A Phase II Study of Dovitinib in Patients with Recurrent or Metastatic Adenoid Cystic Carcinoma

Adenoid cystic carcinomas (ACC) are rare malignancies that afflict 1,200 individuals per year in the United States. The most common primary sites are the salivary gland and other regions in the head and neck, but they may also arise in the skin, breast, and uterus (1). Localized ACCs are usually treated with surgery and/or radiotherapy, but local and distant recurrences are frequent due to the tumor's propensity for perineural infiltration and to metastasize to the lung. Although the median survival for patients with metastatic ACC is about 4 years, the disease often becomes more aggressive later in the course, such that the median survival for patients whose disease had progressed within the previous 6 months is in the range of 18 to 20 months (1–4). Unfortunately, there is no effective systemic treatment for advanced ACC, as standard cytotoxic chemotherapy agents produce tumor response rates in the range of 5% to 15% without a clear impact on patient survival (5–7). Targeted agents, including cetuximab, erlotinib, gefitinib, vorinostat, and others, have also shown limited activity against ACCs (3, 4, 8, 9). One potential target, the tyrosine kinase receptor KIT, is overexpressed in the majority of these tumors, but treatment of ACC patients with the KIT inhibitor, imatinib, produced no or little clinical benefit (10–12). Similarly, the multikinase inhibitors sunitinib and sorafenib, whose targets include VEGFR and PDGFR, rarely cause objective responses, although they may confer a modest improvement in progression-free survival (PFS; refs. 3, 4).

Genetic analyses have shown that ACCs frequently acquire chromosomal translocations such as t(6;9)(q22–23;p23–24) that result in the overexpression of the oncogenic transcription factors MYB or MYBL1 (13–19). The relevance of MYB to the pathogenesis of ACCs is supported by the observation that inhibition of its transcriptional regulatory functions by the BET domain inhibitor JQ1 suppresses the growth of lower grade ACCs propagated as primary xenografts (15). An alternative is to target downstream effectors of MYB, which may include the FGFRs. Among the potential consequences of MYB overexpression appears to be upregulation of FGFR2 and its ligand FGF2, perhaps related to MYB DNA-binding sites that are located near the corresponding genes (15, 17, 20). This, perhaps in conjunction with elevated FGFR1 expression, could result in autocrine receptor signaling and promote tumor growth (21). This scenario is consistent with proteomic studies of primary ACC xenografts that detected spontaneous phosphorylation of peptides derived from the carboxy terminus of FGFR1 (22). In addition, 4% to 12% of ACCs harbor mutations that are predicted to activate or enhance FGFR signaling independently of MYB overexpression (16, 19). A role for FGFR signaling in the pathogenesis of ACC is further supported by the observation that inhibitors of the receptor kinase can suppress the growth of primary ACC xenografts (23).

Dovitinib (TKI258) is a multityrosine kinase inhibitor that not only inhibits the VEGFR, PDGFR, c-Kit, CSF-1R, RET, TrkA, and FLT3 receptor kinases, but also FGFRs 1–3 (IC₅₀ of 10 nmol/L; refs. 24, 25). This contrasts with sorafenib and sunitinib that have similar activity against the VEGFRs and other shared targets but have only weak activity against the FGFRs (IC₅₀ = 580 and 880 nmol/L, respectively). Given that MYB gene alterations are found in the majority of ACCs and may drive autocrine activation of FGFRs, we postulated that the ability of dovitinib to inhibit signaling by these receptors would result in a high tumor response rate in patients with advanced and progressive tumors. To test this hypothesis, we conducted a phase II study of dovitinib in patients whose tumors had progressed in the past 6 months. Objective tumor response was set as the primary endpoint, but we also measured alterations in the tumor growth rate (TGR) as determined by change point analysis to detect more subtle antitumor drug effects that may be clinically relevant. Other secondary endpoints included estimation of the PFS, overall survival (OS) and clinical benefit rate, evaluation of the adverse events, and descriptions of the early metabolic response rate, changes in quality of life (QOL), and correlation between biomarkers and clinical outcome.

