Overexpression of Tissue VEGF-A May Portend an Increased Likelihood of Progression in a Phase II Trial of Bevacizumab and Erlotinib in Resistant Ovarian Cancer

Abstract

Purpose

This phase II trial evaluated bevacizumab plus erlotinib in platinum-resistant ovarian cancer; exploratory biomarker analyses, including that of tumor vascular endothelial growth factor (VEGF-A), were also performed.

Experimental Design

Forty heavily pretreated patients received erlotinib 150 mg/day orally and bevacizumab 10 mg/kg intravenously every 2 weeks until disease progression. Primary end points were objective response rate and response duration; secondary end points included progression-free survival (PFS), toxicity, and correlations between angiogenic protein levels, toxicity, and efficacy.

Results

Grade 3 toxicities included skin rash (n=6), diarrhea (n=5), fatigue (n=4), and hypertension (n=3). Grade 4 toxicities were myocardial infarction (n=1) and nasal septal perforation (n=1). Only 1 grade 3 fistula and 1 grade 2 bowel perforation were observed. Nine (23.1%) of 39 evaluable patients had a response (median duration 36.1+ weeks, 1 complete response), and 10 (25.6%) patients achieved stable disease, for a disease control rate of 49%. Median PFS was 4 months, and 6-month PFS was 30.8%. Biomarker analyses identified an association between tumor cell VEGF-A expression and progression (P=0.03); for every 100 unit increase in the VEGF-A score, there was a 3.7-fold increase in the odds of progression (95% CI: 1.1–16.6).

Conclusions

Bevacizumab plus erlotinib in heavily pretreated ovarian cancer patients was clinically active and well tolerated. Erlotinib did not appear to contribute to efficacy. Our study raises the intriguing possibility that high levels of tumor cell VEGF-A, capable of both autocrine and paracrine interactions, are associated with resistance to bevacizumab, emphasizing the complexity of the tumor microenvironment.

INTRODUCTION

Although approximately 70% of patients with newly diagnosed advanced ovarian cancer respond to standard platinum- and taxane-based chemotherapy regimens, the majority of patients experience disease recurrence (1). For recurrent ovarian cancer, current treatment regimens containing topotecan or liposomal doxorubicin have limited efficacy, especially in patients with platinum-resistant disease for whom response rates have ranged from 10% to 20% (2). Development of targeted therapies is aimed at improving these outcomes.

In the stroma, vascular endothelial growth factor (VEGF-A), by binding to the VEGF receptors (VEGFR), plays a central role in mediating the formation and differentiation of tumor-associated vasculature (3–5). In the tumor cell, the expression of VEGF has the potential to act in a paracrine manner to promote angiogenesis through this mechanism. More recent attention has been focused on autocrine pathways of activated VEGF signaling within ovarian cancer cells, which may contribute to cancer behavior and outcome independent of an effect on angiogenesis (6, 7). In fact, in a mouse model, VEGF-A expression was shown to modulate resistance to cisplatin via an autocrine mechanism (8).

Studies examining the role of VEGF-A from tissue or biological fluids as a prognosticator in ovarian cancer show conflicting results (9–12). Circulating VEGF-A levels, which receive contributions from both endothelial and tumor cell compartments, were not found to be a predictive biomarker for bevacizumab response in ovarian cancer (13, 14).

Recent immunohistochemical evaluation of VEGF-A in ovarian cancer cells has shown strong expression in a minority of cases (7%–13%), with this staining correlated with a poor prognosis (15, 16). The expression of VEGFR-1 and -2 within ovarian carcinomas have also been shown to be higher than levels within normal ovarian tissue, suggesting that anti-VEGF therapies may have direct antitumor activity in addition to suppressing angiogenic mechanisms that sustain ovarian tumor growth (6, 17–19).

Bevacizumab (Avastin, South San Francisco, CA), a humanized monoclonal antibody against VEGF-A, is approved for the treatment of a number of tumor types on the basis of improved progression-free survival (PFS) and/or overall survival (OS) outcomes (20–23). The clinical activity of bevacizumab as a single agent is notably greater in ovarian cancer than in most other cancers; response rates of 16% and 21% have been reported in phase II clinical trials in patients with refractory ovarian cancer, including those with platinum-resistant disease (24, 25). Response rates of up to 78% have also been achieved with combinations of bevacizumab and chemotherapy in patients with recurrent platinum-sensitive or platinum-resistant ovarian cancer (13, 26, 27). Bevacizumab is generally well tolerated in patients with advanced ovarian cancer (13, 25). In heavily pretreated patients with refractory disease, the use of bevacizumab was originally associated with high rates of gastrointestinal perforations (GIPs) (10, 14, 28), although recent reports suggest a GIP rate no greater than that seen in the control ovarian cancer population of <7% (29, 30).

