Biomarkers and Bone Imaging Dynamics Associated with Clinical Outcomes of Oral Cabozantinib Therapy in Metastatic Castrate-Resistant Prostate Cancer

Prostate cancer is the most common cancer in males with an estimated 164,690 new cases in 2018 in the United States with an anticipated mortality of 29,430 (1). Although most cases are treated when localized, others present as disseminated disease or become metastatic after definitive treatment. In metastatic castrate-resistant prostate cancer (mCRPC), various agents such as sipuleucel T, abirateraone, enzalutamide, docetaxel, cabazitaxel, and radium-223 have demonstrated survival benefit (2, 3). Despite demonstrating impressive efficacy in early trials, cabozantinib encountered a rocky road during the development of an indication in metastatic prostate cancer. Initial phase I/II study of cabozantinib revealed tremendous promise with an unprecedented normalization of bone scans that had never been observed even with the most effective treatment to date such as androgen deprivation therapy (4, 5). In addition, these effects were seen in a refractory pretreated patient population, and measurable disease responses were noted. The phase II trial results with this agent in prostate cancer led to the conduct of 2 large registration trials. Unfortunately, the first trial of cabozantinib plus prednisone versus placebo plus prednisone showed no benefit in overall survival (OS), which was the primary endpoint (5). The second trial comparing cabozantinib and prednisone to mitoxantrone and prednisone with predefined pain palliation endpoint was halted early due to results demonstrating lack of benefit (6). These events led to further drug development of cabozantinib being put on hold in prostate cancer. The mechanisms underlying the remarkable bone scan responses associated with significant clinical palliation have not been studied in depth. In addition, the evaluation of biomarker or imaging changes in correlation with clinical outcomes was not conducted.

Cabozantinib is an inhibitor of tyrosine kinases, including MET, AXL, and VEGFR, that results in abrupt clinical changes in bone metabolism represented as an abrogation of ^99mTc-MDP uptake on bone scan. This is likely due to inhibition of osteoclast function and decrease in osteoblast activity. We hypothesized that the agent uniquely targets the cross-talk between C-MET and vascular endothelial growth factor receptor (VEGFR) axis and modulates bone turnover via downstream cathepsin K–driven pathways and activity of novel receptor tyrosine kinases, such as DDR-1 and DDR-2 (refs. 7, 8; Fig. 1). We conducted a pilot trial designed to study the pathophysiology and biomarker changes in bone metastases and correlate these with response and clinical outcome data in metastatic CRPC patients. The study also explored any mechanistic clues for a subset within mCRPC that maybe worthy of targeting with cabozantinib therapy, given the clinical efficacy observed and reported by multiple investigators globally. The MET receptor tyrosine kinase (RTK) for hepatocyte growth factor (HGF) has been implicated as a mediator in many important aspects of tumor pathobiology, including tumor survival, growth, angiogenesis, invasion, and dissemination. (9) Several tyrosine kinase inhibitors of MET have been reported to show antitumor activity in cell lines and animal models (10). The VEGFR2 (vascular endothelial growth factor receptor) is a central mediator of tumor angiogenesis, and several small-molecule and protein therapeutics targeting this receptor are in clinical development. In addition to their individual roles in tumor pathobiology, preclinical data suggest that Met and VEGFR2 play synergistic roles in promoting tumor angiogenesis and subsequent dissemination (9).

Typically, anti-VEGF therapies have not been effective in mCRPC. Randomized trials with bevacizumab or ramucirumab demonstrated lack of benefit when evaluated in combination with docetaxel. Compounds that simultaneously inhibit VEGF and MET RTKs may be more effective anticancer agents than agents that target each of these receptors individually (10). Cabozantinib is a potent RTK inhibitor that targets primarily MET and VEGFR2 and, by this mechanism, is likely to overcome resistance to anti-VEGF therapy. It has activity against other RTKs that have been implicated in tumor pathobiology, including KIT, FMS-like tyrosine kinase 3 (FLT3), and Tie-2. It is known to inhibit RET, an RTK known to be causative for malignancy, such as in hereditary medullary thyroid cancer (11, 12).

