Phase I/II Trial of Vemurafenib in Dogs with Naturally Occurring, BRAF-mutated Urothelial Carcinoma

Mutated BRAF, especially BRAF^V6000E driving constitutive activation of the MAPK pathway, is a promising target for cancer therapy (1). MAPK pathway activation leads to increased cell proliferation and survival, evasion of apoptosis, and emergence of immune inhibitory processes facilitating cancer development and progression (1–4). BRAF^V600E is an important driver in 8% of all human cancer, including 50% of patients with metastatic melanoma, and smaller subsets of patients with bladder, thyroid, liver, colon, ovarian, lung, and pancreatic cancer (5–9). Although targeting BRAF^V600E has led to impressive remission of advanced cancer, particularly melanoma, innate and acquired resistance to therapy is common (2–5, 9). Other cancers with BRAF mutations such as colorectal and thyroid carcinomas are less responsive to BRAF-targeted therapy (8, 9). Resistance can involve reactivation of the classical MAPK pathway or activation of alternative signaling pathways (2–5, 10–13). Multiple studies also implicate a key role of the immune response in the success and failure of BRAF-targeted therapies (2–5, 10–14). In the majority of patients, however, the cause of relapse is unknown (2, 3, 5). Studies to better understand and circumvent the mechanisms of drug failure are high priorities.

Experimental animal models used in well-controlled studies are clearly instrumental in assessing mechanisms of resistance to BRAF-targeted therapy (3, 7, 10, 15). These models, however, often fail to represent the complexities of human cancer (16). Complementary models with diverse features of human cancer including cancer heterogeneity, mutational landscape, aggressive metastatic behavior, and host immunocompetence, are needed to better understand the success and failure of targeted therapies in humans, and to improve and personalize treatment. These complex features are present in a naturally occurring cancer model, pet dogs with invasive urinary bladder cancer, specifically invasive urothelial carcinoma (InvUC; ref. 17). Canine InvUC is a high-grade, heterogenous cancer that harbors molecular subtypes of human InvUC, and that is locally aggressive and metastatic to distant sites such as lung, liver, and bone in 50% of cases (17–19). Intriguingly, >80% of canine InvUC harbors a BRAF^V595E mutation, the canine homologue of human BRAF^V600E (17, 20, 21). Canine InvUC cells are killed in vitro by drugs that target BRAF^V600E (20, 22), although IC₅₀ values are in the μmol/L range. These concentrations are closer to those required to kill colon and thyroid carcinoma cells, in contrast to the nmol/L concentrations that kill human melanoma cells (5, 8, 9). These findings plus the opportunity to obtain cystoscopic tumor biopsies from dogs at multiple points during study, and the favorable acceptance of clinical trials in pet dogs, have enhanced interest in validating canine InvUC as a model to study and improve BRAF-targeted therapy (17). To help establish the relevance of the canine model, a clinical trial was performed to investigate the effects of a BRAF-targeted drug, vemurafenib, in dogs with BRAF-mutated InvUC. The trial included the: (i) evaluation of the safety, MTD, pharmacokinetics, and antitumor activity of vemurafenib, (ii) characterization of the cancer at the molecular level over the course of treatment, and (iii) assessment of target engagement and potential mechanisms of vemurafenib sensitivity and resistance, with comparison to findings reported in humans.

A phase I/II clinical trial of vemurafenib was performed in pet dogs with naturally occurring InvUC harboring BRAF^V595E at the Purdue University Veterinary Teaching Hospital (PUVTH, West Lafayette, IN) with approval of the Purdue Animal Care and Use Committee, and with informed dog owner consent. The pet dogs lived at home and made visits to the PUVTH for evaluation. Cystoscopic biopsies were obtained before and during treatment, and at cancer progression. RNA-sequencing (RNA-seq) analyses were performed to assess transcriptomic changes, and phosphoproteomic and IHC analyses were conducted to detect phosphorylated proteins and changes over time. Vemurafenib concentrations were measured in plasma and tumor tissues.

Inclusion criteria for participating dogs were: the presence of measurable histologically diagnosed lower urinary tract InvUC in which BRAF^V595E was detected in urine by droplet digital PCR (21), expected survival ≥ 6 weeks, and serum creatinine <2.0 mg/dL (normal reference range, 0.5–1.5 mg/dL). A washout period of 4 weeks for chemotherapy and 5 days for cyclooxygenase (COX) inhibitors was required. During the trial, other therapies including COX inhibitors, which can have antitumor activity against canine InvUC (17, 23), were not allowed.

