Purpose: We previously detected promising efficacy of neoadjuvant nintedanib (a multityrosine kinase inhibitor, TKI) in early HER2-negative breast cancer. In a preclinical study, we monitored stromal hypoxia with 18F-fluoromisonidazole-positron emission tomography (18F-FMISO-PET); we found that reoxygenation of tumors (or lack of it) during a window-of-opportunity (WoO) treatment with TKIs correlated with the benefit (or lack of it) from TKI-plus-chemotherapy combinations. We studied the predictive role of 18F-FMISO-PET for the TKI nintedanib in the neoadjuvant setting in a phase II WoO randomized trial.
Experimental Design: Patients were randomized to a 14-day WoO of nintedanib preceded and followed by an 18F-FMISO-PET, followed by nintedanib plus weekly paclitaxel (Arm A) or an 18F-FMISO-PET followed by weekly paclitaxel (Arm B) before surgery. The endpoint was residual cancer burden (RCB). The objective was to detect the patients with no response (RCB-III) on the basis of the baseline or evolutive 18F-FMISO-PET values/changes.
Results: One-hundred and thirty HER2-negative patients were randomized. Seventeen (27.9%), 34 (55.7%), and 8 (13.1%) patients had an RCB of III, II, and I/0, respectively, in Arm A. In this arm, baseline hypoxic tumors had a 4.4-fold higher chance of experiencing RCB = 3 (P = 0.036) compared with baseline normoxic tumors. Nintedanib WoO induced tumor reoxygenation in 24.5% of the patients; those not reoxygenating showed a trend toward higher chance of experiencing RCB-III (6.4-fold; P = 0.09). In Arm B, 18F-FMISO-PET lacked predictive/prognostic value.
Conclusions: Baseline hypoxic tumors (measured with 18F-FMISO-PET) do not benefit from neoadjuvant nintedanib. Clin Cancer Res; 23(6); 1432–41. ©2016 AACR.
Small multikinase inhibitors (tyrosine kinase inhibitor, TKI) targeting tumor stroma have shown activity against breast cancer. However, the development of predictive factors or biomarkers has been elusive. These agents are usually administered with conventional cytotoxics. According to the vascular normalization hypothesis, these agents facilitate chemotherapy delivery, increasing its efficacy. Alternatively, when the TKIs do not exert vascular normalization or cause excessive vascular pruning, TKI treatment would add no benefit to chemotherapy. In a preclinical study, we demonstrated that measuring tumor hypoxia with 18F-fluoromisonidazole PET mirrored the status of vascular normalization. We incorporated this technique in the neoadjuvant setting of breast cancer patients treated with the TKI nintedanib in combination with chemotherapy. Nintedanib was ineffective in patients with baseline hypoxic tumors compared with the remainder. These results and the ease of application of this technique show that it may be used as an imaging modality to assist patient-tailored medicine with this or similar agents.
Nintedanib is a multityrosine kinase inhibitor (TKI) that blocks several axes involved in the maintenance of an abnormal tumor stroma, such as VEGFR1-3, PDGFRß, FGFR1-3, RET, SRC, and FLT-3 (1). We conducted a phase IB trial assessing the safety and the preliminary efficacy of nintedanib in combination with weekly paclitaxel followed by adriamycin plus cyclophosphamide in patients with early HER2-negative breast cancer, which yielded promising results (50% pCR; ref. 2). Nintedanib has also shown survival advantage in NSCLC and progression-free survival benefit in ovarian carcinoma (3, 4). With these encouraging results, we have conducted a randomized pilot trial with a window-of-opportunity (WoO) aiming to define biomarkers of activity/resistance to nintedanib in early breast cancer, before conducting a large phase III that would incorporate those factors in its design.