The conventional bone scan utilizes Technetium ^99mTc methylene diphosphonate (MDP) and is a most widely used standard-of-care method for evaluating skeletal metastases in prostate cancer. The ^99mTc-MDP accumulates in new (woven) bone and is an indicator of changes in bone metabolism especially associated with prostate cancer–induced osteoblastic response. However, ^99mTc-MDP scan findings are nonspecific and are indirect markers of response to treatment (13). Positron emission tomography with ¹⁸F-sodium fluoride (NaF-PET) scans have improved anatomic detail over ^99mTc-MDP scans, a higher accuracy in detecting metastases and potentially allow quantification of the extent of metastatic lesion (14). This imaging modality may be superior to ¹⁸F-fluorodeoxyglucose (FDG) PET for prostate cancer, since the bone metastases in prostate cancer are primary osteoblastic. Osteoblastic metastases tend to exhibit a high rate of fluoride incorporation (13) and may have low FDG uptake (14). Additionally, NaF-PET has improved sensitivity, so earlier detection of changes is feasible and PET imaging offers the potential for rigorous quantification. Unlike conventional bone scans that delineate the presence of a lesion (15), imaging techniques currently being evaluated in association with therapeutic trials in prostate cancer are designed to detect pharmacodynamic effects of novel agents. We utilized the ¹⁸F-fluoride PET in this study to trace the extent of absorption of fluoride ion by bone tissue and to attempt quantification of response in bone metastases.

PET obtained with the thymidine analogue ¹⁸F-(1-(2′-deoxy-2′-fluoro-β-D-arabinofuranosyl) thymine (FMAU PET) scan is used to image tumor metabolism and is based on incorporation of the tracer by mitochondrial thymidine kinase-2 (TK2; refs. 16, 17). The 3 modalities complement each other in distinguishing changes in lesions on imaging. Changes in cellular metabolism effected by therapy were hypothesized to have increased sensitivity in measurement of antitumor effects of cabozantinib than the more conventional approach of observing changes in size as seen on CT scans or just detecting areas of bone turnover or lack thereof, as seen with ^99mTc-MDP conventional bone scans. We have previously reported on the use of FMAU scans for detection of prostate cancer bone metastases from a study conducted at our institution, validating the use of this imaging modality for bone metastases (18).

The primary objective of the study was to evaluate the timing, physiology, and magnitude of changes in tumor imaging, and pharmacodynamics (PDs) markers with cabozantinib treatment in mCRPC. The secondary objectives were to evaluate the clinical safety, progression-free survival (PFS), and OS with this agent and to correlate clinical outcomes with imaging and PD changes observed. This was a single-arm, single-institution, pilot trial of cabozantinib administered at a starting dose of 60 mg orally daily in patients with mCRPC. The study was approved by the Wayne State University institutional review board and written informed consent was obtained from all patients before registration. The study was conducted in accordance with the ethical guidelines of the Declaration of Helsinki.

Eligible patients were 18 years or older and had histologically confirmed mCRPC and objective progression or rising PSA despite androgen deprivation therapy and antiandrogen withdrawal. Patients with rising PSA had to demonstrate a rising trend with 2 successive elevations at a minimum interval of 1 week. A minimum PSA of 5 ng/mL or new area of bony metastases on bone scan were required for patients with no measurable disease. No minimum PSA was required for patients with measurable disease. A maximum of 1 prior chemotherapy regimen for mCRPC was allowed. Any radiotherapy had to be completed at least 2 weeks prior to starting study therapy. All patients had to be documented to be castrate with a testosterone level ≤ 0.5 ng/mL. Luteinizing hormone releasing hormone agonist therapy was continued, if required to maintain castrate levels of testosterone. Patients had to be off antiandrogens for a minimum of 4 weeks for flutamide and 6 weeks for bicalutamide or nilutamide. Patients with ECOG performance status ≤2 and life expectancy of 12 weeks or more were eligible. Patients were required to have adequate bone marrow, liver, and renal function. Other key exclusion criteria included history of bowel perforation or fistula, uncontrolled brain or leptomeningeal metastases, uncontrolled hypertension or diabetes mellitus or history of congestive heart failure.