Vemurafenib (Zelboraf, 240 mg tablets, Genentech) was administered in whole and half tablets with food with a starting dose of 25 mg/kg twice daily orally, based on earlier safety and pharmacokinetic studies in laboratory dogs. Drug tolerability was assessed by medical history, physical exams, complete blood counts, and serum biochemistry profiles. The vemurafenib dose was increased by 25 mg/kg twice daily in subsequent cohorts using a 3+3 trial design. When dose-limiting toxicity (DLT, VCOG-CTCAE grade 3 or 4 toxicity; ref. 24) occurred, the cohort was expanded to 6 dogs, and the dose further adjusted to determine the MTD, the dose at which no more than 1 of 6 dogs had DLT. Grade 3 abnormalities in serum alanine amino transferase (ALT) were not considered a DLT, as increased ALT in the absence of liver dysfunction is common in laboratory dogs and humans receiving vemurafenib (25). Cystoscopy was performed as previously described prior to vemurafenib (pre-vemurafenib), after 1 month of vemurafenib (vemurafenib-1-month), and when possible at cancer progression (vemurafenib-PD; ref. 26). For consistency in biopsy site, the same operator performed each cystoscopy guided by a tumor map incorporating cystoscopy and ultrasonography images. In individual dogs, the vemurafenib dose was reduced by 10% for grade 2 toxicity and 20% for grade 3 or higher toxicity. Dogs that experienced cancer progression or unacceptable toxicity following dose adjustments were eligible to receive other therapies off study.

The antitumor effects against the primary tumor were assessed at 4-week intervals by standardized cystosonography (27), and physical exam including rectal palpation of the urethra, prostate, and adjacent nodes. Thoracic radiography (three views), abdominal radiography (two views), and abdominal ultrasonography were performed at 8-week intervals. Primary tumor responses were defined as: complete remission (CR, no residual cancer detected), partial remission (PR, ≥50% decrease in tumor volume and no new tumor lesions), stable disease (SD, <50% change in tumor volume and no new tumor lesions), and progressive disease (PD, ≥50% increase in tumor volume or development of new tumor lesions; ref. 28). Metastases were assessed by RECIST (29). Permission to perform a necropsy was requested at the time of death.

Blood was collected into tubes containing lithium anticoagulant on day 1, and at 2 and 4 weeks into treatment, with collection before the morning vemurafenib dose, and at 1, 3, 6, and 10 hours after dosing in the phase I portion, and before the morning dose and at 3 and 6 hours after dosing in the phase II portion of the study. Plasma was stored at −80°C until analysis by high-performance LC/MS-MS (30). Vemurafenib concentrations were also measured in paired tumor tissues and blood at vemurafenib-1-month and vemurafenib-PD.

Twenty-four dogs were treated at the MTD to estimate the tumor response (31). Stepwise Cox regression analysis was performed to assess relationships between possible predictor variables (age, gender, high-risk breed for InvUC, tumor stage, grade, BRAF mutational fraction, molecular subtype, drug dosage, drug concentrations, toxicities) and tumor response and progression-free interval (PFI, time from the start of vemurafenib to PD). Univariable and multivariable models were applied. Associations between predictor variables and tumor response or distant metastases at death were assessed by logistic regression models, with ORs and 95% confidence intervals (CI) calculated. To examine associations between predictor variables and PFI, vemurafenib-related survival (start of vemurafenib until death), and overall survival (OS, diagnosis to death), Cox proportional hazards models were constructed, and HRs and 95% CIs calculated. Pearson correlation coefficient was used to determine the strength and direction of association between paired plasma and tumor vemurafenib concentrations with correlation of 0.1–0.3 considered weak, 0.31–0.5 moderate, and >0.5 being a strong correlation.

RNA-seq analyses were performed as previously described to identify genes, pathways, molecular subtype, and immune classification of potential relevance to the antitumor effects and resistance to vemurafenib (18). Cystoscopic tumor samples were immediately placed in TRIzol (Sigma), and RNA was isolated and purified (RNeasy, Qiagen). Normal bladder mucosa control samples were collected at necropsy from dogs without bladder disease who were euthanized because of other reasons. Tumor samples were batched and processed by Q2 Solutions. Directional RNA-seq libraries were created, validated, and run on an Illumina HiSeq system, yielding > 30 × 10⁶ clusters per sample with 2 × 150 paired end reads.