The stroma of breast cancer is abnormal, as in the majority of malignancies (5, 6), and it is characterized by an aberrant activation of endothelial cells, leukocytes, pericytes, and fibroblasts (5–7). Abnormal stroma is thought to induce chemoresistance (5–8). Renormalization caused by agents targeting the stroma would resensitize the tumor to chemotherapy through various mechanisms, including increased chemotherapy delivery to the interstitium, decreased hypoxia, or recovery of the antitumor immune response (5, 9, 10). Theoretically, the positive effects of antistromal agents would be restricted within those patients experiencing normalization. Alternatively, according to the model proposed by Rakesh Jain in 2005 (11), the equilibrium of pro- and antiangiogenic factors naturally existing in the tumor in a certain moment, could be pushed by an agent with antiangiogenic activity toward a situation of “inadequate vasculature.” In this scenario, far from inducing vascular normalization, the treatment would cause hypoxia, lack of perfusion and chemotherapy delivery, and globally detrimental effects (11). Currently, whether drugs with antiangiogenic effect induce hypoxia or normoxia is yet unclear: We have shown how the same cancer model can experience an increase in hypoxia and vessel abnormality in response to a mAb against VEGF but vessel normalization and reoxygenation in response to TKIs (article in press, enclosed with the submission). We have shown as well how the same agent (dovitinib, a TKI) administered to two different models can induce hypoxia in one model and normoxia and normalization in the other (12). Recently, three simultaneous articles with breast, kidney, and neuroenocrine tumor models treated with different antiangiogenics show mixed responses as well (within the same tumor, some areas become normoxic and others hypoxic; refs. 13–15). In the clinical setting, there are studies suggesting both improved (16) and worsened (17) perfusion after antiangiogenic treatment. Taken together, these data suggest that due to the dynamic nature of the equilibrium between pro- and antiangiogenics in the tumor and the different pharmacologic properties of antistromal/antiangiogenic agents, the stromal response is heterogeneous and should be assessed in an individual basis.
We hypothesized that tracing the tumor stromal oxygenation levels with 18F-fluoromisonidazole-positron emission tomography (18F-FMISO-PET) during a WoO would allow detecting the patients that are or are not candidates for nintedanib therapy. The justification of this hypothesis is that it is widely accepted that abnormal stroma is hypoxic as well (5, 18–21). Thus, if hypoxia can be monitored, it could be possible to detect in which patients the stromal normalization phenomenon is occurring (i.e., continue nintedanib) or not (i.e., stop nintedanib). 18F-FMISO is a fluorine-labeled, positron-emitting nitroimidazole that binds to macromolecules in hypoxic areas, allowing its detection by PET (22). In a preclinical study, we have recently established proof-of-concept that 18F-FMISO-PET can monitor stromal normalization. We demonstrated how breast and pancreas cancer models had increased interstitial chemotherapy concentrations and tumor regression when 18F-FMISO-PET showed decreased hypoxia after a WoO priming with a TKI, but not when 18F-FMISO-PET remained stable or there was an increased standardized uptake value (SUV; ref. 12). Similar results were reported elsewhere (23). 18F-FMISO-PET has shown predictive value in head and neck cancer patients undergoing radiochemotherapy (24–26).
Thus, we have conducted this randomized study with the objective of studying the predictive role 18F-FMISO-PET for neoadjuvant nintedanib in breast cancer. We report the efficacy and toxicity data of the trial as well.
Patients and Methods
Study design and patients
This multicenter, phase II, randomized, open-label WoO trial designed for biomarker definition was conducted in 16 hospitals and the Spanish National Cancer Research Center (CNIO) in Spain.
Women ≥18 years old were eligible if they had a treatment-naïve histopathologically confirmed resectable HER2-negative breast cancer of >20 mm as measured by magnetic resonance. Other inclusion criteria included: ECOG performance status 0–1; adequate liver, renal and hematologic function [defined by serum bilirubin <1.25 × upper limit of normal (ULN) and AST/ALT <1.25 × ULN; serum creatinine <1.5 × ULN or creatinine clearance <50 mL/minute; and hemoglobin >10 g/dL, platelet count >100 × 109/L, and granulocyte count > 1.5 × 109/L, respectively); and negative pregnancy test within 7 days before study inclusion plus use of adequate contraceptive methods for fertile women. Patients suffering from bilateral tumors, concurrent severe conditions, taking anticoagulants, having major surgery within 4 weeks of study entry, or with a history of clinically significant bleeding or thromboembolic events within 6 months or study entry were excluded.