Eligible patients were 18 years of age or older with a diagnosis of ACC confirmed by expert review (Drs. C.A. Moskaluk and H. Frierson). All patients had unresectable and/or metastatic measurable disease with evidence of progression within the previous 6 months based on imaging and RECIST 2.0 criteria. All patients were required to submit remote pretreatment cross-sectional films (minimum of 3 months prior to baseline film) to calculate the pretreatment TGR. Up to five target lesions were chosen from scans 3 to 6 months prior to enrollment and assessed by a radiologist. There was no limit on the number of prior therapies, but patients must not have had chemotherapy, radiotherapy, surgery, or investigational treatments in the previous 4 weeks. Other eligibility criteria were as follows: ECOG performance status of 0 to 2, life expectancy of at least 16 weeks, neutrophil count ≥1,500/μL, platelet count ≥100,000/μL, hemoglobin >9 g/dL, total bilirubin ≤1.5× upper limit of normal (ULN), AST and/or ALT ≤3× ULN, creatinine ≤1.5× ULN or creatinine clearance ≥30 mL/minute, absence of brain metastases, no other active malignancy in the past 3 years (adequately treated cervical carcinoma in situ and nonmelanoma skin cancers excepted), no serious medical conditions, left ventricular ejection fraction >45%, and absence of uncontrolled hypertension, malabsorption, cirrhosis, warfarin therapy, pregnancy, and active breastfeeding. Referring hospitals were asked to submit tissue blocks for biomarker analysis. This study was approved by the University of Virginia Institutional Review Board for Health Sciences Research, and patients signed informed consent before study entry.

This was a modified two-step phase II clinical trial in which patients were treated with dovitinib (supplied by Novartis Pharmaceuticals Corporation) at a starting dose of 500 mg taken orally for 5 days on and 2 days off; cycle length was 28 days. Treatment was continued until disease progression, unacceptable toxicity, patient refusal, or at the physician's discretion. Dose reductions and delays were allowed as per protocol. Toxicity was graded according to NCI Common Terminology Criteria for Adverse Events (CTCAE), version 4.03. For intolerable grade 2, or for grade 3 or 4 toxicity, dovitinib was withheld until the toxicity returned to grade 1 or baseline, and the dose was reduced to 400 mg at the same schedule. If the toxicity recurred, the drug was stopped again until the toxicity returned to grade 1 or baseline, and the dose was reduced to 300 mg. For asymptomatic grade 3 hypertension, hyperlipidemia, or cytopenias, dovitinib was withheld until the toxicity resolved to grade 1 and then resumed at the same dose or decreased one dose level at physician discretion. If the grade 3 or 4 toxicity recurred, the drug was withheld until recovery to grade 1, and the dose was again reduced. For a third recurrence of any grade 3 or 4 toxicity, the drug was discontinued. Patients were also taken off treatment if the interruption for toxicity lasted longer than 21 days without recovery to grade 1 or baseline or if they developed a grade 4 toxicity that was judged to be life threatening.

The primary endpoints were objective tumor response rate [ORR; complete (CR) and partial responses (PR)] and drug-induced changes in the TGR using the change point method (26). Tumor response was determined by RECIST 1.1 criteria from CT or magnetic resonance images obtained at 2, 4, and 8 months then every 4 cycles until progression. The modified two-stage trial design incorporated the alternative hypothesis that dovitinib would produce an ORR of at least 18% as compared with 1%, with approximate power of 90% and a one-sided type I error rate of 5%. The first stage allowed the study to be terminated for drug inactivity if no responses were seen in the first 16 patients. If at least one objective response was observed, 5 additional patients would be enrolled. If ≥2 of the 21 patients experienced an objective response, a subsequent protocol amendment recommended the recruitment of 14 additional evaluable patients to more precisely evaluate the change in the TGR. The two stage design was chosen to allow an early “go or no go” decision to minimize patient exposure to dovitinib.

For change point analysis, piecewise linear models of tumor growth were fit to describe a patient's tumor growth profile while on treatment compared with their pretreatment profile (using up to five index lesions from prebaseline films). A preliminary analysis of data from 5 untreated ACC patients allowed us to estimate the variance and power to detect reductions in slopes using change point analysis. In each model, we allowed for one change point, one slope measured from −6 months to time 0 (TG0) compared with the slope (TG1) for the period from study entry to a minimum of 4 months after the start of treatment (time 0 to time 4 months). The effect of treatment on tumor growth was assessed by measuring the change in slope between the pre- and posttreatment periods. Using the estimated variance of change in tumor growth from studies of untreated patients (data not shown), we designed this study to detect a change in the TGR. With a total of 35 patients, there would be 80% power to detect an absolute reduction of 0.62 in the slope of the line that represents the tumor growth rate.

