Approximately 70% of men with advanced prostate cancer harbor osteoblastic bone metastases [1
]. Imaging of bone metastases typically relies on bone scintigraphy and anatomic modalities such as CT and MRI. However, these methods measure qualitative changes in bone turnover (bone scan) or bone structure (MRI, CT) but not direct metastatic tumor cell activity. Clinically meaningful prostate cancer treatment response has been difficult to define quantitatively, as there is no uniformly accepted surrogate marker that correlates with long-term outcomes to optimally guide patient management and new drug development.
The use of positron emission tomography (PET) to monitor response to therapy in prostate cancer is inherently quantitative. PET can measure in vivo tumor and normal tissue biology using tracers to map many metabolic pathways, including bone osteoblastic metabolism using [18
F]-fluoride (NaF) PET [2
]. NaF PET offers a quantitative measure of osteoblastic bone formation and remodeling, and is appropriate for imaging the blastic lesions observed in prostate cancer [4
]. Additionally, when compared to standard 99M
Tc-based bone scintigraphy, NaF PET offers improved sensitivity of detection and when combined with CT, specificity is also improved [5
ACRIN 6687 was a prospective, multi-center phase 2 trial that used NaF PET to probe the response of dasatinib (SPRYCEL®
; Bristol-Myers Squibb) treatment, a SRC kinase inhibitor that decreases bone turnover, in men with metastatic castration-resistant prostate cancer (mCRPC) [8
]. The trial was designed to evaluate differential response of normal and tumor bone to dasatinib treatment using NaF PET using a protocol that began with dynamic single field-of-view (FOV) imaging and then was followed by static whole-body (WB) scans with multiple FOVs. Previous kinetic modelling results from single FOV dynamic imaging found significant differences in changes of the PET kinetic parameters from tumor bone compared to normal bone in response to dasatinib treatment. Changes in the 30–60 min summed SUV metrics from the dynamic acquisition had a modest association (p
= 0.056, n
= 12 patients) with progression-free survival (PFS), where progression was determined by the Prostate Cancer Working Group 2 (PCWG2) [9
Although the initial results for the ACRIN 6687 trial were intriguing, we recognize the potential limitations of dynamic single FOV analyses for general use and widespread adoption. Specifically, although dynamic studies may offer breadth of analysis, the level of complexity and lack of standardization are not practical for broad utilization. In the initial set of analyses using the 30–60 min SUV images, changes in the average SUVmax
for up to 5 tumors (SUVmaxavg
) in a patient not only had significant differential changes to dasatinib therapy in tumor vs. normal bone, but those changes had marginal association with progression free survival (PFS); these were features not displayed by dynamic Ki (metabolic flux) or K1
(tracer transport) kinetic parameters. This lends further credence to the concept of simplifying the NaF PET image analysis with SUV only. Additionally, the previously reported limited FOV may have omitted important information from metastatic lesions outside of the single FOV. As part of a post-hoc analysis not proposed in the original ACRIN 6687 clinical trial, we sought to determine if important information obtained from outside of the dynamic FOV could offer additional clinical and prognostic information, comparable and/or incremental to earlier published dynamic data. Previous reports using WB fluoride analysis also showed a relationship of SUV measures to PFS for patients that received either a docetaxel-based chemotherapy regimen or an androgen receptor pathway inhibitor [10
]. Here we examine SUV analysis results from multi-FOV WB static NaF PET imaging scans, acquired after a one-hour dynamic scan, in mCRPC patients recruited to ACRIN 6687 at baseline and after receiving 12 weeks of dasatinib treatment. Statistical analysis of the clinical and PET imaging data was undertaken in order to identify potentially interesting associations between various biomarkers (PET and blood borne) and patient outcomes. As is the nature of secondary investigations, the reported data analysis and relationships cannot be interpreted in the same way that the analysis for the primary hypothesis of the underlying clinical trial that has been reported [8
Similar to the results in our previous report of ACRIN 6687 evaluating a limited dynamic FOV, NaF PET WB uptake also reveals the distinct patterns of pharmacodynamic changes in bone mCRPC from normal bone in response to therapy with dasatinib, as displayed in Figure 2
. There appears to be a differential effect of dasatinib on normal compared to tumor bone in men with mCRPC, as measured by fluoride uptake and fluoride bone incorporation.
The previous ACRIN 6687 report [8
] showed that SUVmaxavg
from a single FOV NaF image summed exactly from 30–60 min had a large decrease in bone mCRPC uptake in response to treatment with dasatinib, and that a decrease in SUVmaxavg
marginally correlated with shorter PFS (p
= 0.056), indicating that patients with a lower decline in SUVmaxavg
had longer PFS. In the current WB lesion-level SUV analysis, baseline or changes in uptake measures collected later, on average WB imaging starting approximately 75 min after injection (range 53 to 95 min), failed to find significance with PFS or OS in univariate analysis. The later WB scan acquired with a mid-scan average of 90 min after injection (range 65 to 110 min) might be different from the single FOV dynamic scan collected precisely at mid-scan 45 min (30 to 60 min SUV image) due to tracer clearance that is independent of the disease, fewer counts with increased noise and the large variability of uptake time between patient WB scans, that all have the effect of increasing variability. The wide range in the time of WB image acquisition from the injection time in this multi-center trial can increase variability in SUV measurements by as much as 25% for 15 min deviations [22
] and may significantly affect the correlation of WB NaF measures to PFS where uptake times differ by more than 40 min.