Ki67 fraction, estrogen receptor (ER), progesterone receptor (PR), and HER2 receptor status were determined by routine IHC (or FISH if required) at each site on the diagnostic biopsy.
Patients were randomized 1:1 to either of two arms: Arm A consisted of a WoO of 2 weeks of daily nintedanib (150 mg/twice daily) followed by weekly paclitaxel (80 mg/m2) plus nintedanib for 12 weeks (morning nintedanib doses were skipped on the chemotherapy days). The WoO was preceded and followed by an 18F-FMISO-PET, and an image-guided tumor biopsy. Arm B was preceded by an 18F-FMISO-PET and an image-guided biopsy, and consisted of 12 weekly paclitaxel courses (Fig. 1).
Randomization was centrally performed at GEICAM and it was stratified by study site, nodal status (positive versus negative) and hormonal receptors (either positive versus negative). Patients received anthracyclines, hormonal treatment, and radiation treatment upon completion of trial procedures if indicated.
The study was conducted in accordance with the Declaration of Helsinki and Good Clinical Practice standards. Institutional review board approval was obtained from all participating hospitals and CNIO. All patients provided written informed consent. This study (CNIO-BR-03-GEICAM/2010-10) is registered with Clinicaltrials.gov (NCT01484080).
Although the study was conducted in 16 sites, only 10 of them had PET scanners; thus, patients from some study sites were scanned at other study sites (patients from Hospital de Albacete, Hospital 12 de Octubre, Hospital Ramon y Cajal and Hospital de Fuenlabrada were all scanned at Hospital de Fuenlabrada; patients from the 2 Valencia sites were scanned at Hospital Clinico Universitario de Valencia; patients from ICO-Girona, Bellvitge and Vall d´Hebron were scanned at Bellvitge). Three types of scanners were used in 18%, 81.3%, and <0.7% of the patients, respectively: Discovery PET/CT 610 (General Electric); ECAT EXACT (CTI/Siemens) and Siemens ECAT EXACT-HR+ (CTI/Siemens). All of them were operated in a three-dimensional, high-resolution mode.
The PET scanners underwent quality control evaluation each time a study was performed. The cross-calibration between the different scanners was performed by comparing phantom data between institutions (same phantom loaded with the same radioactivity dose). All the phantom calibration data were sent back to CNIO and analyzed by the same technician, using a standardized procedure: briefly, 15-cm wide ROIs centered on the phantom were visually inspected for image artifacts, plane-to-plane sensitivity, variations, and non-uniformities. Then, the phantom activity concentration obtained at each institution was quantitated.
18F-FMISO sterile solution (1H-1-(3-[18F]fluoro-2-hydroxypropyl-2-nitroimidazole) was synthesized by Instituto Tecnologico PET (Madrid, Spain) using an IBA Cyclone C18 cyclotron and an FMISO module (IBA Synthera V2 with HPLC). The obtained radiochemical purity was 99.9%, and the specific activity was 231.9 GBq/micromol. Three hours after intravenous injection of 259 MBq of 18F-FMISO, 20 minutes of static PET images were acquired in prone position.
PET image datasets were reconstructed iteratively by application of computerized tomography data for attenuation correction. Two molecular imaging experts, blinded to the clinical information, displayed coregistered images on a central reader workstation (GE Advantage window) provided with PET VCAR software for image analysis and quantification. 18F-MISO PET images were quantitatively analyzed. An elliptic volume of interest (VOI) was manually drawn following the contour of the primary breast lesions, and the maximum uptake value (SUVmax) was determined. A region of 0.5 cm3 from the spinalis thoracis muscle was selected for normalization purposes. The tumor-to-muscle ration (TMR) was defined as the tumor uptake divided by the muscle uptake. Although a threshold for the definition of hypoxia has not been formally established or standardized, Cheng and colleagues (27) proposed a TMR of 1.2 in a breast cancer study in which 45 tumor lesions (primary and metastatic) from 20 hormone receptor–positive patients were assessed by 18F-FMISO PET. Thus, we also calculated the hypoxic volume (HV) according to this threshold to study the changes in HV induced by nintedanib along the WoO. The 18F-FMISO uptake variation data observed in the spinalis thoracis muscle in an intrapatient basis (pre-WoO to post WoO) from the experimental arm are provided in Supplementary Table S1 and was used to estimate the threshold for considering significant a variation in TMR from the first PET scan to the second one (10%).