The study consisted of open-label daily, oral administration of cabozantinib at a starting dose of 60 mg to eligible patients. This was administered with a full glass of water (minimum of 8 oz/240 mL) after fasting (with exception of water) for a minimum of 2 hours before and at least 1 hour after ingestion. Subjects were advised to record dosing time and doses taken in a study drug dosing diary while on study treatment. The original schedule of assessments continued even if doses were withheld. The subjects were instructed to not make up any missed or vomited doses and to adhere to the planned dosing schedule. The study allowed a maximum of 2 dose reductions of cabozantinib to 40 mg and 20 mg orally daily, respectively. If grade 3 or 4 toxicities or grade 2 toxicity lasting for 7 days or greater was noted, then medication was held until the toxicity resolved to grade 1 or pretherapy baseline. Dose reduction could be considered when resuming therapy. If toxicity persisted after 2 dose reductions and optimal supportive care, then study therapy had to be discontinued.

This study utilized standard-of-care ^99mTc-MDP bone scan using a gamma camera, NaF-PET to measure effects of cabozantinib treatment on bone tissue, and FMAU PET to measure changes in tumor metabolism in response to therapy. Imaging was performed using a gamma camera and PET/CT scanners.

We proposed evaluating patients with NaF-PET pre- and post cabozantinib therapy to determine the optimal timing when bone scan normalization occurs. So approximately 5 patients were evaluated with NaF-PET imaging pretherapy and 2 to 3 weeks posttherapy. The optimal time point to perform imaging with FMAU-PET scans to evaluate for antitumor effect was determined to be at 2 weeks and subsequent patients had scans performed during that timeline. PET imaging was performed using a GE Discovery STE PET/CT system (GE Medical Systems), located at the PET Center, Children's Hospital of Michigan. Patients were positioned on their back on a PET/CT scanner in a high-sensitivity mode. Vital signs were monitored at the beginning and end of each scan. Patients were injected with FMAU PET using an intravenous catheter with doses standardized to body weight (mean, 360 MBq; range, 196–407 MBq) with a specific activity of at least 18,500 MBq/microM and a purity greater than 98%. Dynamic images were acquired at 6 to 11 minutes with 1 bed position over the area of interest of 2 frames: (1 × 5 minutes and 1 × 6 minutes). Thereafter, a whole-body image was obtained using a 2D/3D modality, of 3 bed positions. For NaF-PET scans, patients received an intravenous injection with a mean of 340 MBq (range, 259–407 MBq) and imaging began 45–60 minutes later. Patients were positioned with their arms down and scanned from the vertex to upper thigh and then repositioned to scan the patient's legs. Whole-body acquisition time consisted of 3 minutes per bed position. Reconstructed images were viewed and analyzed using Osirix Imaging Software. Tumor SUVmax values were obtained at baseline and follow-up scans by drawing a 1-cm diameter region of interest over 5 of the most active boney lesions on the NaF-PET scans. We selected no more than 2 active lesions per bone region, including the skull, thorax, spine, pelvis, and extremities. The same lesions were selected on the FMAU PET scans. A decrease in mean SUV by 20% was considered a PET response and was used as the threshold to detect changes.

Serum bone markers were assessed pre- and posttherapy. These included serum bone-specific alkaline phosphatase (BSAP) and N-terminal telopeptide of collagen type I (NTx). High levels of these markers (>146 u/L for BSAP and >100 nmol/mmol for NTx) have been reported to be significantly predictive of higher incidence of skeletal complications (relative risk of 3.32, P < 0.001), prostate cancer progression (RR 2.02, P < 0.001), and death (RR of 4.59, P < 0.001; ref. 19).

A number of other bone turnover markers such as osteocalcin, pyridinoline, and deoxypyridinoline have been implicated to be predictive of therapeutic response. A study evaluating the efficacy of matrix metalloproteinase inhibitors in prostate cancer reported that decline of the bone resorption markers, including NTx, procollagen I NH2-terminal propeptide, osteocalcin, and deoxypyridinoline, correlated with improved PFS and OS outcome (19–21). This led to the hypothesis that detectable changes in bone markers could act as surrogates of therapeutic effect in prostate cancer bone metastases.

We selected the serum NTx and BSAP and urine NTx as the bone turnover markers due to the validation of these markers in prior large studies utilizing zoledronate therapy (22). Decline in the levels was predictive of lower incidence of skeletal events as well as PFS and OS. Hence, the measurement of these markers (NTx and BSAP) pre- and posttherapy was correlated with PET scan findings and clinical outcomes.

