Treatment Response Evaluation Using ¹⁸F-FDOPA PET in Patients with Recurrent Malignant Glioma on Bevacizumab Therapy

This article is featured in Highlights of This Issue, p. 3359

High-grade gliomas carry a very poor prognosis (1). Patients when initially diagnosed are treated typically by the combination of surgery, chemo, and radiation therapy. Inevitably, tumors will recur. The disease status in patients with high-grade gliomas is routinely evaluated clinically and radiographically by contrast-enhanced MRI. However, because of the highly heterogeneous nature of these tumors and the complexity of changes induced by the treatment, the structural information derived from these images does not always adequately reflect both the extent of the actual lesion as well as its malignant potential (2, 3). Biomarkers that are surrogates of proliferative and malignant properties of the tumor and thus can predict treatment responses early and reliably, are under active investigation (4, 5).

Amino acid analogues (6–14) have been studied for brain tumor imaging. ¹⁸F-labeled amino acid analogues have the advantage of easy clinical application because of the favorable half-life (110 minutes) of the ¹⁸F-positron emission tomography (PET) tracer (7). Preliminary studies with ¹⁸F-fluoroethyl-l-tyrosine PET in 21 patients with recurrent high-grade glioma on antiangiogenic treatment have shown promising results (15, 16). The amino-acid-analogue L-3,4-dihydroxy-6-¹⁸F-fluorophenyl-alanine (¹⁸F-FDOPA) has recently also been under investigation in the evaluation of brain tumors (17–24). ¹⁸F-FDOPA uptake in brain tumors is increased compared with normal brain tissue because of the upregulation of amino acid transport across the blood–brain barrier (BBB) by the neutral amino acid transporter. It has been shown previously that ¹⁸F-FDOPA provides excellent visualization of high-grade as well as low-grade brain tumors (17–20). Furthermore, its uptake in tumor tissue correlates with tumor grade as well as Ki-67 proliferation index in newly diagnosed gliomas, allowing for a quantitative assessment of malignancy (23).

This study examines the tumor staging potential and predictive power of ¹⁸F-FDOPA PET in the assessment of treatment response for patients with recurrent high-grade glioma undergoing bevacizumab therapy. First, tumor uptake of ¹⁸F-FDOPA as well as tumor volumes was obtained. Second, the predictive power of ¹⁸F-FDOPA for progression-free survival (PFS) and overall survival (OS) in patients with recurrent high-grade gliomas was assessed. Finally, the predictive power of ¹⁸F-FDOPA was compared with MRI and other survival predictors.

Thirty patients with recurrent high-grade gliomas were enrolled prospectively in this study (Table 1). There were 18 males and 12 females with median age of 57.5 (range: 26–78) years. All patients met the following inclusion criteria: they had a histologically confirmed diagnosis of glioblastoma (GBM) or anaplastic astrocytoma (GBM: N = 24; anaplastic astrocytoma: N = 6) and had previously undergone surgical resection and chemoradiation therapy. All patients had MRI-confirmed recurrent disease by the time bevacizumab treatment was started. Further criteria included a Karnofsky performance score (KPS) ≥70, adequate hematologic values, and sufficient hepatic and renal function. Patients were excluded if there was a bleeding disorder, a recent history of intracranial bleeding or thromboembolism.

All patients gave written consent to participate in this study, which had been approved by the University of California, Los Angeles, Office for Protection of Research Subjects.

All Patients were treated with bevacizumab and irinotecan except for 3 patients who were treated with bevacizumab alone (patient no. 23, 26, and 28; Table 2). Co-administration of corticosteroids was closely monitored. Although 12 patients did not require corticosteroids, 9 patients were maintained on stable or tapering doses of dexamethasone, and 9 patients needed a dose increase after the baseline MRI and PET studies were obtained.

Patient's disease status was evaluated and monitored using gadolinium enhanced as well as nonenhanced MRI obtained within 1 week before and at approximately 6-week intervals after starting bevacizumab therapy. Patients were followed until death with no subjects lost to follow up. The OS was defined as the interval between treatment initiation and death whereas PFS was defined as the interval between treatment initiation and radiographic and/or clinical progression.