Secondary objectives were to estimate PFS, the clinical benefit rate (CR + PR + stable disease >4 months), OS, and to determine the adverse event profile. Exploratory analyses included evaluation of early metabolic tumor response rate (day 12), changes in QOL, and biomarker studies. Early metabolic responses were defined as 25% reduction in the maximal standard uptake (SUV) of 18F-fluorodeoxyglucose (¹⁸FDG) on PET scans done on day 12 as compared with pretreatment values (27). QOL outcomes were monitored by the FACT-G questionnaires at each clinic visit. The biomarker studies looked for potential correlations between MYB:NFIB chromosomal translocation and the expression of c-MYB, phosphorylated FGFR1, and clinical endpoints (Supplementary Data).

Between February 2012 and December 2013, 65 patients were screened and 38 patients were consented (Table 1). Thirty-four patients were evaluable for response and 35 for toxicity. Two patients signed consent but were never treated (financial burden for one and pneumonia for another). The median number of completed cycles was 8 (range, 0–20). Twenty-two patients came off study due to progressive tumor, 8 because of toxicity, 3 due to withdrawals, one due to disease-related adverse event, and one due to death.

Partial tumor responses were observed in 2 patients, including one with a CR of the measurable tracheal lesion without changes in nonmeasurable bone metastases (Table 2). The other experienced a 32% decrease in RECIST measurements at 2 months, primarily due to marked regression of the largest lung metastasis and also a dramatic decrease in cancer pain. However, although this response was apparent on the follow-up scan 4 months later, the appearance of a new ground glass area made this an unconfirmed response by standard RECIST criteria (unclear whether pneumonia vs. tumor per radiologist). Given the extent of the initial response and the relatively long interval between scans, we chose to list this case as a response but denote it as unconfirmed. Overall, 14 patients (41%) experienced any degree of tumor regression at any point (Fig. 1). In addition, dovitinib significantly suppressed the TGR (Fig. 2). Prior to treatment, the slope of the line representing TGR was 1.95, which dropped to 0.63 after four cycles of dovitinib (P < 0.001). In an unplanned analysis, we noted that beyond 4 months, the slope increased to 1.06 but remained significantly lower than the pretreatment level (P < 0.001). This suggested the emergence of partial drug resistance with an ongoing drug effect at least up until the time treatment was stopped, the consequence of the frequent dose reductions of dovitinib, and/or the influence of drug-independent events. Given the low rate of clinical responses, no robust association of TGR with outcome (OR, PFS, or OS) was observed in an exploratory analysis. It is noted that the two partial responders had relatively steep TGRs prior to treatment but so did some nonresponders. It is noteworthy that half of the patients (49%) on this trial had prior systemic therapies (Table 1) for ACC (possibly owing to the lack of effective therapies for ACC). Of the participants, 37% (13/35) had a prior systemic therapy in the 6 months prior to enrollment.

At the last follow-up, 15 patients have died and 20 patients remain alive. The median PFS for the entire group (Fig. 3) was 8.2 months [90% confidence interval (CI), 7.3–11.0] with a median OS of 20.6 months (90% CI, 17.6–unmet). Of 34 evaluable patients, all but one had stable or responsive disease at 2 months, whereas the clinical benefit rate (CR + PR + stable disease ≥4 months) was 70.6%.

In an exploratory study, 15 patients underwent PET imaging with ¹⁸FDG before and 12 days after starting treatment to determine whether FGFR inhibition alters tumor glucose metabolism and whether metabolic tumor response correlates with subsequent clinical outcome (27–29). Three patients (20%) had metabolic responses with no discernable changes in tumor size. Overall, the tumor SUV decreased in 8 patients and increased in 6. However, there was no apparent association between metabolic response and other study endpoints (Supplementary Table S4).

QOL as assessed by FACT-G questionnaires is summarized in Supplementary Fig. S2. In modeling overall QOL, there was a significant association between overall QOL and time to treatment failure, where higher scores of QOL were associated with a smaller instantaneous hazard of end of treatment (P = 0.073). Time is the only significant covariate in the model, where there is an average decrease in QOL of 1.2 points for each additional month in the study (P = 0.014). The subset of patients who experienced pain relief while on dovitinib generally had the least decline in FACT-G scores, but there was no correlation with objective tumor response or PFS. Notably, overall QOL scores trend lower over time, suggesting either morbidity from disease progression or cumulative toxicity from treatment or a combination.