Serum and urine (24-hour urine collection sample) N-telopeptide were assessed using the Vitros ECI Immunodiagnostic System competitive assay (Johnson & Johnson Ortho-Clinical Diagnostics). Serum BSAP levels were assessed using a chemical inhibition and differential inactivation assay.

The levels of bone resorption markers in serum were assayed at baseline and at 8, 15, and 28 days after treatment with cabozantinib. Tartrate-resistant acid phosphatase (TRAcP) levels were measured using Human TRAcP ELISA (RayBiotech). For the evaluation of osteocalcin levels, Quantikine Colorimetric Sandwich ELISA assays (R&D Systems) were used. All samples were assayed in triplicate according to the manufacturer's instructions, and cytokine levels were quantified by colorimetric detection at 450 nm against appropriate standards.

This was measured by the Cell Search method pretherapy, and 2 and 4 weeks after therapy and at progression.

Paraffin sections were deparraffinized in a xylene–ethanol series. Endogenous peroxides were removed by a methanol/1.2% hydrogen peroxide incubation at room temperature for 30 minutes. HIER antigen retrieval with a pH9 EDTA buffer and the BIOCARE Decloacking Chamber. A 40-minute blocking step with Super Block Blocking buffer (Thermo Scientific) was performed prior to adding the primary antibody. Met antibody from Abcam (ab51067) was used at a dilution of 1:100. Detection was obtained using HRP/DAB chromogen and counterstained with Mayer's Hematoxylin. Sections were dehydrated through a series of ethanol-to-xylene washes and cover slipped with Permount. The staining was evaluated, categorized as 0, 1+, 2+, and 3+ by a qualified pathologist who was blinded to clinical data, and reported.

The trial was a prospective pilot adaptive design study to obtain preliminary data. Each patient would undergo up to 4 PET scans, using different radiotracers. It was desired to estimate the mean SUV at any time point to approximately one-third of a standard deviation (SD) with 80% confidence. With 15 patients, the mean SUV could be estimated to within 0.347 SD units of the true mean with 80% confidence. These preliminary estimates would be of sufficient precision for use in designing a subsequent larger study. Because not all patients would undergo all PET scans, the mean SUV may be less precisely estimated at some time points.

The Prostate Cancer Clinical Trials Working Group (PCWG2) criteria were used to determine a response (23). For measurable disease response, RECIST criteria 1.1 were used (24). For the patients imaged on a common schedule, the continuous endpoints (e.g., SUV, bone scan measurements, and all continuously distributed correlatives) were summarized with standard descriptive statistics, separately at each measurement time point. We hypothesized that biomarker changes in bone metastases and imaging correlated with response and clinical outcome data in metastatic CRPC patients treated with cabozantinib. The categorical endpoints such as toxicities, the clinical response, and IHC expression levels were summarized via their frequency distribution, point estimate of the proportion, and the Wilson type 90% confidence interval (CI).

The distributions of percent change in SUV by FMAU PET and percent change in each bone marker were quite nonnormal, despite various transformations applied. Accordingly, the relationship between pre/posttherapy changes in imaging (percent change in SUV by FMAU PET and by NaF) and changes in each of the 3 bone marker levels (percent change from day 1 to week 4) were first assessed using the Spearman rank correlation coefficient (rho, and its 90% CI). Fisher Z-transformation of rho was required in order to calculate the confidence limits for the rank correlation coefficient. To explore the relationships of change in bone markers with change in imaging, we used a nonparametric regression approach. We fit a locally estimated scatterplot smoother (LOESS) curve using the LOESS procedure in SAS 9.4 software. A LOESS curve was fit to 2 selected bivariate relationships of change in bone marker and change in imaging. For exploratory analysis purposes, the selections were the only 2 relationships having rho > 0.20. The default smoothing parameter (percentage of the total observations used in each smoothing neighborhood) was used in the LOESS procedure. These nonparametric LOESS curves better described the nonlinear character of those 2 relationships.

The distribution of censored PFS was summarized via the Kaplan–Meier (K-M) survivorship estimate. Summary statistics (e.g., median, 6-month, and 12-month progression-free rates) were calculated from the K-M life table. Similar analyses will be performed for OS as well.

PFS was measured from treatment start date to the first date of documented progression, whether by PSA or by imaging, or death from any cause, whichever occurred first. Patients not experiencing progression were censored for PFS as of the date of their last PSA or imaging result. OS was measured from treatment start date to the date of death from any cause. Patients were censored for OS as of the last date on which they were confirmed to be still alive.

