Bevacizumab Induces Acute Hypoxia and Cancer Progression in Patients with Refractory Breast Cancer: Multimodal Functional Imaging and Multiplex Cytokine Analysis

Breast cancer commonly exhibits active angiogenesis (1). Although tumor angiogenesis is known to be an important prognostic factor, the degree of the activity has not yet been stratified for tumor subtypes (2, 3). Emerging functional imaging technologies that can noninvasively visualize several parameters regarding cancer physiology can provide essential information on the therapeutic response (4, 5). ¹⁸F-fluorodeoxyglucose (FDG)–positron emission tomography (PET)/computed tomography (CT) provides information on glycolytic activity and is currently the most widely used functional imaging modality for early detection and evaluation of therapeutic efficiency (6, 7). ¹⁸F-fluoromisonidazole (FMISO) PET/CT has been validated as an effective method of imaging hypoxia and can capture hypoxic tissues by selectively taking an analog of nitroimidazole; as a result, it reflects the degree of intracellular hypoxia in cancer cells (8). Concurrent FDG- and FMISO-PET/CT scans are reportedly useful for evaluating tumor glycolysis linked to hypoxia (9). Diffuse optical spectroscopic imaging (DOSI) using near-infrared light is another noninvasive approach to measure tissue oxygen levels, which reflect the relative difference between oxy (O₂)-hemoglobin (Hb) and deoxyhemoglobin (HHb), and consequently, tissue oxygen consumption. Total hemoglobin (tHb) and oxygen saturation (SO₂) are markers of blood volume and tissue oxygen level, respectively (10, 11).

Bevacizumab, a human recombinant antibody against VEGF-A, is widely used in cancer treatment. In a randomized phase III clinical trial (ECOG2100) in patients with advanced and/or metastatic breast cancer, the addition of bevacizumab to paclitaxel treatment yielded significant improvements in progression-free survival but not prolongation of overall survival (OS) when compared with paclitaxel treatment alone (12). Combination therapy of bevacizumab and paclitaxel was consequently approved in the metastatic setting; however, it is controversial because several pivotal trials have failed to demonstrate an improvement in OS for patients treated with bevacizumab (13). Recent preclinical data have indicated that bevacizumab induces vascular remodeling and subsequent tissue oxygenation, which improve drug delivery and stabilize the tumor microenvironment (14, 15). In contrast, failure of the vascular remodeling may disrupt the vascular structure and trigger acute hypoxia (16, 17). However, there is a lack of clinical evidence showing the negative effect of antiangiogenic therapy (18, 19). Despite comprehensive research, the mechanism of bevacizumab resistance remains poorly understood. One possible reason is the difficulty of identifying molecular biomarkers associated with the synergistic effect of bevacizumab and cytotoxic chemotherapy in humans (20, 21). We believe that intrinsic resistance to bevacizumab is more dependent on physiologic changes during treatment than genomic alterations (22, 23). Thus, direct, quantitative, and continuous measurements of tumor vascular remodeling and hypoxia in clinical practice are necessary to monitor the therapeutic response in terms of the antiangiogenic strategy. To elucidate the mechanism of bevacizumab resistance, we conducted a clinical study with six courses of bevacizumab combined with paclitaxel. As we previously reported data of initial experience of 7 patients in the study protocol, paclitaxel was omitted on the first day of the first cycle and thus, a single-agent bevacizumab was initially administrated (24). Although RECIST or histopathologic response criteria are the gold standard, recent studies demonstrated that binary classification according to FDG-PET–based metabolic response could help distinguish nonresponders to chemotherapy from responders and avoid unnecessary adverse effects and ineffective treatment among nonresponders (25, 26). Furthermore, the information from FDG-PET/CT is useful because patients with metastatic breast cancer, the majority of whom develop bone metastasis, often cannot be evaluated according to RECIST. Because metabolic changes to chemotherapy precede tumor shrinkage, early PET assessment has recently been adopted in many clinical trials (27–29).