A minimum of 4 weeks after the last nintedanib dose (or last paclitaxel dose in Arm B), the patients were scheduled for surgery. Pathologic complete response (pCR) and residual cancer burden (RCB) determination were centrally performed according to the Symmans and colleagues method (28). Adverse events were assessed and graded according to the CTCAE v4.0.
RCB was the primary efficacy endpoint, but, despite the randomized design, the trial was not powered to compare the RCB rate between both arms (as it was a phase II trial) but to study potential predictive markers. The predictive markers under exploration were TMR values and phosphoprotein profiles. The safety of the combination was a secondary endpoint.
The pCR rate to single-agent weekly paclitaxel can be inferred from data of previous neoadjuvant studies and is approximately 10% to 15% (29, 30). A 15% pCR rate was assumed for the control arm (Arm B). A sample size of 65 patients would provide a power of 80% to detect a 15% variation in RCB index with a statistical significance of 0.05 (two-tailed). A number of 10 RCB-0/I and 30 RCB-III was deemed as the minimum number to detect discriminative power of the correlative studies. Thus, 65 patients had to be enrolled per arm to observe at least 10 RCB-0/I in each arm.
Descriptive statistics of main clinical/pathologic variables, response data, PET data and toxicity data were performed. All randomized patients were included in the intent-to-treat (ITT) population except for two, as after trial registration the staging work-up revealed M1 stage. The safety analysis was performed in the ITT population that received at least one dose of the study treatment. The efficacy population was the ITT except for three additional patients (1 from Arm A and 2 from Arm B) whose tumor measurements were 20 mm (inclusion criterion: >20 mm). All the comparisons were performed using a bilateral alpha of 0.05. The ability to predict the RCB with the PET results was studied with different logistic regression models, categorizing the response variable depending on the studied outcome (i.e., ability to predict obtaining RCB 3 or 0/1 vs. others). The covariates tumor size (T), nodal status (N), grade (G), Ki67, and status of hormonal receptors (HR) were included in the models, and encoded as follows: T, N, 1 to 4 or 0 to 3, respectively, according to their AJCC staging; G: 1 to 3; Ki67: 1 if Ki67 >14% and 0 if Ki67 <15%; HR: 1 if either receptor positive and 0 for triple-negative breast cancer (TNBC) cases. The percentage of TMR variation along the WoO was calculated as follows: [(TMR2 - TMR1)/TMR1)] × 100: where “2” stands for the PET performed on day +15 and “1” for the one performed at baseline. According to the TMR changes during the WoO, a new variable (“TMR Change”) was encoded as +1, 0, or −1 when the tumors experienced >10% decrease, <10% variation, or 10% increase, respectively, in their TMR. All analyses were performed with the SAS Enterprise Guide 5.1 and the SPSS v.19 software.
Patient demographic and clinical characteristics
From July 2012 to November 2013, 130 patients were randomly assigned (Table 1). The CONSORT diagram describes trial flow in Supplementary Fig. S1.
Safety and efficacy
The mean paclitaxel dose intensity was in 94.6% in Arm A and 95.7% in Arm B (P = 0.71). Regarding nintedanib, the mean dose intensity was 92.2%. Fifty-seven and 60 patients received the four complete courses of paclitaxel in Arm A and B, respectively. Fifty-six patients received the four complete courses of nintedanib.
Toxicity was generally low, and similar for both treatment arms (Table 2). There were five serious adverse events (SAE) in Arm A [grade 2 pneumonia (two patients), grade 3 hypokalemia, gastric microperforation, and cholestatic hepatitis] and two in Arm B (grade 3: hypersensitivity and migraines).