Twenty-six patients were consented, of which 20 were eligible and enrolled; 1 withdrew from the study and 5 were screen failures. The median age was 69 years (range, 56–76 years; Supplementary Table S1). Thirteen patients had bone pain at the time of enrollment. Sixteen patients discontinued therapy due to progression and 4 discontinued due to toxicities which consisted of hand–foot syndrome, fatigue, urinary infection, and elevated creatinine each in 1 patient. No unexpected toxicities or treatment-related prolonged morbidity or mortality were noted. Each cycle consisted of 28 days of therapy. The median number of cycles received was 4 (range, 1–17 cycles) and 6 (30%) patients received 8 or more cycles of therapy. Three patients did not require dose reductions and 11 and 6 required 1 and 2 dose reductions, respectively.

Response was assessed in all 20 patients based on PSA as well as measurable disease per RECIST1.1. Six of 9 patients with measurable disease showed tumor shrinkage per RECIST 1.1 criteria (median −20%, range, −10 to −38%). Eight of 20 patients demonstrated a PSA decline with a median of 21% (Fig. 2) with absolute change of 22.3 ng/mL (range, 6.4% to 70.8%, absolute decline range 0.9–282.9 ng/mL). Two patients had a 50% or greater reduction in PSA levels. The PSA decline did not correlate with PFS. Ten patients received therapy for ≥4 cycles and 6 patients continued on therapy for ≥8 cycles (maximum of 17 cycles). All patients were evaluable for PFS and OS. Median PFS was 4.1 months (90% CI, 2.3–5.3 months), and median OS was 11.2 months (90% CI, 15.0–29.7 months).

Nineteen of 20 patients demonstrated a decline in the SUV max and mean pre- and posttherapy. Seventeen patients had the FMAU PET scan pre- and posttherapy and 14 showed a SUV decline of ≥20%. Two patients showed a decline of <20% at 14.2% and 18.7%. Only 1 patient showed a 23% increase in SUV max. Maximum SUV decline was 63.7%. Examples are shown in Fig. 3. The NaF-PET scan was evaluated in 19 patients and 18 (1 patient showed a 5% decline) showed a decline in SUV max of >20%. Maximum change in uptake was a 85.4% decrease. The timeline of changes and decline in tracer activity on both imaging techniques was very rapid and seen within 1 to 2 weeks of therapy. Serum NTX showed minimal change from pretherapy to week 4 of therapy, median decline of 2.8% from pretherapy median levels of 12.5 to 11.4 at week 4 of therapy. Urine NTX revealed an appreciable decrease from median of 29 pretherapy to 15 posttherapy, median decrease of 41.2%. Unfortunately, no correlation was noted between the magnitude of decline in SUV max by imaging with clinical benefit. Eleven patients had CTC > 5 pretherapy, of which 5 patients converted to CTC< 5 after therapy. Table 1 summarizes the changes dichotomized by patients receiving fewer than 8 cycles, or 8 or more cycles of therapy.

As an exploratory analysis only, Spearman rank correlation coefficients (rho values and their CI) for all pairs of 5 variables (percent change in each of FMAU PET SUV, NaF SUV, BSAP, serum NTx, and urine NTx) are given in Supplementary Appendix Table S2. The 2 pairs of imaging change and bone marker change with the largest (and positive) correlation (rho > 0.20) were NaF SUV with BSAP (rho = 0.22), and NaF SUV with urine NTx (rho = 0.26). The nonparametric LOESS curve fit for percent change in NaF SUV as a function of percent change in BSAP is shown in Fig. 4A. There is a positive relationship overall, especially in the range of negative percent changes in BSAP. There, large decreases in BSAP tended to relate to large decreases in NaF SUV. A similar but weaker positive relationship was found in which large decreases in urine NTx tended to relate to only modest decreases in NaF SUV (Fig. 4B).

Tumor biopsies were conducted pretherapy and 2 weeks after therapy. C-MET testing was conducted by IHC on tumor biopsies (Fig. 5A and B). No consistent changes in c-MET expression were noted pretherapy and posttherapy, and the extent of CMET expression did not correlate with clinical benefit (Fig. 5C and D).