Epidermal growth factor receptors (EGFRs) have been widely implicated in the development and progression of cancer, correlate with advanced-stage disease and poor survival, and are overexpressed in approximately 70% of ovarian tumors (31, 32). Erlotinib (Tarceva, South San Francisco, CA) is an orally available small-molecule EGFR tyrosine kinase inhibitor (33, 34). Single-agent erlotinib has demonstrated little activity in ovarian cancer, although 44% of patients in one phase II study experienced stable disease [SD] (35).

Significant crosstalk occurs between the VEGF and EGFR signaling pathways, providing a compelling rationale for combining agents that separately target these pathways (36). EGFR regulates VEGF via the PI3 kinase dependent pathway, and cells that develop resistance to EGFR inhibition upregulate both VEGFR-1 and VEGF (37, 38). In ovarian cancer, mechanisms for resistance to EGFR inhibitors have yet to be fully defined (31). K-Ras mutations, which are associated with resistance to EGFR directed therapies in colon cancer, are rare in ovarian tumors, with the exception of transitional or mucinous tumors (39).

Resistance to the actions of anti-angiogenic therapy, including bevacizumab, has as its basis both tumor and host (tumor microenvironment) mediated pathways (recently reviewed by Ebos et al [40]). In light of the complexity of these networks and level of cross-talk, it is unlikely that expression of a single targeted molecule such as VEGF-A could be a reliable predictive marker for bevacizumab resistance or response.

Combination therapies targeting both EGFR and VEGF pathways have been studied in other cancers and, after initiation of our trial, one small trial in ovarian cancer (14). Based on this rationale, we initiated a nonrandomized phase II trial evaluating the safety and efficacy of erlotinib and bevacizumab in women with heavily pretreated platinum-resistant ovarian or primary peritoneal cancer.

METHODS

Patient eligibility

Patients with histologically or pathologically confirmed epithelial carcinoma of the ovary or primary peritoneal carcinoma were eligible if their disease was refractory to primary treatment, they had relapsed within 6 months of completing treatment with taxane- and platinum-based therapies, or they developed platinum-resistant disease. Patients were required to have elevated CA-125 levels (>2 times the institutional upper limit of normal [ULN]) or measurable disease. In addition to the primary chemotherapy and those for platinum-sensitive disease, 2 other chemotherapy regimens were allowed. Prior hormonal or radiation therapy was allowed. Debulking surgery for relapsed disease must have been completed at least 28 days prior to the first day of study therapy.

Additional inclusion criteria were a Zubrod performance status of 0 or 1, adequate hepatic function (serum bilirubin ≤1.5 times the ULN), serum glutamic oxaloacetic transaminase or serum glutamic pyruvic transaminase (≤2.5 times the ULN), adequate renal function (serum creatinine ≤1.5 times the ULN; urine protein:creatinine ratio <1.0 at screening or urine dipstick for proteinuria <2+), hemoglobin ≥10 mg/dL, white blood cell count ≥2500/µL, and platelets ≥80,000/µL.

Exclusion criteria included mixed Müllerian tumors; low malignant potential; major surgical procedure within 4 weeks of starting therapy (or within 6 weeks for high-risk procedures); bowel obstruction or short bowel syndrome; treatment with chemotherapy, biologic therapy, or any other investigational drug within 28 days prior to the start of therapy (or the presence of a related grade ≥1 adverse event other than alopecia); pregnancy; or lactating.

Additional exclusion criteria related to bevacizumab-specific concerns were inadequately controlled hypertension (systolic blood pressure ≥150 mm Hg and/or diastolic blood pressure ≥100 mm Hg on antihypertensive medications); history of hypertensive crisis or hypertensive encephalopathy; New York Heart Association grade II or greater congestive heart failure; history of myocardial infarction, cerebrovascular accident, transient ischemic attack or unstable angina within 6 months of study enrollment; uncontrollable nausea; clinically significant peripheral vascular disease; evidence of bleeding diathesis or coagulopathy; presence of central nervous system or brain metastases; proteinuria; history of abdominal fistula; GIPs or intra-abdominal abscesses; serious nonhealing wound, ulcer, or bone fracture; diagnosis of any other malignancy in the past 5 years except nonmelanomatous skin cancer; and known hypersensitivity to any component of bevacizumab.

All patients gave informed consent before study enrollment. The study was approved by the Institutional Review Board at the Arizona Cancer Center, University of Arizona, and was conducted in accordance with institutional and federal guidelines.