PET imaging was performed on a dedicated PET system (ECAT HR or HR⁺; refs. 25 and 26). Patients were asked to fast for at least 4 hours before image acquisition. ¹⁸F-FDOPA was synthesized according to a previously reported procedure (27, 28) and was injected intravenously at a mean dose of 1.89 ± 0.37 MBq/kg. Data were acquired in the 3-dimensional mode. The emission scan was started 10 minutes after tracer injection. To correct for photon attenuation, 5-minute transmission scans were acquired after the emission scans for all patients. Images were acquired for 30 minutes in the 3-dimensional mode. Image data acquired between 10 and 30 minutes were summed to obtain a 20-minute static image. PET images were reconstructed using iterative techniques with ordered-subset expectation maximization consisting of 6 iterations with 8 subsets (29, 30). A Gaussian filter with a full width at half maximum of 4 mm was applied.

Images were first inspected visually with the axial PET image slice displaying the maximum tumor ¹⁸F-FDOPA uptake selected. The radiotracer concentration in the regions of interest (ROI) was normalized to the injected dose per kilogram of patient's body weight to derive the standardized uptake values (SUV).

Tumor ROIs were defined on summed images in 2 ways. First, a standardized 10 mm circular region was placed over the area with the maximum activity. This was used to derive SUV_max and SUV_peak. Then, metabolic tumor volume (MTV) was obtained by including all voxels that fall within an SUV threshold determined by the mean SUV of the contralateral striatum (19). In cases where the tumor involved the striatum bilaterally, a threshold of 1.5 mean SUV of normal hemispheric background was used (19). Mean ¹⁸F-FDOPA uptake within this volume (SUV_mean) was determined. Absolute MTVs and SUVs as well as their changes were correlated with clinical outcome.

Data were collected on a 1.5T MRI system (General Electric Medical Systems) using pulse sequences supplied by the scanner manufacturer. Standard anatomical MRI sequences included axial proton density, T1, and T2 weighted fast spin-echo images, along with fluid-attenuated inversion recovery (FLAIR) images, all obtained with a 5-mm slice thickness with 1-mm interslice distance, 2 excitations, matrix size of 256 × 256, and a field-of-view of 24 cm. In addition, gadopentetate dimeglumine enhanced (Magnevist; Berlex) axial and coronal T1 weighted images were acquired after contrast injection.

Regions of FLAIR abnormality were chosen based on Response Assessment in Neuro-Oncology (RANO) recommendations (31). The regions of postcontrast T1-weighted image (T1+C) hyperintensity were defined on postcontrast T1-weighted images, excluding any T1 shortening from blood products on precontrast T1-weighted images as well as cystic and surgical resection cavities. The volumes of FLAIR and T1+C were calculated using a semiautomated procedure as described previously (32).

MRI-based response at 6 weeks was defined according to the RANO criteria. However, the 4-week sustained response requirement for complete and partial response was not considered because the goal of the study was to compare the predictive values of MRI and PET at 6 weeks after starting treatment (31). Progressive disease was defined as more than a 25% increase in the sum of the products of perpendicular diameters, a significant increase in nonenhancing tumor, or neurologic decline. Complete response was defined as neither enhancing- nor nonenhancing tumor, without steroid use. Partial response was defined as more than a 50% decrease in the sum of the products of perpendicular diameters on stable steroids, without new lesions. Nonenhancing tumor was identified by mass effect and/or architectural distortion, including blurring of the gray-white boundary. All scans that did not qualify above were considered stable disease. MRI assessment was performed by a board certified neuroradiologist (W.B. Pope).

In addition to assessment using the standard MRI response criteria, tumor volumes by MRI at baseline and follow up scans were also evaluated for response predication.

¹⁸F-FDOPA PET uptake threshold values for treatment response were established by receiver-operating characteristic (ROC) curve analysis. Threshold values derived from this method were scrutinized by leave-one-out cross-validation analysis to assess how robust and sensitive these cut-points were to individual values. The differences between groups of patients were established by the Student t test. Kaplan Meier curves were subsequently generated to obtain survival estimates (33).

