancer sample: 1. CNV-seq 39 and 2. CNVKit 40 . For both analyses, a fixed window size of 2 Mb was used and tumour samples were compared against two normal reference samples obtained from feline healthy mammary frozen tissue. Control tissue was collected from two recently euthanised (for medical reasons) intact female Domestic Shorthair cats (five-and seven-years-old) with no pathologies of the mammary gland. DNA was isolated and stored using the same methodology employed for neoplastic FT samples (described above).
The log2-ratios obtained by the CNVKit method "WGS" were corrected for the highly aneuploid tumour samples using the method "centre mode" to shift the true neutral regions closer to zero. Second, the results (log2-ratios) of both algorithms were smoothed using a circular binary segmentation algorithm implemented in the program "Copynumber" 143 . For the CNV-seq data a gamma and kmin of 20 were used, while a gamma of 10 was used for the CNVKit data, because different from CNV-seq, CNVKit does not implement a sliding window binning approach resulting in only half the number of total bins. Third, from the segmented output only regions with smoothed-log2-copy-number-ratios of >0.2 or <−0.2 and >0.2 or <−0.1 for the CNV-seq and CNVKit data, respectively, were scored as significantly aberrant. Significant CNV-seq regions with >80% overlap to a significant CNVKit region were scored as validated. To refine the region boundaries, the start position of a validated CNV region was set to the maximum of both start positions and the stop position was set to the minimum of both stop positions. The validated and refined CNV regions of all samples were used for generating the frequency plots with the R package GenVisR for data generation 144 , and the Circos program 145 for plotting.
Common CNG/CNL segments were defined by frequency-changepoints, i.e. every region with a consistent frequency of >20% gains or losses across the population was defined as a common CNG or CNL segment that was used as independent variable in multivariate regression testing. Moreover, structural rearrangements detected (CNGs and CNLs) were compared with somatic-CNVs detected in HBCs 19,27,28,[41][42][43] , somatic-and germline-CNVs validated by the CGC 59 , EMT-related genes affected by somatic-CNGs in multiple human cancer types 21 , CNVs affecting FMC-and HBC-derived cell lines 19,84 , and additional cancer-associated genes previously reported on FMCs.
Genes in regions affected by CNVs were identified in the reference genome sequences provided by the UCSC Genome Bioinformatics Site (http://genome.ucsc.edu/). Additionally, to determine enriched biological terms, lists of the genes harboured in genomic regions most commonly affected by CNVs, and those correlated with poor outcome (reduced DFS and/or cancer-specific OS) were uploaded to the DAVID Bioinformatics server (http://david.abcc.ncifcrf.gov), which was employed for functional annotation clustering and KEGG pathway analysis 146,147 . Statistical analysis. The statistical software SPSS (IBM SPSS Statistics for Windows, Version 23.0. Armonk, NY, USA) was employed to perform all statistical analyses. For all statistical analyses, a p ≤ 0.05 was considered significant. Epidemiological, clinical, and histopathological variables included were age, breed, reproductive status, tumour size, lymph node metastasis (confirmed by histopathology), distant metastasis, modified WHO clinical stage 8 , and histological grading (EE, MMEE, and Mills-2015). CNVs, CNGs, and CNLs were categorised (CNVs score) into high or low according to whether the percentage of affected bins was lower or greater than their respective median values. Presence or absence of specific CNVs (CNGs and/or CNLs) affecting >20% of the sample size was included as categorical variables. Local recurrence, distant metastases, and death were considered as follow-up variables. Descriptive statistics of patients allocated in histological categories TC and SC regarding all epidemiological, clinical, histopathological and follow-up variables were performed. Percent agreement calculation was used to compare histological grades assigned (I, II and III) using the three systems employed and Ki-67 scores (low or high 34 ). Moreover, the level of agreement using the three different histological grading systems employed was compared using Cohen's Kappa test 148 . DFS and cancer-specific OS were defined as months