Study Design and Treatment

The study was a nonrandomized, open-label, single-center phase II trial using a Simon two-stage design. The primary objective of the study was to evaluate the objective response rate (ORR; confirmed complete response [CR] + partial response [PR]) and response duration. Secondary end points were PFS; toxicity; and correlations between toxicity (diarrhea, rash, hypertension), angiogenic protein levels (tumor VEGF-A and VEGFR-1 and urinary VEGF-A), and efficacy (ORR, progression rate, PFS, and 6-month PFS rate).

Patients were treated with erlotinib 150 mg/day orally and bevacizumab 10 mg/kg by intravenous infusion every 2 weeks (±1 day) until disease progression, intolerable toxicity, or physician/patient decision.

Dose reductions of bevacizumab were not permitted. For predetermined bevacizumab-related toxicities, including grade 3 proteinuria, treatment was held until improvement in clinical parameters. Bevacizumab was discontinued for uncontrolled grade ≥3 hypertension, grade ≥2 hemorrhage, symptomatic grade 4 venous thrombosis, an arterial thromboembolic event of any grade, grade 4 congestive heart failure, grade 4 proteinuria, grade 4 nasal septal perforation, grade 4 fistulae, and wound dehiscence requiring medical or surgical therapy.

Erlotinib was held for persistent grade 2 diarrhea, grade 3 diarrhea, or intolerable grade ≤3 rash, then restarted at a reduced dose of 100 mg/day after resolution of the adverse event to grade ≤1. Regardless of the reason for holding study drug treatment, the maximum allowable length of treatment interruption was 6 weeks. A second dose reduction to 50 mg/day was permitted, but erlotinib was discontinued if patients were unable to tolerate this lower dose. Erlotinib treatment was discontinued for grade 4 diarrhea or rash. Topical treatments (eg, clindamycin cream, triamcinolone ointment, and other over-the counter herbal ointments) and oral medications (eg, doxycycline, loratadine) were used to ameliorate symptoms of erlotinib-related rashes.

Patient Evaluation

Patients were evaluated every 4 weeks, by assessment of clinical symptoms, and physical including pelvic examinations. Additionally, serial serum CA-125 levels were evaluated when they were deemed to be an appropriate tumor marker for the individual patient. If there was any suspicion for disease progression on the basis of any of these factors, then computed tomography imaging of the chest, abdomen, and pelvis were obtained at that time. Radiological evaluations of disease were therefore based on standard of care provided to each subject. In the absence of suspicion for progression, then the computed tomography imaging was obtained at least every 3 months. Responses were assessed using the Response Evaluation Criteria in Solid Tumors (RECIST) by physical examination and/or computed tomography imaging of the chest, abdomen, and pelvis. Toxicity was measured according to the Common Terminology Criteria for Adverse Events, version 3.0.

Determination of Angiogenic Protein Levels

VEGF-A and VEGFR-1 expression levels were scored within the tumor cells in the ovarian carcinoma samples obtained from the primary surgery. A study of VEGFR-2 expression was also attempted; the staining was so minimal, that it could not be reliably scored. Blocks for 36 of 40 (90%) cases contained sufficient tumor and demonstrated immunohistochemical staining suitable for scoring of VEGF-A and VEGFR-1. K-Ras mutations were not measured in these tumors since there were no mucinous or transitional histologies among the 40 patients in this study. Urine VEGF-A levels were obtained at baseline for 27 (67.5%) cases.

Immunohistochemistry (IHC)

Hematoxylin and eosin (H&E) stains were performed on three micron sections of tissue cut from the formalin fixed, paraffin-embedded blocks. IHC was performed using VEGF-A mouse monoclonal antibody (Santa Cruz Biotech #sc-7269) diluted 1:70, and VEGFR-1 rabbit polyclonal antibody (NeoMarkers LabVision #RB1527) diluted 1:20. Tissue sections were stained on a Discovery XT Automated Immunostainer (Ventana Medical Systems, Inc. [VMSI], Tucson, Arizona). All steps were performed on this instrument using VMSI validated reagents, including deparaffinization, cell conditioning (antigen retrieval with a borate-EDTA buffer), primary antibody staining, detection and amplification using a biotinylated-streptavidin-Horseradish Peroxidase and Diaminobenzidine (DAB) system and hematoxylin counterstaining. VEGF-A was detected using a goat anti-rabbit secondary antibody and an UltraMap DAB detection kit (VMSI). Images were captured using a Paxcam 3 camera with PAX-it Digital Image Management and Image Analysis (Midwest Information Systems, Inc., Villa Park, IL), and standardized for light intensity. Each case was evaluated and given a semiquantitative histologic score (RBN, pathologist) using a previously described method (42). Briefly, staining intensity for neoplastic cells was scored as: 0 negative, 1 weak, 2 moderate, and 3 intense. In addition, the percentage of positive neoplastic cells was evaluated for each intensity. The overall scores were calculated by multiplying the intensity by the corresponding percentage of positive cells, resulting in values ranging from 0 to 300.