In the efficacy population, the results of the primary outcome were as follows: five [8.2%; 95% confidence interval (CI), 3.1%–18.8%] and 4 (6.5%; 95% CI, 1.8%–15.7%) patients experienced RCB-0 in Arms A and B, respectively (P = 0.71). In the hormone receptor–positive subpopulation, there was a trend to increased response rate: five of 49 patients (10.2%; 95% CI, 3.8%–23%) experienced RCB-0 in Arm A, and one of 48 patients in Arm B (2.1%; 95% CI, 0.1%–12.5%; P = 0.09). The percentage of patients with RCB-0 or 1 was almost 2-fold higher in the hormone-positive patients in Arm A [8 (16.3%; 95% CI, 7.8%–30.2%) vs. 4 (8.3%; 95% CI, 2.7%–20.9%)], but not statistically significant (P = 0.23). There were 17 and 23 patients that had RCB-III in Arms A and B, respectively (P = 0.27). The responses are summarized in Supplementary Table S2.
18F-FMISO-PET and primary outcome
The average 18F-FMISO injected dose was 267.4 MBq (SD 15.3 MBq; range, 213.1–325.6 MBq), and the uptake times ranged from 150 to 207 minutes (average: 182 minutes; SD: 8.2 minutes). Forty-nine patients had pre- and post-WoO 18F-FMISO-PET and RCB data available in Arm A, whereas 58 had baseline 18F-FMISO-PET and RCB data available in Arm B (Supplementary Fig S1). The description of the TMR variable among the trial population, the definition of the threshold for hypoxia, and the variation of the TMR during the WoO (“TMR Change”) are shown in Table 3. The status of baseline hypoxia highly correlated with outcome in the Arm A. The patients whose tumors were hypoxic at the baseline assessment had a 4.4-fold higher chance of having RCB-III than the rest of the patients (P = 0.036); the results of the logistic regression models for RCB-III based on different predictive hypoxia models are shown in Table 4. After a threshold of 10% was established to consider a TMR change as clinically significant during the WoO, nintedanib was able to induce a reduction in 18F-FMISO TMR uptake (i.e., reoxygenation) in 24.5% of the patients. The remainder mostly stayed within the plus, or minus, 10% boundaries, with the exception of one patient that had a significant increase in TMR uptake (i.e., increased hypoxia). Patients whose tumors did not experience a TMR decrease of at least 10% during the WoO (i.e., not reoxygenation or lack of stromal renormalization induced by nintedanib) had a 6.4-fold increase in the probability to obtain a RCB-III (i.e., no response) compared with the remainder, although this correlation did not reach statistical significance (P = 0.09). Because the number of patients with RCB-0 or -I (pCR) was low, we did not identify any statistically significant association between the opposite parameter (TMR decrease - reoxygenation - along the WoO and RCB-0/I). The box-plot diagrams shown in Fig. 2A suggest a trend toward a worse RCB outcome as the magnitude of the TMR decrease during the WoO (i.e., the degree of reoxygenation) is smaller.
According to the Symmans and colleagues (28) classification, patients experiencing a RCB = 3 response belong to a unique disease subgroup with dismal (>50% relapse rate) prognosis. The patients that obtained RCB = 0, 1, or 2 (n = 35) had >15-fold decrease in the TMR (better reoxygenation) during the WoO than the patients that experienced RCB-III (n = 14; average decrease of 7.75% vs. 0.5%, P = 0.014). All the patients with pCR (RCB-0 or -I) experienced some degree of reoxygenation during the WoO (i.e., 0% of the patients with no TMR change experienced pCR). However, the threshold for significant TMR decrease was set in 10% on the basis of the inter-instrument variability (Supplementary Table S1); thus, the relevance of these results should be taken with caution.
Finally, in Arm B, baseline 18F-FMISO-PET did not show any type of association with RCB, thus suggesting a predictive, not prognostic, value of 18f-FMISO-PET for nintedanib. PET images of baseline hypoxic tumors and tumors experiencing hypoxia worsening and improvement during the WoO are shown in Fig. 2B–D, respectively.
Hypoxic tumor volumes, according to a threshold proposed elsewhere (27), were highly variable and did not correlate with clinical outcomes (Supplementary Table S3).