Tumor samples from a subset of patients were available for biomarker and genetic testing. Tissue samples were frequently insufficient or had been depleted by prior pathologic studies during clinical analyses of this rare tumor type. Serial biopsies were not pursued for safety considerations; in many patients, this would require lung biopsies. Of the 15 tested tumors, 10 had MYB gene rearrangements as determined by the break-apart FISH assay. Despite the limited sample sizes, we found significant associations between MYB translocation and higher MYB expression (P = 0.021) as determined by IHC; however, there was no apparent association with FGFR1 phosphorylation (P = 0.99) or clinical endpoints. Nine tumor samples yielded DNA suitable for exome sequence analysis of the Foundation One gene panel. This revealed gene mutations previously reported to be altered in ACCs but no alterations in the FGF or FGFR genes (Supplementary Table S4).

Dovitinib was generally tolerable, although dose reductions were required in all but 2 patients and 8 stopped treatment due to toxicity, most often due to fatigue and anorexia. Grade 1 and 2 toxicities that were seen in more than 50% of patients included fatigue, nausea, diarrhea, anorexia, and acneiform rash (Supplementary Table S3). Hyperphosphatemia was not observed. Grade 3 or higher toxicity was observed in 63% of the patients, including 9 with hypertriglyceridemia (Table 3). One patient with locally recurrent stage III ACC of the parotid developed central necrosis of tumor after 2 months of therapy but died from pneumonia while on study.

This study revealed that dovitinib has demonstrable but limited antitumor activity in patients with recently progressed ACC. Although the ORR (6%) was not sufficient to reject the null hypothesis, dovitinib significantly suppressed the overall TGR as determined by change point analysis. Consistent with this, about 40% of the patients had some reduction in tumor volume, and 65% had stable disease for greater than 4 months. We also observed that 3 of 15 evaluable patients had early metabolic responses as determined by ¹⁸FDG PET imaging. Although a high proportion (43%) of our patients had liver involvement, an adverse prognostic factor, the median PFS (8.2 months) and OS (21 months) compared favorably with those reported for ACC patients treated with other kinase inhibitors. For instance, median PFS for patients treated with cetuximab, lapatinib, imatinib, or regorafenib in phase II studies is in the range of 2.5 to 6 months, whereas the values for those given sunitinib or sorafenib were 7.2 and 11.3 months, respectively (3, 4, 8, 12, 16). The prolonged PFS in the sorafenib study may reflect the inclusion of patients without evidence of recent tumor progression; also, despite the reported PFS, the OS (19.6 months) was similar to that seen in our dovitinib-treated patients (4). The antitumor effects of dovitinib were achieved at the expense of moderate toxicity, as dose reductions were required in almost all patients, and one in five stopped treatment due to adverse effects. That dovitinib has limited activity against ACC is supported by a recently reported trial in which only one of 32 treated ACC patients responded to this drug, although 69% did exhibit minor responses (30).

The results of our trial leave open the question of whether or not FGFR is an appropriate therapeutic target in ACC. We had postulated that FGFR signaling acted downstream of the frequently acquired MYB gene mutations, and inhibition of this signaling would produce at least a moderate objective tumor response rate. This was not observed; however, treatment was associated with a reduction in TGR and a favorable PFS. Because dovitinib is a multikinase inhibitor, we cannot determine whether the observed antitumor effects were due to inhibition of signaling by the FGFR or other target receptors, or a combinatorial effect. In this regard, sunitinib and sorafenib, which lack activity against FGFRs, nonetheless confer PFS or OS similar to that seen in our dovitinib-treated patients. Thus, the similarities in clinical outcome seen with these three kinase inhibitors may stem from suppression of shared targets, such as the VEGFRs.

One possible explanation for the low ORR of ACCs to dovitinib is failure of the drug to adequately suppress oncogenic FGFR signaling in tumor tissues. Although dovitinib has been shown to increase the blood levels of FGF23, a biomarker of systemic FGFR inhibition, the drug produces little or no hyperphosphatemia, which is a marker of more robust receptor inhibition (31–33). That dovitinib may not produce sufficient FGFR suppression in vivo is supported by phase II studies of patients with bladder cancer who were treated with dovitinib or the more potent and specific FGFR inhibitor BGJ398. Dovitinib produced only one response in 54 treated patients and none in 12 patients whose tumors harbored FGFR3 gene mutations (34). In contrast, BGJ398 caused hyperphosphatemia in up to one half of the treated patients and an ORR of 36% in those whose tumors contained FGFR gene alterations (31). Thus, it is possible that the more potent and specific FGFR inhibitors, such as BGJ398, may be more effective than dovitinib in the treatment of ACC.