The assessment of up to the 5 hottest tumors with a threshold SUV, is similar to prior methods, but may not be as useful as the selection of tumors and imaging FOV by local clinicians that utilized information based on their clinical impression of the patients in the ACRIN 6687 primary aim report [8
]. Averaging the SUVpeak
over 5 tumors may capture the intensity, but not the spatial distribution of a tumor and [10
] therefore may be unable to determine total tumor burden, as the QTBI analysis offers. Using QTBI analysis, Harmon et al. [10
] have found that total tumor burden determining a SUVtotal
metric via bone segmentation followed by thresholding the NaF SUV at 15g/mL has been valuable in assessing response in mCRPC patients using an effective therapy, such as androgen receptor pathway inhibitors or a docetaxel-based chemotherapy regimen. Patient-level WB assessment using QTBI software for the patients presented here did show that large baseline total tumor burden (qSUVtotal
) and tumor volume fraction (qVF) were significantly associated with shorter PFS in univariate analysis, suggesting that a large, intense tumor burden at baseline indicates poor clinical outcome. However, in univariate analysis the change in patient-level PET parameters from QTBI analysis failed to show a relationship to PFS, and no patient-level parameter showed association with OS.
The inability to observe a definitive relationship between changes in NaF PET uptake and PFS or OS may also be because the effect of dasatinib in mCRPC patients is marginal. Dasatinib has not been successful in demonstrating overall survival benefit in phase 3 trials of men with mCRPC [24
]. Although the effects of dasatinib on bone have been clearly documented, it does not appear to offer significant anti-tumor efficacy [25
]. The lack of association of changes in PET parameters to PFS or OS may be that the disease burden was so high in these mCRPC patients, that any response was buried in either PET measurement variability or dasatinib is an ineffective antineoplastic treatment against mCRPC.
However, a multiple variable statistical model that has covariates of age, a clinical biomarker (baseline ln(BAP)) and an NaF PET uptake measure showed that lesion-level baseline SUVmaxavg
, baseline SUVpeakavg
and patient-level baseline qSUVpeak
were all significantly associated with longer PFS in this small cohort (Table 4
). No PET parameter used in multivariate modeling analysis showed significant association with OS. The major multivariate model driving component is ln(BAP), which along with age and measures of NaF uptake aids in optimizing the estimates of progression. BAP and NaF uptake are expected to be closely related, as bone turnover (BAP) goes hand-in-hand with new bone formation and matrix mineralization (fluoride uptake on NaF PET). High baseline BAP and high NaF uptake might indicate a more favorable blastic phenotype and longer progression, while baseline BAP and lower NaF might indicate a more lytic phenotype and more aggressive clinical behavior.
The statistical results for the multivariate analysis might be affected by the large variation in image acquisition times between patients (see Supplementary Materials Figure S2
), which can increase variability by as much as 75% for over 40 min deviation in uptake time between patient scans [22
]. Outcomes using NaF PET have been different when more efficacious agents, with proven survival benefit, such as androgen axis inhibiting therapeutics or docetaxel chemotherapy have been used. In prior published studies with a larger cohort of patients (n
= 56), mid-treatment findings with NaF imaging alone have association with PFS [10
]. This suggests that NaF PET imaging has potential for assessment of treatment efficacy of some therapies in men with mCRPC.
Interestingly, we observed a negative correlation between a decreasing change in lesion-level SUV parameters (ΔSUVmaxavg
, ΔIndex SUVmax
) and an increase in bone alkaline phosphatase (ΔBAP). This relationship was noted in the initial report on the ACRIN 6687 trial that patients with the largest decrease in PET uptake parameters had worse outcome than those that stayed the same or increased [8
]. An increase in BAP levels may be due to dasatinib treatment, which has been shown previously to promote osteoblast differentiation [27
] and mineralization that could lead to a relative activation and a transient increase in BAP levels indicative of a healing or reparative response [28
]. Increased osteoblastic activity would also be expected to lead to a relative increase of NaF uptake. We did not follow these patients after completion of dasatinib treatment with repeat measurements of BAP, thus it is speculative to associate a decrease in BAP levels in this small cohort of patients with better outcome; however, this finding indicates some mechanistic consistency between prior findings based on dynamic imaging and the current WB analysis. Change in uNTX and PSA had no correlation with changes by NaF PET.
Given the very limited capacity of the dataset (n = 17 at baseline, n = 14 with an additional scan at 12 weeks into therapy) and the many measurements carried out, there is no real scope to carry out any type of internal cross-validation. The bootstrapping approach used in evaluating the relationship between PET variables and outcomes (PFS and OS) provides more defensible estimates of statistical significance of the reported effects and provides some measure of adjustment for the limited sample size. Nevertheless, our exploratory analysis is mainly offered to provide some guidance on what relationships may be worth future investigation via a prospective clinical trial. The most glaring limitation of this study, however, was the small number of evaluable patients recruited and an even smaller subset that completed the second PET scan during dasatinib treatment, limiting statistical power for prediction of PFS and OS.