This study was designed to monitor 18F-FMISO-PET during a WoO with nintedanib, a novel TKI with interesting results to date, and to determine whether 18F-FMISO PET would show predictive/prognostic capacity or shed light on the mechanisms of resistance to nintedanib. The randomized design was adopted to distinguish predictive from prognostic properties of the biomarker. To avoid confusion in the biomarker interpretation, most of the standard chemotherapy that is usually administered in this setting (a taxane plus an anthracycline-based regimen) was removed from the trial schedule; the anthracycline-based regimen was administered after trial completion in the post-operative setting. Thus, despite the randomized design, the trial was not powered to compare the RCB rate between treatment arms or determine the true efficacy of nintedanib plus standard neoadjuvant chemotherapy. In any case, doubling the rate of pCR (RCB-0 and –I) in the hormone-positive population, which constituted the 79% of the randomized patients, by adding nintedanib to paclitaxel seems promising. The assumption of a pCR between 10% and 15% for single-agent paclitaxel was made from data from old trials in the neoadjuvant setting (ref. 30; actually, only one of them tests single-agent paclitaxel, but in a three-weekly schedule; ref. 29). This assumption would require a similar distribution of hormone-positive and TNBC cases than that in those trials. Those trials included HER2-positive patients as well but reporting the HER2 status was not standard back then; thus, the specific response rate broken down by disease subtype is not available. Excluding the HER2-positive population, the distribution of hormone-positive/TNBC incident cases can be estimated in 80%/20%, which was approximately the proportion included in our trial (Table 1). The pCR should be higher in the TNBC cases; and, within TNBC cases, the response is higher in the basal-like cases and lower in the luminal androgen receptor cases (31). The size of our trial precludes breaking down the response rate by disease subtype; however, despite the limitations in the pCR assumptions, we achieved a pCR rate within the pre-established boundaries of 10% to 15% (Supplementary Table S2). The results in the experimental arm were at expense of an increased incidence of SAEs probably related to nintedanib (pneumonia, gastric microperforation, hypokalemia and cholestatic hepatitis); toxicity should be carefully monitored in future trials.
A novel feature of this work is the incorporation of an imaging test to monitor if stromal renormalization and reoxygenation occur during a WoO, as proposed by the hypothesis of vascular normalization. We have found a biomarker, 18F-FMISO-PET, that can detect at baseline evaluation which patients (those with hypoxic tumors) will not experience benefit from nintedanib. Targeted agents added to chemotherapy regimens in breast cancer usually yield small numeric improvements in outcomes (32–35); thus, identifying the population that does not benefit from targeted agents is very appealing.
This study was conducted to test the model of vascular normalization in the clinical setting, proposed by Rakesh Jain in 2005 (11): The vascular network that adequately supports cell metabolism and turnover, and warrants sufficient oxygenation and nutrients delivery in normal tissues is supported by the equilibrium between pro- and antiangiogenic factors. This equilibrium is disrupted in tumors; the disruption can be originated by either an increase or a decrease in pro- and/or antiangiogenic factors (i.e., both sides of the balance), leading to decreased perfusion and hypoxia. The equilibrium is dynamic in time, and drugs can interact with the equilibrium in various ways depending on the time of administration and their pharmacological properties. Theoretically, restoring the equilibrium back to normal or "almost normal" would be beneficial for chemotherapy coadministration (better chemotherapy delivery in a normoxic environment), whereas pushing it to more abnormal would be detrimental (11). Preclinical studies have shown that different antiangiogenics can cause different effects in this equilibrium, and even within the same tumor, the effects may be nonhomogeneous (12–15, 23, 36, 37). However, the model seems proved at the preclinical level; moreover, the positive effects of antiangiogenics in stromal normalization seem to be traceable with 18F-FMISO-PET (12, 23, 38). Unfortunately, in the clinical setting, the positive effects of antiangiogenic treatment in chemotherapy delivery are yet unproved. Actually, an elegant study with labeled docetaxel administered after bevacizumab showed decrease docetaxel delivery to