Another explanation for the few objective responses in our study is that the FGFRs are not major drivers of tumorigenesis in the majority of ACCs, irrespective of MYB gene status. In the ancillary studies, we found that 10 of the 15 tumors that were studied contained detectable MYB gene rearrangements, and this correlated with higher MYB expression as determined by IHC. However, in this limited sampling, there was no apparent association between MYB mutation or expression and the level of FGFR1 phosphorylation (P = 0.99) or clinical outcome. This suggests that MYB translocation is not tightly associated with FGFR1 activation and therefore may not be a reliable marker of sensitivity to FGFR inhibitors. However, our observations do not exclude the possibility that the response rate to dovitinib would be much higher in the small subset of ACC patients with tumors that contain activating mutations in the FGFR or FGF genes (16, 19).

This study did confirm that change point analysis of TGR of ACCs is feasible and capable of detecting statistically significant reductions in the TGR following within 4 months of initiating drug treatment. However, additional studies will be needed to determine the degree of TGR suppression required to confer a true clinical benefit. Once established, the specified reduction in TGR could be used as an early endpoint in trials of ACC designed to identify agents likely to be clinically effective. In this regard, reductions in the TGR (measured by a distinct method) were linked to improved OS in patients with renal cancer treated with kinase inhibitors and with favorable PFS in patients entered into a series of phase I trials (35–37).

In summary, we found that dovitinib has limited but demonstrable antitumor activity against ACCs. However, it is not clear whether these limitations are related to inability of the drug to adequately suppress oncogenic FGFR signaling or the fact that FGFRs are not valid therapeutic targets in the majority of tumors. These issues can be clarified by additional preclinical studies to confirm that MYB gene translocation is associated with autocrine FGFR signaling. If so, this would justify a follow-up phase II trial of a potent FGFR-specific inhibitor along with companion genetic and pharmacodynamic studies to verify that most tumors contain MYB mutations and that FGFR signaling is effectively suppressed. In addition, a trial of FGFR inhibitors for the treatment of the 4% to 12% of ACCs with FGFR or FGF gene mutations is also appropriate. However, because of their rarity, these tumors would probably be best studied in the context of a “basket trial,” which enrolls patients with different types of tumors that contain similar FGFR gene alterations.

P.M. Dillon reports receiving other commercial research support from Novartis Pharmaceuticals. G.R. Petroni is a consultant/advisory board member for NeuroHabilitation Corp. C.A. Moskaluk reports receiving commercial research grants from Philips. C.Y. Thomas is a consultant/advisory board member for Novartis. No potential conflicts of interest were disclosed by the other authors.

Conception and design: P.M. Dillon, G.R. Petroni, C.A. Moskaluk, P.M. Fracasso, C.Y. Thomas

Development of methodology: P.M. Dillon, G.R. Petroni, P.M. Fracasso, C.Y. Thomas

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P.M. Dillon, C.A. Moskaluk, P.M. Fracasso, M.G. Douvas, S. Zaja-Milatovic, C.Y. Thomas

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P.M. Dillon, G.R. Petroni, B.J. Horton, C.A. Moskaluk, M.G. Douvas, N. Varhegyi, C.Y. Thomas

Writing, review, and/or revision of the manuscript: P.M. Dillon, G.R. Petroni, B.J. Horton, C.A. Moskaluk, P.M. Fracasso, C.Y. Thomas

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P.M. Dillon, C.A. Moskaluk, C.Y. Thomas

Study supervision: P.M. Dillon, P.M. Fracasso, C.Y. Thomas

The authors thank all of the patients and their families for their participation. We also thank the University of Virginia Cancer Center and the Adenoid Cystic Carcinoma Research Foundation.

This study (NCT01524692) was funded by Novartis Pharmaceuticals. Some biomarker studies were supported by the UVA Cancer Center, Charlottesville, VA (to P.M. Dillon and C.A. Moskaluk). Statistical analysis was done by the Biostatistics Shared Resource, University of Virginia Cancer Center, University of Virginia (P30 CA044579).

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.