Radiologic response and progression represent key decision-making endpoints in oncology therapeutics. Frequently, decisions to continue or change therapy hinge on changes noted in imaging. In the case of bone metastases, the predominant site of spread in prostate cancer, this endpoint is flawed and leads to erroneous decisions. A prime example of this was noted within the imaging changes after cabozantinib administration. Conventional (Technetium) ^99mTc-MDP bone scans showed remarkable response in a majority of the patients and led to excitement in phase II trials, which was subsequently not matched by clinical efficacy seen in the randomized setting (25). The current study was designed to incorporate other imaging techniques such as NaF-PET and FMAU PET scans to evaluate correlation with clinical endpoints. The results reveal that these scanning methods were not predictive of efficacy. Novel imaging techniques, such as choline/acetate scans and fluciclovine scans based on amino acid radiotracer, have demonstrated exquisite sensitivity to detect recurrent disease at low PSA levels, and high positive predictive values (26). These scans are able to detect metastases in both bone and soft tissue; however, they have not yet been assessed for monitoring effects of systemic therapy (27). Bone turnover markers have been explored as potential predictors of response or clinical benefit from therapy. Unfortunately, the results of most studies have been disappointing, and these biomarkers are not utilized in routine clinical practice. Alkaline phosphatase changes have been reported to be predictive of response in radium-223 therapy (28). In fact, changes in alkaline phosphatase are likely to be better predictors of benefit than changes in PSA for radium-223 treatment. Cabozantinib had a major impact on the bone turnover in prostate cancer bone metastases in initial studies. It also demonstrated promising clinical response rates in measurable disease (5). Unfortunately, later studies showed that the remarkable efficacy noted did not lead to an OS impact. Randomized double-blind controlled trials revealed lack of OS benefit when compared with prednisone therapy alone. However, investigator-assessed PFS, which was an exploratory endpoint of the study, was statistically significantly improved by cabozantinib treatment (HR = 0.48; P < 0.001). The result was that the clinical efficacy of cabozantinib in randomized trials could not be proven.

The anti-VEGF therapies have demonstrated a consistent pattern of promising efficacy in single-arm trials, which subsequently did not translate into OS benefit in the randomized setting. Sunitinib demonstrated lack of OS benefit in a double-blind placebo-controlled trial with median OS of 13.1 months with sunitinib and 11.8 months with placebo (29). CALGB 90401 was a trial that compared the combination of bevacizumab and docetaxel to docetaxel alone, and although the combination showed improved response rates (43.4% vs. 35.4 %) and median PFS of 9.9 vs. 7.5 months (stratified log-rank P < 0.001), the OS (median of 22.6 months for the combination vs. 21.5 months, P = −0.18) was no different (30). A meta-analysis of 9 randomized control trials of docetaxel and antiangiogenic therapy as compared with docetaxel and prednisone confirmed lack of clinical benefit and possibly increased toxicity with combination therapy in mCRPC (31).

The current study was designed to detect the mechanism of the bone changes and to explore if bone biomarkers and imaging changes could select a patient population that was likely to benefit. The results of our study show that initial response and sustained clinical benefit for over 8 months was noted in 30% of the patients. The observed reduction in serum marker levels was likely due to the reduced activity of cathepsin K, a protease highly present in metastatic prostate cancer (32) and the only enzyme capable of completely degrading helical and nonhelical collagen I, the main component of bone matrix (33). We also suspect the involvement of DDR1 and DDR2, 2 novel RTKs that are ligands for bone matrix collagen (34, 35). Whether cabozantinib mediates its action on bone turnover via this axis needs further investigations.

Table 1 reveals that the serum and urine biomarkers tested, such as CTC, BSAP, and serum and urine Ntx, were not distinctly indicative of early prediction of clinical benefit with cabozantinib therapy. The MET overexpression and phosphorylated MET also did not reveal any clear trend of being predictive markers of efficacy. C-terminal MET protein expression was absent in hormone-naïve prostate cancer and, in contrast, was present in CRPC in 23% of palliative transurethral resection specimens and in 72% of bone metastases (36). This was also not related to MET polysomy or amplification. C-MET is phosphorylated after nuclear translocation, and the staining noted indicates the activity; however, the level of expression was not associated with clinical benefit (36). On imaging, the decline in SUV was very rapid and occurred within a short period of time, in less than 1 week. This was in concordance with that observed in preclinical studies, but unfortunately no differential emerged to predict clinical outcomes. MET overexpression on IHC staining was also not predictive of clinical benefit. Thirty percent of patients had a durable benefit and continued on therapy for longer than 32 weeks. The imaging changes were seen in a majority of the treated patients, regardless of clinical benefit with cabozantinib.