Sequential scans of FMISO-PET were used to quantitatively measure the hypoxic activity of the tumors both during pretherapy and after the second course of concurrent treatment with bevacizumab and paclitaxel; further, DOSI was utilized several times to determine the changes in the tissue levels of Hb and SO₂ from baseline to immediately after the initiation of single-agent bevacizumab. Moreover, blood samples were collected from the patients to observe specific variations of multiple cancer-related cytokines between responders and nonresponders during treatment using single-agent bevacizumab.

From October 2012 through September 2016, 30 patients with locally advanced stage III/IV HER2-negative breast cancer who received combination chemotherapy with paclitaxel and bevacizumab were registered. No patients experienced to receive cytotoxic chemotherapy prior to the study participation, and 12 patients had received prior endocrine therapy. This study was approved by the Institutional Review Board of the Saitama Medical University International Medical Center, and written-informed consent was obtained from each patient prior to participation in the study. The study was registered in the UMIN clinical Trial Registry (UMIN000015837, 000006802).

All patients received bevacizumab (10 mg/kg body weight) intravenously on days 0 and 14 in combination with paclitaxel (90 mg/m² body surface area) on days 0, 7, and 14 of every cycle. Dexamethasone (6.6 mg) and an H₂ antagonist were used for supportive treatment during the course of chemotherapy. In the regimen, paclitaxel and the supportive treatment were omitted on the first day of the first cycle. Therefore, the first drug administered was bevacizumab alone. Treatment continued for six courses unless there was disease progression, unacceptable toxicity, or withdrawal of consent. If the study treatment was discontinued, further systemic and/or local therapy including surgery was permitted at the investigator's discretion.

Histologic evaluation of the primary tumor tissue resected from patients treated with bevacizumab and paclitaxel was performed according to the histopathologic classifications to categorize the tumor response as follows; pathologic complete response (pCR), defined as no histologic evidence of invasive tumor cells in the breast; pathologic partial response, indicating histologic evidence of marked degeneration of residual invasive carcinoma cells; and pathologic no response, indicating almost no change in cancer cells after treatment.

Serial scans of PET/CT (Biograph6, Siemens) with FDG and FMISO were scheduled before chemotherapy initiation and within 3 weeks after two courses of chemotherapy. The baseline PET/CT scan was taken after waiting for at least 1 week after performing a diagnostic core biopsy, but no later than 1 week before the initial infusion of the drugs. Patients were intravenously administered FDG (3.7 MBq/kg) and FMISO (7.4 MBq/kg) on separate days. With the patients positioned prone on the whole-body PET/CT scanner couch, FDG PET/CT and FMISO PET/CT were performed 1 hour and 2 hours after the administration of the tracers, respectively. Regions of interest (ROI) were drawn over the primary tumor, and the maximal standardized uptake values (SUV_max) were recorded from the target. SUV_max was calculated according to the following formula: SUV = activity concentration in ROI (MBq/mL)/injection dose (MBq/kg body weight). Dynamic contrast-enhanced (DCE)-MRI of the breasts was used to measure the maximal size of the tumors at baseline and after two courses of chemotherapy, as well as after six courses of chemotherapy. Detailed methodologies for the PET/CT scans with FDG and FMISO, as well as DCE-MRI, were obtained from a previous study (24).

The TRS breast imaging system (TRS20; Hamamatsu K.K.) uses time-correlated single-response profiles of tissue against optical pulse inputs and enables quantitative analysis of light absorption and scattering in tissues according to the photon diffusion theory (30, 31). This approach enabled us to quantify the concentration (μmol/L) of O₂Hb and HHb in the breast tissue. The concentration of tHb (O₂Hb + HHb) and percentage (%) of SO₂ (O₂Hb/tHb × 100) in the tissue were calculated from these parameters. The distribution of Hb in the breast lesion tissue and the contralateral normal breast was reconstructed using a custom-made software (Gridbreastviewer; Sincere Technology). The mean values of tumor Hb and SO₂ were calculated from the TRS measurements taken for the breast tissue corresponding to the ROI using a previously described method (11). The ROIs were determined using ultrasound on the tumors. Changes in the concentration of Hb and the percentage of SO₂ before therapy (day −1 or day 0) and on days 1, 3, 6, 8, and 13 in the tissue were measured after the first infusion of bevacizumab, as well as after the first (post C1) and second courses (post C2) of combination chemotherapy. Details of the system and the measurement procedure have been previously described (11, 24, 31).