Univariate and multivariate Cox proportional hazards models were constructed to test the relationship between ¹⁸F-FDOPA metabolic response and other predictors with regard to survival (OS and PFS). The forward stepwise selection technique was utilized to reduce the multivariate models (P < 0.10). Harrell C-index was used to compare the strength of predictive performance for each model. Furthermore, a permutation simulation study was performed to evaluate the effect of cut-point selection bias on the strength of the association between FDOPA and survival. Survival outcomes were randomly permuted and we recomputed the optimal cut-point for F-FDOPA for prediction of survival. For each of these cut-points, we computed the C-statistic. This was done 10,000 times to form an empirical distribution for the optimal C-statistic. The P value for the observed C-statistic was computed based on the proportion of the simulated C-statistics that were larger than the observed C-statistic. Statistical analyses were performed in SAS (version 9.2; SAS Institute) and R (version 2.14.2 www.rproject.org).

Thirty patients were enrolled for the ¹⁸F-FDOPA PET study (Table 1). All 30 patients completed the baseline ¹⁸F-FDOPA PET scan whereas 28 patients completed the second scan 2 weeks later and 24 patients were able to complete the third ¹⁸F-FDOPA PET scan 6 weeks after starting treatment.

Tumor ¹⁸F-FDOPA SUVs for all scans were obtained. Tumor SUV_max on baseline scans varied between 1.32 and 4.14. Following therapy, no significant changes of SUVs were seen (Fig. 1; SUV_max 2 weeks: −4.3% ± 15%; 6 weeks: −2.7% ± 19%; 2–6 weeks: +3.5% ± 22%).

MTVs, as delineated by ¹⁸F-FDOPA uptake, were determined and are listed in Table 2. Patients' median MTVs were 24.4 (range 1.6–272.3 mL) at baseline, 11.3 (range 1.0–181.0 mL) at 2 weeks, and 7.6 (range 0.83–115.5) at 6 weeks (Table 2). Significant changes of ¹⁸F-FDOPA-PET MTV were found (mean values: −40% ± 27% at 2 weeks and −47% ± 42% at 6 weeks (Fig. 1).

¹⁸F-FDOPA uptake values (at baseline, 2 weeks, or 6 weeks) both as SUV_max as well as SUV_peak were initially analyzed but were found not to be predictive of survival. Changes of SUV over the course of therapy were not indicative of treatment success, either.

However, ¹⁸F-FDOPA MTV changes were predictive of treatment response. Using ROC analysis, threshold values for MTV changes between baseline and 2 weeks, and baseline and 6 weeks, were obtained. Subsequently, Kaplan–Meier analysis was performed to assess their predictive power for PFS and OS.

When comparing ¹⁸F-FDOPA MTV changes from baseline to 2 weeks using the ROC-derived threshold value of ≥35% reduction, 16 responders and 12 nonresponders were identified with ¹⁸F-FDOPA MTV changes of −58% ± 16% for responders and −15% ± 17% for nonresponders, respectively (P < 0.001). The mean OS was 13.7 months for responders versus 7.0 months for nonresponders (P = 0.02). At 6 weeks, 17 responders and 7 nonresponders were identified with ¹⁸F-FDOPA MTV changes of −67% ± 18% for responders and +0.4% ± 48% nonresponders, respectively (P = 0.01). Mean OS was 14.1 months for responders and 7.6 months for nonresponders (P = 0.02). ¹⁸F-FDOPA MTV change at 6 weeks is a stronger predictor of OS (P < 0.001) than MTV change at 2 weeks (P = 0.001), whereas ¹⁸F-FDOPA MTV change at 2 weeks (P < 0.001) is a stronger predictor for progression free survival than ¹⁸F-FDOPA MTV change at 6 weeks (P = 0.003).