The cabozantinib experience in mCRPC highlights the major challenge of using imaging response as a surrogate for clinical outcomes in this disease. Bone as a site of metastases alone comprises about 90% of the patients with CRPC. The conventional bone scan is severely limited in response determination, as the extent of tracer uptake and changes thereof does not correlate with clinical response. The MDP bone scan also cannot be utilized for measurement of lesions. In addition, due to the bone targeted, osteoclast inhibitory activity of cabozantinib, bone marker, and bone scan changes were misleading and not useful to predict clinical outcomes. Our study depicts that even novel imaging methods such as NaF-PET and FMAU PET scans were unsuccessful in detecting responses that would predict clinical benefit with this agent. NaF-PET is thought to detect bone metabolism and formation and may also bind to calcium phosphate, and hence not likely to provide a clear measure of treatment response (37). FMAU PET was developed with the intent to image tumor proliferation, but may be trapped in the cell by mitochondria by TK2, limiting the ability to assess changes in tumor growth (18). With the advent of genomic testing, next-generation sequencing–based results should be investigated as guides to therapeutic decisions.

Cabozantinib-based combinations are being explored in mCRPC, such as a clinical trial conducted by the NCI with nivolumab and an ongoing trial in combination with atezolizumab (NCT03170960). Synergy has been reported with multiple agents, such as immune-checkpoint inhibitors, and with radiotherapy. The phenotype of mCRPC is likely to change dramatically over the next decade. The advent of early indication and utilization of chemotherapy and abiraterone in the metastatic hormone-naïve setting and use of agents such as enzalutamide and apalutamide in nonmetastatic disease will alter the configuration of the mCRPC state. Incidence of neuroendocrine features within mCRPC is likely to increase. Cabozantinib has demonstrated efficacy in neuroendocrine tumors and is worthy of evaluation in prostate cancers that manifest neuroendocrine features (38). Application of next-generation sequencing will provide future clues in predicting clinical benefit with cabozantinib. With review of specific activity of cabozantinib in medullary thyroid cancer, it can be hypothesized that the RET gene mutations in prostate cancers could possibly predict for response (39). Future investigations of cabozantinib in mCRPC should focus on genomic markers that are representative of MET upregulation and RET mutations.

In conclusion, cabozantinib represents a unique mechanism of action that is distinct from currently approved therapies in mCRPC and continues to be worthy of deeper investigation. It holds potential in the treatment of patients with refractory mCRPC. The imaging changes occurred indiscriminately and were unable to indicate clinical benefit with the agent.

U. Vaishampayan reports receiving commercial research grants from and is a consultant/advisory board member for Exelixis. E.I. Heath reports receiving speakers bureau honoraria from Sanofi. A.F. Shields reports receiving commercial research grants from Exelixis. No potential conflicts of interest were disclosed by the other authors.

Conception and design: U.N. Vaishampayan, I. Podgorski, L.K. Heilbrun, E.I. Heath, A.F. Shields

Development of methodology: U.N. Vaishampayan, I. Podgorski, J. Boerner, K. Stark, E.I. Heath, A.F. Shields

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): U.N. Vaishampayan, J.M. Lawhorn-Crews, J. Boerner, E.I. Heath, A.F. Shields

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): U.N. Vaishampayan, L.K. Heilbrun, J.M. Lawhorn-Crews, D.W. Smith, E.I. Heath, A.F. Shields

Writing, review, and/or revision of the manuscript: U.N. Vaishampayan, I. Podgorski, L.K. Heilbrun, J.M. Lawhorn-Crews, K.C. Dobson, E.I. Heath, J.A. Fontana, A.F. Shields

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): U.N. Vaishampayan, J.M. Lawhorn-Crews, K.C. Dobson, J. Boerner

Study supervision: U.N. Vaishampayan, A.F. Shields

This study was partially supported by Department of Defense National Oncogenomics and Molecular Imaging Center grant W81XWH-11-1-0050, Exelixis Inc., and the NIH Cancer Center Support grant CA-22453.

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.