Peripheral blood samples obtained from patients treated with single-agent bevacizumab on day 0 (before therapy) and day 1 and days 3 and 4 were analyzed for plasma biomarkers. The plasma samples were separated by centrifugation and then aliquoted and stored at −80°C. ELISA was performed for free (non–bevacizumab-bound) VEGF-A, basic fibroblast growth factor (bFGF), FLT-3L, EGF, G-CSF, TNFα, IL1β, IL4, IL6, IL8, IL12p40, and IFNγ with a chemiluminescent immunoassay–certified multiplex protein array using the MILLIPLEX MAP Human Cytokine/Chemokine Magnetic Bead Panel. ELISA was also carried out for the TGF family using the MILLIPLEX MAP TGFβ1,2,3 Magnetic Bead Kit using the Luminex 200TM System. All samples were assayed in duplicate.

Schedules for the imaging studies using FDG-PET/CT, FMISO-PET/CT, MRI, and DOSI are shown in Fig. 1A. A cutoff value of 20% decrease of the SUV_max from the baseline in the early metabolic assessment by FDG-PET/CT was adopted to distinguish nonresponders from responders, taking into consideration therapeutic and survival outcome (Supplementary Fig. S1). Figure 1B shows the SUV_max changes in the tumors, which were classified into responders and nonresponders.

We considered that at least 10 patients in each group were required to compare the variables between the responder and nonresponder groups. We reported changes in the plasma cytokine concentrations from baseline as ratios; these changes were assessed using the one-sample, two-sided, exact Wilcoxon test. Blood samples for which cytokine measurements were missing were excluded from the analysis. The t test was used to analyze the other continuous variables. A value of P < 0.05 was considered statistically significant. The data were analyzed using the Medcalc software.

Of the 30 patients enrolled, 2 patients were excluded because of their preference for surgical treatment and a blood transfusion owing to severe anemia. Thus, a total of 28 patients were included in the study. The patients' characteristics are presented in Table 1. Among 18 patients who underwent mastectomy with axillary dissection, 13 (72.2%) patients with clinical stage 3 who had advanced breast cancer underwent definitive surgery, and 5 (27.8%) patients with clinical stage 4 who had distant metastasis underwent salvage surgery owing to local control of the primary tumor and evaluation of therapeutic response. Another 10 patients continued chemotherapy after the study. Histologic response and overall clinical response were shown in Table 2. All 3 patients achieving pCR were included in a responder group. The histologic and clinical responses were correlated with the early metabolic response with statistical significance (P = 0.02 and 0.01, respectively).

Using dataset of the present study, an optimal cutoff value of change in FDG-SUV_max from before therapy until after the second course of therapy was calculated by the receiver operating characteristic curve analysis, which can divide patients into a poor survival group and a favorable survival group. The Kaplan–Meier analysis of OS was shown in Supplementary Fig. S1. The cutoff value resulted in 19.8% reduction in FDG-PET SUV_max without regard to distant metastasis. Therefore, the tumor metabolic response classified by the change in FDG-SUV_max with a cutoff value of 20% determined the patients to be divided into responder (18 patients, 65.5%) and nonresponder groups (10 patients, 34.5%; Fig. 1B).

The differential relationship between tumor glycolysis and hypoxia indicated by uptake levels of FDG and FMISO according to breast cancer subtypes is shown in Supplementary Fig. S2. In patients with triple-negative breast cancer (TNBC; n = 7), a linear correlation was found between FDG and FMISO, indicating that nonresponders have both higher glycolysis and more severe hypoxia, whereas responders have less glycolysis and hypoxia. The difference in FMISO uptake level between responders (n = 4, mean 2.1, SD 0.3) and nonresponders (n = 3, mean 3.3, SD 0.6; P = 0.02) was statistically significant. In contrast, among patients with luminal breast cancer (n = 20), a discrepancy between glycolysis, hypoxia, and metabolic response was shown. Although some hypoxic tumors with higher FMISO uptake were categorized as nonresponders, there were no significant differences in FMISO uptake level between responders (n = 14, mean 1.7, SD 0.2) and nonresponders (n = 6, mean 2.1, SD 0.7; P = 0.1).