The RNA-seq analyses were performed using Strand NGS (Agilent Technologies). Raw sequence was passed through quality control and filtering, and aligned against canine genome CanFam3.1 (University of California, Santa Cruz Genome Database). Genes differentially expressed between normal bladder and InvUC were assessed pre-vemurafenib, vemurafenib-1-month, and at vemurafenib-PD. Pairwise comparison between response groups was used to detect differential expression by applying DESEQ2 on DESEQ normalized data and edge R on TMM normalized data (fold change ≥2, P_corr < 0.05) of genes at the three timepoints (32). Ingenuity Pathway Analysis (IPA) and Gene Ontology (GO) analyses were conducted. The RNA-seq data were also specifically interrogated for: genes involved in MAPK signaling and genes reported to be associated with vemurafenib treatment results (2–5, 10–13, 33), genes used to assign InvUC into luminal and basal molecular subtypes (18, 19, 34), and genes used to characterize the tumor immune state (35). The immune-related genes were those reported to classify human bladder cancer as non-T cell–inflamed (not infiltrated by T lymphocytes, immune “cold”), or T cell–inflamed (infiltrated by T lymphocytes, immune “hot”; ref. 35). A semiquantitative approach based on the percentage of immune genes expressed in each tumor was used to assign each tumor an “immune score” as follows: 1 = cold, 2 = cool, 3 = neutral, 4 = warm, and 5 = hot. In additional analyses, CIBERSORT (Alizadeh Lab, Stanford University, Stanford, CA) was used to estimate the abundances of specific immune cell types within the immune infiltrate (36). The RNA-seq data and correlative clinical data were deposited in the NCI's Integrated Canine Data Commons: Accession ID ICDC000004.

Biopsy samples were homogenized, treated with protease and phosphatase inhibitors, and 500 μg of protein was used for phosphopeptide enrichment. Reduced, alkylated, and trypsin-digested peptides were desalted (Pierce Peptide desalting spin columns, Thermo Fisher Scientific) and enriched for phosphopeptides (Spin-Tip PolyMAC Phosphopeptide Enrichment kit, Tymora Analytical) as described previously (37). Samples were analyzed by a Dionex UltiMate 3000 RSLC Nano System coupled to Q Exactive High Field Hybrid Quadrupole-Orbitrap Mass Spectrometer (Thermo Fisher Scientific; ref. 37). LC/MS-MS raw data were searched against Canis familiaris protein database using MaxQuant, and the MaxQuant output analyzed with Perseus software (38, 39). Proteins and phosphosites identified as “contaminants,” “reverse,” and “only identified by site” were filtered out. Samples were grouped, and proteins present in two or more replicates in at least one group were retained for further analysis. Missing values were replaced with values based on the normal distribution, and ANOVA test was used for significance analysis.

IHC to detect phosphorylated ERK (pERK; Clone 137F5, Cell Signaling Technology) was performed on formalin-fixed paraffin-embedded InvUC sections. The percentage of immunoreactive tumor cells and staining intensity were recorded (40, 41).

The subject and tumor characteristics of 34 dogs enrolled in the trial are summarized in Table 1. Four other dogs were presented for the study, but did not meet enrollment criteria. The MTD of vemurafenib was 37.5 mg/kg twice daily orally. Three, 5, and 6 dogs were treated with 25, 50, and 37.5 mg/kg, respectively, with the DLTs being gastrointestinal upset. The most common adverse event above the MTD was anorexia (grade 2 in 3 dogs, grade 3 in 2 dogs). In the phase II portion of the trial, mild to moderate anorexia was common (10% grade 1, 30% grade 2, and 10% grade 3) which resolved with dose reduction. All dogs except 1 dog treated at the MTD had elevated ALT (17% grade 1, 4% grade 2, 79% grade 3). Changes which could indicate liver dysfunction such as reduction in blood glucose, blood urea nitrogen, or albumin were not observed. Six dogs (18%) including 4 dogs receiving 37.5 mg/kg twice daily and 2 dogs receiving 50 mg/kg twice daily vemurafenib, developed multiple new skin masses including squamous cell carcinoma and squamous papillomas during vemurafenib treatment, akin to pharmacodynamic effects in humans related to paradoxical MAPK pathway activation (Supplementary Fig. S1; ref. 43).