Next, absolute MTVs at baseline and follow up scans were assessed for their predictive performance. As expected, baseline MTVs do not correlate with survival (PFS and OS). Interestingly, MTVs at follow up scans do correlate with survival and ROC analysis identifies an optimal threshold tumor volume of 18 mL. This surprising result was scrutinized by performing a leave-one-out cross-validation analysis to assess how robust and sensitive this cut point was to individual values. A total of 28 of 30 cutpoint result thus derived generate a value of 18 mL, with the remaining 2 being 16.55 and 18.89 mL. When applying the optimal threshold value ¹⁸F-FDOPA MTV ≤ 18 mL at 2 weeks, 17 responders and 11 nonresponders are identified. Examples of individual responses are shown in Fig. 2. The median survival for the responders was 12.1 months versus 3.5 months for nonresponders, a 3.5-fold survival advantage (P < 0.001). ¹⁸F-FDOPA MTV at 2 weeks is the most significant predictor of OS as well as PFS by Kaplan–Meier analysis (P < 0.001 and 0.001, respectively; Fig. 3). ¹⁸F-FDOPA MTV at 6 weeks is also predictive of OS and PFS (P = 0.03 and 0.02, respectively).

Response by MRI was evaluated at approximately 6 weeks (5.9 ± 2.7) after starting treatment based on RANO criteria and was available for 29 patients. By MRI criteria, 22 patients were classified as responders (76%), these include 2 complete response, 7 partial response, and 13 patients with stable disease. The remaining 7 patients (24%) had progressive disease and were considered nonresponders. MRI response was predictive of OS as well as PFS (P = 0.01 and P < 0.001, respectively). Median OS for responders based on MRI criteria was 1.5 times longer than for nonresponders (11.4 months vs 7.7 months, P = 0.03). Comparing responders by PET and MRI criteria, 8 discrepant cases were identified. Of these, 6 patients showed a response by MRI (5 stable diseases, 1 partial response), but no treatment response by ¹⁸F-FDOPA. As these patients had a median survival of only 3.4 months, ¹⁸F-FDOPA predicted treatment failure earlier than MRI. A median time benefit of 7.2 weeks (range 2–20 weeks) for earlier detection of treatment failure with ¹⁸F-FDOPA was demonstrated. The remaining 2 discrepant cases were classified as nonresponders (progressive disease) by MRI but showed response with ¹⁸F-FDOPA. These 2 patients lived for 9.5 and 12 months.

Contrary to the PET study, MRI-derived absolute tumor volumes did not show survival correlation.

Multiple clinical variables were tested by univariate and multivariate analyses (Tables 3 and 4). Measurements of MTV at baseline, 2 weeks, and 6 weeks in addition to changes between these time points were modeled. Baseline characteristics such as age, number of prior recurrences, and RANO response at 6 weeks were considered in the multivariate models. Modeling all MTV measurement combinations at each time point would induce a survival bias because those who died after 2 weeks would not have had 6-week measurements (and thus be excluded from the model). Therefore, models were run with baseline data, data up through 2 weeks, and data up through 6 weeks separately (Table 4). By univariate analysis, OS was better if patients had fewer recurrences. Patients' age, baseline KPS, dexamethasone treatment, did not predict survival (Table 3). ¹⁸F-FDOPA MTVs at 2 weeks and 6 weeks, as well as their changes from baseline to 2 weeks and baseline to 6 weeks were all predictive of OS (P < 0.001, P = 0.04, P = 0.01, P = 0.01; Table 3). However, ¹⁸F-FDOPA absolute MTV at 2 weeks and ¹⁸F-FDOPA MTV changes at 2 or at 6 weeks provided the highest HRs (HR, 9.05, 2.94, and 4.02, respectively). Response by MRI was also predictive of survival (P = 0.016).

By multivariate analysis, ¹⁸F-FDOPA MTV at 2 weeks (P < 0.05; HR, 7.79) and MTV changes at 6 weeks (P < 0.05; HR, 4.09) were the most significant predictors of OS. Interestingly, no apparent added value of 6-week ¹⁸F-FDOPA MTV for survival prediction was seen with HR of 10.71 (P < 0.01) for 2-week ¹⁸F-FDOPA MTV response and HR of 4.09 (P < 0.05) for 6-week ¹⁸F-FDOPA MTV response (Table 4).