The MRI scans showed that nonresponders had significantly lower tumor shrinkage [mean, –3.8%; 95% confidence interval (CI), –17.1 to 9.4; SE, 5.8; SD, 18.5 vs. mean –25.9%; 95% CI, –36.1 to –15.8; SE, 4.8; SD, 21.0; P = 0.009) than responders as early as after the second course of chemotherapy. Tumor shrinkage during the sixth course of chemotherapy marginally differed between nonresponders and responders (mean, –11.9%; 95% CI, –36.7 to 12.9; SE, 10.5; SD, 29.7 vs. mean, –38.7%; 95% CI, –57.7 to –19.6; SE, 8.8; SD, 32.9; P = 0.07; Fig. 1C). The FMISO-PET/CT study showed that nonresponders had a significantly higher baseline SUV_max (mean, 2.5; 95% CI, 1.8–3.2; SE, 0.3; SD, 0.9 vs. mean, 1.8; 95% CI, 1.7–2.0; SE, 0.07; SD, 0.3; P = 0.008) and a higher SUV_max after the second course (mean, 2.1; 95% CI, 1.6–2.7; SE, 0.2; SD, 0.6 vs. mean, 1.4; 95% CI, 1.3–1.6; SE, 0.07; SD, 0.3; P = 0.001) than responders (Fig. 1D).

DOSI measurement showed that baseline Hb parameters were significantly higher in cancer lesions than those in normal breasts. No significant differences were found in the baseline parameters neither between the subtypes nor between responders and nonresponders (Table 3). During the 1 week of observation after the initiation of single-agent bevacizumab, the tissue concentrations of tumor O₂Hb, HHb, and tHb in the nonresponders were significantly higher than those in the responders on day 3 and/or day 6, as well as after the first and/or second courses of chemotherapy (Fig. 2). Nonresponders tended to exhibit decreased tumor SO₂ immediately after treatment, but responders did not. The percentage of tumor SO₂ in the nonresponders was significantly lower than that in the responders from day 1 to day 3.

We analyzed the levels of 15 cancer-associated cytokines, including proangiogenic factors (VEGF-A, bFGF, TGFβ1), epithelial–mesenchymal transition (EMT)–induced factors (TGFβ family), growth factors (EGF, G-CSF), and inflammatory factors (IL1β, IL6, IL8, TNFα). Nonresponders exhibited upregulation of proangiogenic factors and inflammatory factors from day 1 to day 4 after the initiation of single-agent bevacizumab, whereas the cytokine levels in responders appeared to decrease during this period (Fig. 3).