The results of the pharmacokinetic analyses are summarized in Fig. 1, with considerable variation noted between dogs. Two dog owners reported at times finding vemurafenib pills largely intact in the feces indicating inconsistencies in drug absorption.

The tumor responses in 31 dogs across dose groups are summarized in Fig. 2 including PR in 38% and SD in 54% of 24 dogs treated at the MTD. Vemurafenib was discontinued in 3 dogs due to gastrointestinal toxicity before tumor response could be determined. The odds of remission significantly increased with increasing vemurafenib dose (OR, 1.16; 95% CI, 0.99–1.35; P = 0.019), and trended up nonsignificantly in high-risk breed dogs (53%) versus other dogs (25%; OR, 3.43; 95% CI, 0.75–15.67; P = 0.103). The median PFI was 181 days (range, 53–608 days) for dogs receiving 37.5 mg/kg vemurafenib, and 179 days (range, 29–767 days) for all dogs (Supplementary Fig. S2). Following vemurafenib failure, 19 dogs received COX inhibitors, and 10 dogs received chemotherapy plus COX inhibitors. The median vemurafenib-related survival was 272 days (range, 18–1,252 days) and median OS was 354 days (range, 19–1,272 days), with 2 dogs alive at the end of the study (Supplementary Fig. S2). Male dogs had significantly longer vemurafenib-related survival (median, 314 days; range, 42–980 days) than female dogs (median, 187 days; range, 18–1,252 days; HR, 0.36; 95% CI, 0.14–0.91; P = 0.031). Dogs that had received other therapies prior to vemurafenib had longer OS (median 398 days; range, 127–1,261 days) than dogs that had not received prior therapy (median, 258 days; range, 19–1,272 days; HR, 0.22; 95% CI, 0.07–0.75; P = 0.011).

Distant metastases were present in 14 of 26 dogs (54%) with known tumor–node–metastasis stage at death, including 21 dogs undergoing necropsy. Metastases were significantly more common in male dogs (11/14 dogs, 79%) than female dogs (4/12 dogs, 33%; OR, 5.83; 95% CI, 0.98–34.64; P = 0.042), but did not differ between male dogs with prostatic involvement of the cancer versus those without. Distant metastases were less common in dogs that had PR (50%) than in those without PR (79%; P = 0.020).

Tumor tissues were available for RNA-seq analyses from 29 dogs pre-vemurafenib, 25 dogs at vemurafenib-1-month, and 17 dogs at vemurafenib-PD. In IPA and GO analyses, differentially expressed genes involved signaling pathways, immune processes, membrane channels, transporter activity, cell metabolism, and energy (Supplementary Table S1).

Heatmaps were generated to observe changes in the classical MAPK, stress-JNK, and PI3K/AKT/mTOR pathways, and other genes linked to vemurafenib resistance (2–5, 10–13, 44), with heterogeneity noted within and between cases (Fig. 3; Supplementary Fig. S3; Supplementary Table S2). One or more MAPK pathway genes was upregulated in all dogs pre-vemurafenib (Fig. 3). ERK, which is downstream of BRAF and a marker of MAPK pathway activity, was upregulated to some extent in 22 of 25 cases (Fig. 3; Supplementary Fig. S4). ERK expression decreased in 20 of these 22 cases by vemurafenib-1-month, and increased in 10 of 17 dogs with available samples at vemurafenib-PD (Supplementary Fig. S4). BRAF expression was upregulated in 10 of 25 cases pre-vemurafenib (Fig. 3; Supplementary Fig. S4). BRAF expression decreased in 9 of these 10 dogs by vemurafenib-1-month, and increased in 10 of 17 dogs with samples available at vemurafenib-PD. Changes in BRAF and ERK expression were not associated with tumor response. There was a general increase in RRAS, MRAS, and PI3K/AKT/mTOR pathway genes, and a decrease in NRAS expression with vemurafenib treatment (Fig. 3). Changes in other genes implicated in vemurafenib resistance were noted in subsets of dogs including enhanced expression of CCND1, PDGFRB, EGFR, AXL, and EphA2, and loss of PTEN and NF1 at progression (Fig. 3; Supplementary Table S2; refs. 2, 3, 12, 13, 44). Neither the initial urine BRAF mutation fraction (median 32% in dogs with PR, 25% in dogs with SD/PD) nor changes during treatment, were associated with outcome.