lung tumors (17). This observation is complex to explain within the context of the model of vascular normalization and the positive effects of adding bevacizumab to chemotherapy in lung cancer (39). Other studies suggest that antiangiogenic agents improve the perfusion (16, 40, 41); one of these studies showed an improved vascular network at the microscopic level in the patients that experienced favorable outcomes; however, the low number of patients and the nonrandomized nature of the study precludes definitive conclusions (40). We expected to observe a heterogeneous response, and that the patients showing improved oxygenation or hypoxia worsening would be those with better and worse outcomes respectively, as an indirect proof of vascular normalization. The statistical significant association between baseline hypoxia and RCB = 3 (Fig. 2B, Tables 3 and 4) could suggest that nintedanib exerts detrimental effects in the vasculature (vessel pruning, turning the vasculature inadequate and thus interfering with chemotherapy delivery), what fits with the Rakesh Jain's model suggesting a dynamic balance shifting from vascular abnormality to vascular normalization or inadequate vasculature. This observation is strengthened by the fact that baseline hypoxia/normoxia was unrelated with the outcome in the standard arm. However, the serial evaluation along the WoO did not capture the hypoxia worsening in this subgroup. The association between lack of oxygenation and RCB = 3 was not statistically significant (Tables 3 and 4; Fig. 2A and C) either. In addition, although all patients showing RCB = 2/1/0 showed some degree of TMR decrease, and that TMR decreased >15-fold during the WoO in the patients with RCB = 2/1/0 vs. those with RCB = 3 (P = 0.014), the TMR values were within the pre-established boundaries for “noise” (7.7%). Taken together, these results indirectly suggest that the hypothesis of vascular normalization (in the clinical setting) is not universally true in that in some cases that treatment may exacerbate hypoxia, and high levels of pretherapy hypoxia can portend a poor outcome in subsets of patients. In addition, the results suggest that patients with baseline hypoxic tumors should not receive nintedanib treatment (potentially because of leading to an inadequate vasculature); however, we were not able to establish definitive (indirect) proof, suggesting that some degree of hypoxia correction/normalization is necessary to achieve a good response or that hypoxia worsening is linked to an adverse outcome. The potential reasons for it might be: First, the study was designed with a WoO duration similar to that in the animal studies; however, the “tumor lifespan” is much longer in humans. Possibly, a longer WoO would be able to detect statistically significant associations between TMR changes and outcomes; however, such a trial may be questionable from the ethics point of view. Second, the RP2D for single-agent nintedanib is 200 mg/twice daily. It could happen that the lower dose that we used was not sufficient to induce the expected changes; however, the higher dose might have induced higher toxicity when chemotherapy was introduced. Third, one of the key features for defining accurate prognostic/predictive factors is adopting a randomized design, whereas another one is the percentage of patients experiencing a positive primary outcome. The main strength of this study is its randomized nature. The low number of pCRs, and the moderate methodologic failure in the experimental arm is its main weakness and has probably affected the statistical power of the study. The number of missing PETs was higher in the experimental arm (11 patients missing one of the two scans) than in the standard arm (5 patients missing baseline scan; Supplementary Fig. S1). Because we only administered weekly paclitaxel, a low number of RCB 0 or -I was expected; but the predicted methodologic failure (10%) expected only 5 to 6 missing scans per arm, and thus it may have affected the lack of observed correlations between pCRs and TMR changes during the WoO. However, we were able to detect the patients experiencing RCB = 3 with the baseline PET, which could be very useful in the future as it translates into therapeutic resistance. Studying the “adverse outcome” is quite relevant, because it allows formulating hypotheses related with therapeutic resistance that may lead to new therapeutic opportunities (i.e., targeting hypoxia in combination with nintedanib for the patients not showing PET improvement). Fourth, and last, it could be that vascular normalization is less relevant in the clinical setting (for achieving a response) than inadequate vasculature (for treatment failure).