Similarly, longer PFS was predicted if patients had fewer prior recurrences. ¹⁸F-FDOPA MTV at 2 weeks, 6 weeks, and their changes from baseline to 2 weeks and 6 weeks, as well as MRI response were all predictive of PFS (P = 0.004, P = 0.02, P < 0.001, P = 0.002, P = 0.01). By multivariate analysis, ¹⁸F-FDOPA MTV change at 2 weeks (P < 0.01; HR, 4.38) is the most significant predictor of PFS.

This study examines the value of ¹⁸F-FDOPA PET in assessing treatment response in patients with recurrent high-grade gliomas on bevacizumab therapy. First, this study demonstrates that MTV measured by ¹⁸F-FDOPA uptake is predictive of OS in patients with recurrent high-grade glioma. Absolute MTV as measured by ¹⁸F-FDOPA uptake as early as 2 weeks after starting treatment is the best predictor of OS (P < 0.001). Evaluation of ¹⁸F-FDOPA MTV at 2 weeks stratifies patients into 2 subgroups, ¹⁸F-FDOPA responders (17/28, 61%) and nonresponders (11/28, 39%). ¹⁸F-FDOPA responders survived for 12.1 months, which is 3.5 times longer than the 3.5 months survival for nonresponders (P < 0.001). Second, this study shows that change of ¹⁸F-FDOPA MTV at 2 weeks and at 6 weeks (P = 0.002 and 0.02) is also predictive of OS, albeit being not as strong a predictor as the absolute MTV at 2 weeks. Third, ¹⁸F-FDOPA MTV as well as MTV change at 2 weeks after treatment initiation are predictive of PFS (P < 0.001 and P = 0.001 respectively). Fourth, ¹⁸F-FDOPA SUV values as well as their changes at any time point are not predictive of survival. Finally, this study compared metabolic responses by ¹⁸F-FDOPA with MRI response based on RANO criteria. MRI identifies 22 responders (76%; 2 complete response, 7 partial response, and 13 stable disease), and 7 patients with progressive disease (24%) and is predictive of OS. Median OS is 1.5 times longer for MRI responders than for MRI nonresponders (11.4 months vs. 7.7 months, P = 0.03).

MRI contrast enhancement primarily reflects a disrupted BBB, which can be influenced by changes in corticosteroid dose as well as other treatment effects such as inflammation, radiation necrosis, and postsurgical change (3). In addition, nonenhancing T2-weighted and fluid-attenuated inversion recovery sequence changes can reflect tumor recurrence as well, especially in patients on antiangiogenic therapy (31, 32). Thus, it has been proposed by the recently published Response Assessment in Neurooncology (RANO) that changes in both enhancing and nonenhancing areas should be considered in evaluating treatment response by MRI (31, 32).

All patients in this study had also undergone ¹⁸F-fluoro-thymidine (FLT) PET scans as previously reported (5). Current result showed while response criteria is different with ¹⁸F-FLT PET (maximum tumor SUV changes rather than tumor volume), both ¹⁸F-FDOPA PET and ¹⁸F-FLT PET have similar predictive value in treatment response in this group of patients. OS for responders and nonresponders by ¹⁸F-FLT PET is 12.5 months versus 3.8 months (P < 0.001) and OS by ¹⁸F-FDOPA PET is 12.1 months versus 3.5 months (P < 0.001).

As amino acid uptake in tumors is not limited to the disruption of BBB, amino acid PET imaging is particularly advantageous in evaluating noncontrast enhancing infiltrating tumors. Amino acid imaging with ¹⁸F-FET PET in monitoring treatment in 11 and 10 patients with glioma were reported previously (15, 16). In was shown that ¹⁸F-FET PET generally demonstrated treatment failure earlier than MRI. However, follow up ¹⁸F-FET-PET was done at 6 to 8 weeks after starting treatment in that study, much later than our study (15, 16). Similar to our study, it was shown that ¹⁸F-FET PET tumor volume change, but not the SUV change, was predictive of PFS. Interestingly, it was noted in our study that not only ¹⁸F-FDOPA tumor volume changes but also the ¹⁸F-FDOPA tumor volumes at follow up study are predictive of treatment response. As this finding has not been previously reported, rigorous statistical analyses were performed using a leave one out cross-validation analysis to assess how robust and sensitive the cut point to individual values. In addition, a simulation to assess whether this finding was likely a function of selection bias was performed. The process involves breaking up the relationship between ¹⁸F-FDOPA MTV and survival variable by randomly permuting the observations and then optimizing a cut-point to create an empirical distribution of C-statistics to compare our result to.