The phosphoproteome analysis revealed upregulation of the phosphorylated forms of several kinases involved in the classical MAPK, stress-JNK, and PI3K/AKT/mTOR pathways consistent with the RNA-seq analyses. The most abundant phosphorylated sites mapped to specific proteins and changes in these with treatment are summarized in Fig. 4, Supplementary Fig. S5, and Supplementary Table S3. InvUC tissue samples for pERK IHC were available from 8 dogs before and after vemurafenib treatment. Marked pERK expression was noted in 5 dogs pre-vemurafenib. The expression decreased in 4 of the 5 dogs at vemurafenib-1-month (Fig. 4). The IHC findings confirmed target engagement, and were consistent with RNA-seq and phosphoproteomic analyses.

Examination of the pre-vemurafenib RNA-seq data revealed nine basal and 20 luminal subtype tumors (Supplementary Fig. S6). Male dogs were underrepresented in the basal tumor group, with one of nine cases being male, but not in luminal tumor group (11 male, 9 female dogs). The remission rate did not differ significantly between dogs with basal tumors (33%) and dogs with luminal tumors (44%; OR, 0.63; 95% CI, 0.12–3.32; P = 0.5772). A basal subtype at vemurafenib-PD was associated with shorter PFI (HR, 11.08; 95% CI, 0.95–129.6; P = 0.055) in multivariate analyses independent of immune changes discussed below, and a trend toward a higher distant metastatic rate at death (OR, 5.25; 95% CI, 0.4–68.9; P = 0.1787).

The expression of immune-related genes (n = 1,797 genes) used to classify the immune state of human InvUC (35) are summarized in Fig. 5. The RNA-seq immune scores at diagnosis were from coldest to hottest: 1 (n = 3 dogs), 2 (n = 7), 3 (n = 7), 4 (n = 9), and 5 (n = 3). Although significant differences were not found, the median PFI was 152 days (range, 82–482 days) for dogs with immune scores ≤2, and 248 days (range, 53–605 days) for dogs with immune scores ≥4. Remission occurred in 4 of 11 dogs (45%) with an immune score ≥ 4, and in 3 of 9 dogs (33%) with an immune score ≤2. A higher tumor immune score pre-vemurafenib was associated with the basal subtype (Fig. 5). Nine of 12 tumors (75%) with an immune score ≥4 were basal, and three tumors (25%) were luminal. All of the 17 tumors with an immune score of ≤3 were luminal (Fig. 5).

The immune score increased between pre-vemurafenib and vemurafenib-1-month in 17 dogs (68%), remained the same in 4 dogs (16%), and decreased in 4 dogs (16%). Remission occurred in 53% of dogs with an increasing immune score, and 25% of dogs with stable or declining immune score (OR, 3.38; 95% CI, 0.42–21.73; P = 0.181). A higher immune score at vemurafenib-1-month was significantly associated with longer survival (HR, 0.14; 95% CI, 0.03–0.62; P = 0.0338), being most apparent when comparing scores ≤2 to ≥4. In 10 dogs with an immune score ≤2, the median vemurafenib-related survival was 185 days (range, 18–921 days), and OS was 259 days (range, 19–1,261 days). In 12 dogs with an immune score ≥4, the median vemurafenib-related survival was 301 days (range, 106–1,252 days), and OS was 357 days (range, 158–1,272 days). Between vemurafenib-1-month and vemurafenib-PD, the immune score decreased in 16 of 17 dogs (94%), and increased in one dog (6%), with examples of transcriptome changes in Fig. 5 and Supplementary Table S2.

To identify patterns in immune-related genes, heatmaps were generated from lists of genes of importance in the immune response to cancer, including those reported to be affected by vemurafenib (Fig. 5; refs. 2–5, 10–13, 48). There was a notable increase in the expression of immune enhancing genes at vemurafenib-1-month and downregulation of these genes at vemurafenib-PD (Fig. 5; Supplementary Table S2). At vemurafenib-PD, the majority of cases (76%) had upregulation of one or more immune checkpoints including CTLA-4, PD-L1, PD-1, and B7X (Supplementary Table S2). CIBERSORT analyses (Fig. 5) confirmed heterogeneity between cases, but demonstrated trends for an increase in CD8 T-cell signatures at vemurafenib-1-month, and for a decrease in CD8 T cell and type 1 macrophage signatures and an increase in regulatory T cell signatures at vemurafenib-PD.