Based on the antiangiogenic concept of cancer therapy, bevacizumab should be sustainably delivered to malignant tumors as it remains oxygenized microenvironment (29). However, the behavior of bevacizumab is a double-edged sword; it induces reoxygenation under the presence of proper vascular remodeling, otherwise it can lead to severe hypoxia because of destruction of the tumor vasculature (30). In this study, we found that one of the mechanisms of bevacizumab resistance is that refractory tumors are intrinsically hypoxic with ample vascularity. The FMISO-PET/CT scans showed that nonresponding tumors treated with bevacizumab exhibited significantly higher FMISO-SUV_max at baseline, and then sustained higher FMISO-SUV_max after the second course of chemotherapy than responding tumors. The DOSI scans also showed that nonresponding tumors sustained higher tumor tHb than responding tumors during treatment. In hypoxic environments, resistance to VEGF blockade could be associated with the activation of the hypoxia-inducible factor (HIF)-1a, which induces the release of hypoxia-mediated proangiogenic factors, such as VEGF, bFGF, and TGFβ, from tumor cells (32, 33). These factors collectively promote endothelial cell proliferation and migration. Although it appears that the DOSI data suggesting higher O₂Hb before therapy in nonresponders compared with responders were at variance with the FMISO data, the finding would be explained by the fact that a severe hypoxic tumor induces angiogenesis and increases arterial inflow (i.e., an increase in tumor O₂Hb level). Therefore, nonresponders barely sustained high SO₂ levels comparable with those in responders at pretherapy. We believe that tumor HHb level is substantially associated with hypoxic status because accumulation of HHb is considered a marker of impaired venous outflow and elevated interstitial fluid pressure. From our DOSI and cytokine analysis findings, we found that nonresponding tumors exhibit markedly decreased tumor SO₂ levels 1 to 3 days immediately after the initiation of bevacizumab and secrete higher levels of multiple proangiogenic cytokines than responding tumors. The DOSI scans also demonstrated that nonresponding tumors had transiently decreased concentrations of tumor tHb day 1 after the initiation of bevacizumab; however, the blood volume immediately recovered to the pretreatment level from day 3 to day 6. On the contrary, a sustained decrease in the concentration of tumor tHb was found in the responding tumors over this period. This result suggests that the resistance mechanism of tumor angiogenesis may be dependent on induction of proangiogenic factors. Our findings validate the previous notion that failure of vascular remodeling deconstructs the tumor microenvironment, leading to acute hypoxia and further angiogenic promotion. An interesting observation in our study is that the nonresponding tumors showed enhanced secretion of hypoxia-mediated inflammatory cytokines, such as IL1β, IL6, IL8, and TNFα, which decreased in the responding tumors. Given the fact that bevacizumab-induced acute hypoxia and the subsequent activation of HIF1α triggered the secretion of inflammatory cytokines from the tumor cells and immune cells, it is highly plausible for cancer cells to acquire a potential of EMT and metastasis to escape from the deteriorated microenvironment (33). In contrast, a fall in angiogenic/inflammatory cytokines could occur after bevacizumab-induced vascular normalization, resulting in oxygenated microenvironment and decreased edema and inflammation. Another intriguing finding of our study is that adding paclitaxel following bevacizumab treatment recovered the tissue oxygen levels similar to those before therapy. After the second course of paclitaxel combined with bevacizumab, drastically improved tumor SO₂ levels along with decreased tumor HHb were found in the nonresponding tumors. We hypothesize that paclitaxel itself has the potential of remodeling vasculature with reoxygenation regardless of off/on treatment of bevacizumab and that the antiangiogenic effect of paclitaxel is synergistic with the effects of bevacizumab even in nonresponders. Some patients may benefit from the synergetic effect of tissue reoxygenation by combining paclitaxel with bevacizumab due to improved drug delivery. In contrast, another explanation may be the following: Despite tissue reoxygenation of the tumor after the combined therapy, cancer cells may be still be hypoxic at the intracellular level and remain resistant to paclitaxel. This hypothesis may be supported by the discrepancy between the two results of increased SO₂ and higher FMISO uptake occurring in nonresponders that received the combined therapy. Because FMISO-PET and DOSI can measure intracellular hypoxia and tissue oxygenation, respectively, we need to carefully interpret the separate statuses of tumor hypoxia. Bevacizumab combined with cytotoxic chemotherapy is routinely used to treat metastatic breast cancer, but thus far, it has only demonstrated a limited benefit in terms of extending OS compared with chemotherapy alone. Our findings suggest that bevacizumab treatment may result in negative effects in some patients such as shortening survival by triggering hypoxia and promoting cancer progression (34, 35). It is unclear if nonresponding tumors can be treated with an antiangiogenic drug while at the same time avoiding unnecessary side effects for the patients. However, we suggest that the introduction of a PET-based response classification method, such as PET Response Criteria in Solid Tumors, may be a reasonable approach for early assessment of the therapeutic response (25, 26). The results of our PET scans showed that tumors with early metabolic response characteristically had lower levels of hypoxia than nonresponding tumors, and our findings are supported by several clinical studies, which used FDG-PET/CT scans to show that early metabolic response to chemotherapy is associated with the patients' prognosis and treatment strategy (36, 37). Therefore, it is evident that using only the FDG-PET imaging technique can already provide the critical information that reflects the cumulative effects of underlying biological parameters including glycolysis, angiogenesis, and hypoxia (38). FDG-PET imaging can thus play an essential role in managing the majority of cancer patients, minimizing efforts by the medical personnel and patients, and decreasing the costs associated with multiple procedures that have limited value for most patients.