In our study, baseline ¹⁸F-FDOPA MTV is not predictive of OS (P = 0.08) or PFS (P = 0.31) but the MTVs in the follow up studies are. This is probably because of the fact that whereas tumor amino acid uptake is not dependent on a permeable BBB, it is enhanced by a broken BBB (34–37). High-grade gliomas have pronounced tumor neovascularization through the VEGF pathway and anti-VEGF therapy results in an apparent normalization of the highly permeable BBB (38, 39). It is likely that baseline ¹⁸F-FDOPA uptake is in part because of this disturbance of the BBB, thus reflecting a larger than actual tumor volume and therefore is not prognostic of survival. After “normalization” of the disturbed BBB with bevacizumab, ¹⁸F-FDOPA uptake in the follow-up studies becomes more reflective of tumor metabolism and outlines an effective tumor volume. Both the decrease of the tumor volume after treatment initiation as well as the resultant tumor size are prognosticators of survival. The fact that only the follow-up ¹⁸F-FDOPA MTVs but not the baseline ¹⁸F-FDOPA MTV are prognostic, must to some extent reflect the effects of anti-VEGF therapy irrespective of the therapy's actual mechanism.

This study has several limitations. First, the study population was small, thus these result need to be validated in a larger prospective study. Second, the patient population is relatively heterogeneous. There was a mix of grade 3 (N = 6) and grade 4 (N = 24) patients. Patients also had a various number of recurrences (median 1.77, range 1–5). One could argue that with a patient population at different stages of disease, one would see different prognosis at baseline. Indeed, the number of recurrences before starting treatment is predictive of OS and PFS. However, it is not as strong a predictor as ¹⁸F-FDOPA metabolic response, so a significant added predictive value of ¹⁸F-FDOPA metabolic response is seen (Table 4).

This study demonstrates that metabolic imaging with ¹⁸F-FDOPA PET provides a powerful prognostic tool in assessing treatment response in patients with recurrent high-grade glioma on bevacizumab therapy. This response already becomes apparent 2 weeks after starting treatment. As absolute MTV can be used for assessment, 1 ¹⁸F-FDOPA PET at 2 weeks suffices to assess treatment response, thus potentially making treatment response assessment with ¹⁸F-FDOPA PET both early and cost effective. A further and larger validation study is needed to test the potential of ¹⁸F-FDOPA PET in guiding treatment decisions.

J. Czernin has ownership interest (including patents) in Sofie Biosciences. T.F. Cloughesy is a consultant/advisory board member for Genentech and Roche. W.B. Pope is a consultant/advisory board member for F. Hoffmann-La Roche. No potential conflicts of interest were disclosed by the other authors.

Conception and design: J. Schwarzenberg, J. Czernin, T.F. Cloughesy, W.B. Pope, D.H.S. Silverman, W. Chen

Development of methodology: J. Schwarzenberg, J. Czernin, T.F. Cloughesy, W.B. Pope, D.H.S. Silverman, W. Chen

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T.F. Cloughesy, W.B. Pope, W. Chen

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J. Schwarzenberg, B.M. Ellingson, T. Grogan, D. Elashoff, C. Geist, D.H.S. Silverman, W. Chen

Writing, review, and/or revision of the manuscript: J. Schwarzenberg, J. Czernin, T.F. Cloughesy, B.M. Ellingson, W.B. Pope, D. Elashoff, D.H.S. Silverman, M.E. Phelps, W. Chen

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J. Schwarzenberg, T.F. Cloughesy, B.M. Ellingson, C. Geist, W. Chen

Study supervision: J. Czernin

This study was supported by grants P50 CA086306 from the National Institutes of Health — National Cancer Institute, and U.S. Department of Energy contract DE-FC03-87-ER60615, NIH/National Center for Advancing Translational Science (NCATS) UCLA CTSI Grant Number UL1TR000124.

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.