The important findings from the study include: (i) the responses to vemurafenib in dogs which mimic those reported in humans including good initial antitumor activity followed by drug resistance, as well as pharmacodynamic cutaneous effects, (ii) immune signatures in dogs that parallel those in humans in regards to immune hot signatures early in vemurafenib treatment and immune cold signatures at cancer progression, and (iii) no new safety signals (2–5, 10–13, 33). These findings support the relevance of the canine model, and inclusion of dogs with naturally occurring InvUC in studies to help inform advances in BRAF-targeted therapy in humans.

Vemurafenib demonstrated good antitumor activity in dogs with InvUC. The 38% remission rate and 6-month PFI compare favorably with remission rates of <25% and PFIs of approximately 4 months resulting from other single-agent therapies in dogs with InvUC with most remissions being PR across all studies (17, 23, 28). While this is considered a positive response in a cancer like InvUC where drug resistance is common, the remission rate appears lower than that reported in human melanoma patients in which 50% remission rates are typical with some CRs observed (1). It is possible that the effects of vemurafenib in canine InvUC are more akin to those in human thyroid and colon carcinomas in which upstream drivers can temper the response leading to more modest remission rates in the 10%–30% range (8, 9). It is also recognized that the vemurafenib formulation has been optimized for humans, not dogs (49). The pharmacokinetics varied more in dogs than in humans (49), which could be due in part to the shorter vemurafenib half-life in dogs and inconsistent absorption. While the vemurafenib concentrations in dogs (e.g., median plasma steady state concentration calculated from a partial AUC at 2 weeks of 20 μg/mL or 40.8 μmol/L) were often higher than those required to kill canine InvUC cells in vitro (IC₅₀ 10 μmol/L in BRAF-mutated cells, 30 μmol/L in BRAF wild-type cells; ref. 20), the concentrations were approximately half of the target vemurafenib plasma concentrations in human patients with cancer (31–39 μg/mL; ref. 49), and concentrations were difficult to maintain for constant target inhibition. New skin lesions including cutaneous squamous cell carcinoma and squamous papillomas in 18% of dogs receiving vemurafenib were of particular interest because these are considered manageable pharmacodynamic effects of vemurafenib that occur in 10%–30% of humans (43). These are attributed to paradoxical pathway activation in cells with underlying RAS mutations (43). The findings in dogs support effective target engagement, and a similar frequency of this paradoxical pathway activation in dogs and humans (43). Elevation in ALT occurred more frequently in dogs than in humans (25), but liver dysfunction was not detected.

The RNA-seq immune patterns were one of the more interesting findings. These patterns were present regardless of the expression of MAPK pathway genes and other signaling genes. Importantly, the immune score increased, that is, became more immune hot, in two-thirds of the dogs at vemurafenib-1-month. At cancer progression, the immune score decreased, that is, became more immune cold, in 94% of the dogs. These findings correlate with studies of tumor tissues from humans treated with BRAF inhibitors and from experimental animal work lending further validity to the canine InvUC model (2–5, 10–14, 33). Further study is indicated as the immune state is thought to have profound effects on the response to several different types of cancer therapy and outcomes in humans (35, 36, 48). In general, immune cold tumors are typically less responsive, and immune hot tumors more responsive to cancer drugs, especially immunotherapies (35, 48). Interest is high in identifying drugs that enhance immune signatures and which could improve the efficacy of other cancer therapies.

There are several possible mechanisms by which vemurafenib could enhance immune signatures. On a general level, there is growing evidence that the anticancer effects of immunotherapy, chemotherapy, and radiation are due at least in part to the release of neoantigens and damage associated molecular proteins from dying cancer cells that then trigger or enhance an immune attack against the cancer (48). In melanoma, signaling activity driven by mutated BRAF has been associated with decreased expression of tumor-associated antigens, and BRAF inhibitor therapy has resulted in enhanced expression of tumor antigens and MHC, increased CD8⁺ TILs, reduction in immunosuppressive cytokines IL6 and IL8, and increased markers of T-cell cytotoxicity (2–5, 11, 13, 14, 33). In addition to apoptotic cell death, melanoma cells treated with BRAF inhibitors can undergo pyroptosis in which pores release inflammatory cytokines activating dendritic cells (4). Interestingly, as the antigen expression and TILs increase with BRAF inhibitor treatment, increased T-cell exhaustion markers such as TIM-3 and PD1 also occurs, suggesting an optimal window for enhanced immune effects should be further elucidated (2). At melanoma progression on BRAF-targeted therapy, reduced antigen expression, decreased TILs, and influx of tumor associated macrophages have been observed (2, 3, 11–13, 33). Similar changes were observed in the tumors in the dogs in the trial. This provides support for including dog studies to potentially help optimize and validate combination immunotherapy-BRAF inhibitor therapies.