Even if FDG-PET/CT is not available in routine practice, early tumor shrinkage (e.g., size reduction 1–2 months after the start of therapy) as assessed by conventional imaging could be used as a yardstick to determine whether treatment should be continued or not. The results of our MRI study suggested less-shrinkage tumors at early phase should be considered to potentially acquire intrinsic hypoxia. Several clinical studies have reported that dynamic and functional MRI could provide unique information on the characterization of tumor perfusion and hypoxia (39). Although hemodynamic measurement using DOSI is a promising method of noninvasively monitoring tissue oxygen levels, it is only applicable to surface lesions. DOSI can measure the change in tissue concentrations of O₂Hb and HHb during a short period of time, which were translated to the dynamics of blood flow. The results of our DOSI study contribute to the literature on the interpretation of blood flow dynamics induced by bevacizumab. Figure 4 presents a hypothetical illustration showing the relationships among angiogenesis, hypoxia, and glycolysis. Thus, the DOSI technique could be clinically useful to understanding the physiologic mechanism of vascular remodeling in breast tumors.

There were limitations to this study. To begin with, this was a small clinical study including 18 patients who underwent surgery and 10 patients who continued chemotherapy after the study, and a limited number of the patients had postsurgical histopathology. Twelve patients received prior endocrine therapy before participating in the study. Although no association between therapeutic response and history of prior endocrine therapy was found, we did not perform a new biopsy before bevacizumab treatment to confirm hormonal receptors and HER2 statuses. Some physiologic and genomic changes after endocrine therapy could have affected the study results. What is the best timing for PET scans remains unsettled. Although many investigators adopted response assessment after one or two cycles of chemotherapy or targeted therapy (27–29) in clinical studies, it is unclear whether hampered drug delivery after anti-VEGF therapy could result in a false-negative response by FDG uptake or not. An absolute need exists for more subtype-specific (luminal breast cancer and TNBC) PET assessment. Our data on the baseline relation between glycolysis and hypoxia suggest that therapeutic response in TNBC is linked more to hypoxia and glycolysis than it is in luminal breast cancer (shown as Supplementary Fig. S2). Finally, this study raises more questions about the significance of different methods for assessing tumor hypoxia. We need to recognize the diverse results obtained from multimodal images, such as those obtained using MRI, PET, and DOSI, which should be carefully interpreted by physicians and integrated into practice.

In conclusion, our imaging research in this study has provided new mechanistic insights on bevacizumab-induced hypoxia and cancer progression. We revealed that bevacizumab-resistant breast cancer features severe hypoxia along with a high degree of angiogenesis. Thus, the functional imaging technologies utilized in the present study will likely be used to measure important in vivo biomarkers in future, which complement conventional tissue sampling methods by allowing molecular and physiologic characterization of tumors and the assessment of treatment response.

No potential conflicts of interest were disclosed.

Conception and design: S. Ueda, T. Saeki, A. Osaki

Development of methodology: S. Ueda, A. Osaki

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Ueda, T. Yamane, I. Kuji

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Ueda

Writing, review, and/or revision of the manuscript: S. Ueda, T. Saeki, T. Yamane, I. Kuji

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Yamane

Study supervision: T. Saeki, A. Osaki, I. Kuji

Other (research fund): T. Saeki

We are grateful to all the patients who participated in this study. The authors thank Yukio Ueda, Yutaka Yamashita, Hiroyuki Suzuki, and Kenji Yoshimoto for their technical support regarding the TRS20 device (Hamamatsu K.K.); Noriko Wakui and Midori Nakajima for their help in optical measurement; and Hiroko Shimada, Ikuko Sugitani, and Eiko Hirokawa for their contribution to patient enrollment.

The research was supported by KAKEN 26282144, 22591351, 16K10293, 16K10361, Japan Research Foundation for Clinical Pharmacology, 2016 Hidaka Research Project.

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.