An additional point of interest is what 18F-FMISO uptake parameter should be used in future studies. On the basis of 45 tumor lesions from 20 breast cancer patients (combining 13 primary tumors and 32 metastases) Cheng and colleagues (27) defined a cutoff value of 1.2; this value was the one that better discriminated the response to hormonal therapy. We applied this value to determine the percentage of hypoxic tumor volumes, but we did not find a correlation between the relative hypoxic volume, or the changes in the relative hypoxic volume during the WoO, with clinical outcomes (Supplementary Table S3). Part of the explanation might be due to the fact that the value of 1.2 (or any other) lacks a correlate with tissue oxygenation determinations (which, in the preclinical setting can be performed); in addition, the mechanism of action of hormonal therapy is quite different to that of the multi-TKIs. In our study, we set the hypoxic threshold in the 75th percentile (a pre-specified cutoff set in the statistical analysis plan, before the data distribution was known). The 75th percentile was 1.73. We also tested the correlations between hypoxic tumor volume (baseline, or change during WoO) and clinical outcome, but we did not find any association (data not shown, because our study was not powered for multiple comparisons). Alternatively, it may occur that the TMR values (TMR = normalized SUVmax values = tumor SUVmax normalized by muscle SUVmax) translates more accurately the tumor biology than parameters related to the hypoxic volumes (similarly to the case of FDG), evidenced by the association found in our study. We encourage other researchers to compare both parameters (TMR and hypoxic tumor volumes) in future studies in breast cancer, because the volumes and degree of hypoxia reported for head and neck cancer, the pathology where 18F-FMISO PET has been most widely used, are quite larger than in breast cancer (24) and probably not extrapolable.
In the CNIO-BR-003/GEICAM/2010-10 study, we were able to show a predictive role of 18F-FMISO-PET in HER2-negative early breast cancer treated with nintedanib, detecting the patients that do not experience benefit from it based on baseline tumor oxygenation levels. The results suggest as well that adding nintedanib to paclitaxel improves efficacy in hormone receptor–positive patients. A randomized study combining nintedanib with polichemotherapy incorporating 18F-FMISO-PET in the design to exclude patients with low chance of benefit from nintedanib is warranted.
Disclosure of Potential Conflicts of Interest
M. Quintela-Fandino reports receiving commercial research grants from Boehringer Ingelheim. J. Cortes reports receiving speakers bureau honoraria from Celgene, Eisai, Novartis and Roche, and is a consultant/advisory board member for Celgene and Roche. A. Llombart-Cussac and E. Alba are consultant/advisory board members for Pfizer and Roche. No potential conflicts of interest were disclosed by the other authors.
Conception and design: M. Quintela-Fandino, L. Manso, A. Llombart-Cussac, F. Mulero, R. Colomer
Development of methodology: M. Quintela-Fandino, N. Martinez-Jánez, A. Llombart-Cussac, F. Mulero, R. Colomer
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): M. Quintela-Fandino, A. Lluch, I. Calvo, J. Cortes, J.A. García-Saenz, M. Gil-Gil, A. Gonzalez-Martin, E. Adrover, R. de Andres, G. Viñas, E. Alba, J. Guerra, B. Bermejo, E. Zamora, F. Moreno-Anton, S. Pernas Simon, A. Carrato, F. Mulero
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Quintela-Fandino, I. Calvo, E. Adrover, A. Llombart-Cussac, M.J. Escudero, E. Carrasco, F. Mulero, R. Colomer
Writing, review, and/or revision of the manuscript: M. Quintela-Fandino, A. Lluch, I. Calvo, J. Cortes, J.A. García-Saenz, M. Gil-Gil, A. Gonzalez-Martin, E. Adrover, A. Llombart-Cussac, E. Alba, J. Guerra, B. Bermejo, E. Zamora, F. Moreno-Anton, S. Pernas Simon, A. Carrato, M.J. Escudero, R. Campo, E. Carrasco, J. Palacios, R. Colomer
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): R. Campo
Study supervision: M. Quintela-Fandino, E. Adrover, G. Viñas, B. Bermejo, A. Lopez-Alonso, R. Campo, E. Carrasco, R. Colomer
This study (CNIO-BR-003 GEICAM/2010-10) was funded by the following grants: FIS PI13/00430, FIS PI10/00288, SEOM 2015 and AECC Scientific Foundation 2010 (all of them awarded to M. Quintela-Fandino). AVON Spain Inc. contributed to this work with a philanthropic donation. Boehringer-Ingelheim Spain provided funds for the study.
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.
Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).
- Received March 28, 2016.
- Revision received July 12, 2016.
- Accepted July 19, 2016.
- ©2016 American Association for Cancer Research.