In analyzing the RNA-seq data for changes in pathways of expected importance in vemurafenib response and mechanisms of resistance reported in humans (2, 3, 5, 10, 11), all of the dogs had one or more changes in these genes or pathways at vemurafenib-PD. A specific consistent change driving remission or short versus long PFI across the majority of cases, however, was not observed. BRAF^V600E causes constitutive activation of the MAPK pathway, mainly through the classical MAPK pathway. When this pathway is blocked, the stress-JNK, PI3K/AKT/mTOR, and other alternative pathways become important. All three pathways converge on shared nuclear targets. As expected, one or more genes in the classical MAPK pathway were overexpressed in the majority of dogs pre-vemurafenib. Similarly, ERK, which is downstream of BRAF and an indicator of MAPK pathway activity, was upregulated in 84% of cases pre-vemurafenib. As expected, ERK expression decreased in 91% of cases at vemurafenib-1-month. ERK expression increased again in 59% of cases at vemurafenib-PD consistent with restored MAPK pathway–related signaling as resistance developed. The pERK IHC results were also consistent with these findings, as were the phosphoproteomic results of reduction in phosphorylated BRAF and related MAPK proteins at vemurafenib-1-month and enhanced expression at cancer progression. There are many complexities, however, in interpreting the results. Upregulation of most genes in the MAPK pathway was not observed, suggesting multiple entry and exit points to alternative signaling pathways and complex pathway regulation, which has been suggested in other studies (50). Similarly, the phosphoproteomic results suggest vemurafenib induces diverse regulatory alterations in the MAPK pathways indicating complex interactions across the pathways.

In addition to the increase in expression of various genes in the classical MAPK, stress-JNK, and PI3K/AKT/mTOR pathways and phosphorylated proteins at cancer progression, additional transcriptomic changes were found in the dog tissues that have been associated with vemurafenib resistance in humans. These included loss of PTEN and NF1, and increased expression of CCND1, PDGFRB, EGFR, AXL, and EphA2(2, 3, 5, 10, 13, 44). Other genes and processes not previously linked to vemurafenib effects were also identified (Supplementary Table S1), including upregulation of genes involving membrane channels, transporter activity, other signaling, and cell metabolism and energy. Vemurafenib, like many drugs, could have a wide range of effects.

Evidence is building for the role of molecular subtypes in human bladder cancer (19, 34), and thus the intriguing association between molecular subtype and immune state in the dogs warrants further study. In humans, basal subtype bladder cancer is associated with a hotter immune state, and luminal tumors with a colder immune state, as was observed in the dogs in this study (19, 34). In other cross-species similarities, male dogs were underrepresented in the basal tumor group as occurs in men with InvUC (19, 34). In parallel with the more aggressive cancer behavior of the basal subtype in humans (19, 34), the basal subtype in dogs, especially at progression was associated with shorter PFI and higher rate of distant metastases at death. It is interesting that while BRAF mutations are not common in human InvUC, different mutational events can converge into similar molecular subtypes (18, 19, 34).

The heterogeneity in signaling and immune pathways in the dogs is viewed as a positive finding, and a strength of the canine model in regards to its relevance to the human condition and studies to personalize cancer therapy in a very complex disease. The canine model can complement other preclinical models that are essential in offering a controlled more homogenous environment for mechanism studies. Dog studies can be conducted to test combination therapy approaches that are more likely to address multiple oncogenic and immunologic mechanisms. In addition, as the “big data science” field continues to develop, the canine trial data, which have been deposited in the NCI's Integrated Canine Data Commons, offers a resource for further data analysis methods and application, especially when integrating many types of complex data, and performing cross-species analyses.

In conclusion, the study demonstrates many similarities in the outcomes of BRAF-targeted therapy in dogs and humans. This supports the relevance of the canine model and further evaluation and application of naturally occurring InvUC bearing the BRAF^V595E mutation to help inform BRAF-targeted therapy approaches and combination therapies in humans. The development of canine large molecule therapies similar to those in humans for cross-species combination therapy studies is critical. This collective work will also be beneficial in a veterinary oncology setting as well as being valuable for translational research.

P. Rossman reports grants from Puppy Up Foundation and other support from Genentech, Inc. during the conduct of the study. D. Tuerck reports other support from F. Hoffmann La-Roche AG, Basel, Switzerland, in the United States also known as Genentech outside the submitted work; and in addition to being an employee of F. Hoffmann La-Roche in Basel, Switzerland, (manufacturer of Zelboraf, INN: vemurafenib). D. Tuerck owns a limited amount of nonvoting shares of his employer. R. Mohallem reports other support from Genentech and grants from Puppy Up Foundation during the conduct of the study. M. Merchant reports personal fees from Genentech/Roche outside the submitted work. C.M. Fulkerson reports grants from Puppy Up Foundation and non-financial support from Genentech during the conduct of the study. E. Murray reports other support from Genentech, Inc. during the conduct of the study; other support from Genentech, Inc. outside the submitted work. N. Dybdal reports other support from Genentech, Inc. outside the submitted work. L.M. Fourez reports other support from Genentech during the conduct of the study. D.W. Knapp reports other support from Genentech and Scottish Terrier Club of America, and grants from Puppy Up Foundation and NCI during the conduct of the study. No disclosures were reported by the other authors.

P. Rossman: Conceptualization, data curation, formal analysis, investigation, writing–original draft, writing–review and editing. T.S. Zabka: Conceptualization, resources, supervision, funding acquisition, investigation, methodology, writing–original draft, project administration, writing–review and editing. A. Ruple: Data curation, software, formal analysis, validation, investigation, methodology, writing–original draft, writing–review and editing. D. Tuerck: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–original draft, writing–review and editing. J.A. Ramos-Vara: Data curation, formal analysis, validation, investigation, visualization, methodology, writing–review and editing. L. Liu: Data curation, formal analysis, investigation, methodology, writing–review and editing. R. Mohallem: Resources, data curation, software, formal analysis, investigation, visualization, methodology, writing–original draft, writing–review and editing. M. Merchant: Resources, data curation, investigation, methodology, writing–review and editing. J. Franco: Data curation, software, formal analysis, validation, investigation, methodology, writing–review and editing. C.M. Fulkerson: Conceptualization, data curation, investigation, writing–review and editing. K.P. Bhide: Conceptualization, resources, software, formal analysis, investigation, methodology, writing–review and editing. M. Breen: Data curation, software, formal analysis, validation, investigation, writing–review and editing. U.K. Aryal: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, project administration, writing–review and editing. E. Murray: Investigation, writing–review and editing. N. Dybdal: Conceptualization, resources, data curation, software, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. S.M. Utturkar: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. L.M. Fourez: Data curation, software, formal analysis, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. A.W. Enstrom: Data curation, software, formal analysis, investigation, methodology, writing–review and editing. D. Dhawan: Conceptualization, data curation, software, formal analysis, supervision, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing. D.W. Knapp: Conceptualization, resources, data curation, software, formal analysis, supervision, funding acquisition, validation, investigation, visualization, methodology, writing–original draft, project administration, writing–review and editing.

The authors would like to thank the dedicated clinicians and staff of the Purdue Comparative Oncology Program for their work on this project. The authors also acknowledge the important contributions of Dawn Colburn and Tony Contreras of the global Roche vemurafenib team for their support to release clinical-grade drug; Lesley Cunningham a prior DMPK Genentech scientist for enabling pharmacokinetic analysis; Aaron Fullerton of Genentech Safety Assessment for enabling mRNA submission and analysis; John Moffat of Genentech research for his initial input on cell line biology; Rochelle Pando of the Genentech contracts team for establishing the framework of our collaboration; and Silvia Kimpfler of Roche to organize shipment of clinical drug. We also wish to express our sincere gratitude to the dog owners for allowing their dogs to participate and for supporting the study. All the LC/MS-MS data for phosphoproteome analysis were collected at the Purdue Proteomics Facility, Bindley Bioscience Center of Purdue University.

This study was supported in part by the Puppy Up Foundation and by F. Hoffmann-La Roche AG.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.