Enzyme-linked Immunosorbent Assay

A spot urine sample was collected at baseline for 27 cases, and stored at −80°C until batch analyzed. Urine VEGF-A from centrifuged samples was quantitated using the QuantiGlo system (R&D Systems, Minneapolis, MN) and normalized to creatinine, which was measured using the Creatinine assay kit (Cayman Chemicals, Ann Arbor, MI). Results were calculated by dividing VEGF (pg) by creatinine (mg), and performed in triplicate.

Statistical Analysis

The study was designed to enroll a maximum of 40 patients, with an initial accrual of 20 patients. If ≤1 confirmed responses were observed in the first 20 patients (stage I), the trial was to be closed and the agents were to be considered inactive. If ≥2 responses were observed, 20 additional eligible patients were to be accrued in stage II. Eight or more responses (20%) in 40 patients were to be considered evidence that the combination warrants further study, provided PFS appeared favorable. This design had a significance level (probability of falsely declaring an agent with a 10% true response probability to be one warranting further study) of 0.04 and a power (probability of correctly declaring an agent with a 30% response probability to warrant further study) of 0.94.

PFS was defined as the time from the start of therapy to the time of the first documentation of progression, symptomatic deterioration, or death due to any cause; PFS was estimated using the Kaplan-Meier method, and differences between groups were tested using the log-rank test. Exploratory analyses compared the toxicity levels and patient outcomes. Fisher’s exact test was employed to measure differences in progression rate (versus stable disease or response) and 6-month PFS rate by toxicity levels, and the association between CA-125 response/progression and response/progression by RECIST. Biomarker analyses were also exploratory. Associations between the expression of tumor VEGF-A and VEGFR-1, as well as urinary VEGF-A and toxicity and efficacy outcomes, were assessed using linear regression for the toxicity categories, logistic regression for progression and 6-month PFS rate, and the Cox proportional hazards model for PFS. Progression rate, rather than ORR, was studied in association with these biomarkers because of sample size considerations. All statistical tests were two-sided.

RESULTS

Patients

Fifty-six patients were deemed eligible for the study and provided consent. A total of 40 patients were treated with bevacizumab and erlotinib after the criteria for proceeding to stage II were met. Thirty-two patients were eligible for the study based on measureable disease, with only 8 patients eligible on CA125 criteria alone. Patients did not receive study treatment because of the following reasons: failed to meet chemotherapy criteria (n=5), failed to meet performance criteria (n=3), had recent or active small bowel obstruction (n=2), were ineligible owing to pathology (n=2), death (n=1), failed to meet resistance criteria (n=1), brain metastases (n=1), or decreasing CA-125 levels (n=1).

Baseline characteristics are shown in Table 1. The median age of patients was 61.0 years (range, 31.9–77.5), and 80% of patients presented with International Federation of Gynecology and Obstetrics (FIGO) stage III disease. Patients were heavily pretreated (median of three prior regimens; range, 1–7), with 80% having received ≥2 prior regimens. Seven patients had a history of bowel resections with or without bowel obstruction, ostomies, or fistulas.

Table 1.

Patient and Disease Characteristics

Characteristic	Patients (N=40)
Median age, years (range)	61.0 (31.9–77.5)
Baseline Zubrod performance status, no. (%)
0	21 (52.5)
1	19 (47.5)
Race, no. (%)
White	39 (97.5)
American Indian	1 (2.5)
FIGO stage, no. (%)
II	3 (7.5)
III	32 (80.0)
IV	5 (12.5)
Cell type, no. (%)
Serous	26 (65.0)
Endometrioid	6 (15.0)
Clear cell	4 (10.0)
Epithelial (not otherwise characterized or mixed)	4 (10.0)
Tumor grade, no. (%)
1	1 (2.5)
2	9 (22.5)
3	22 (55.0)
High-grade	5 (12.5)
Unknown	3 (7.5)
Primary site, no. (%)
Ovarian	32 (80.0)
Peritoneal	8 (20.0)
Prior chemotherapeutic regimens; no. (%)
Median (range)	3 (1–7)
1	8 (20.0)
2	11 (27.5)
3	13 (32.5)
4–5	7 (17.55)
6–7	1 (2.55)
Prior pelvic radiation, no. (%)	2 (5)
Prior hormonal therapies, no. (%)	6 (15)

FIGO, International Federation of Gynecology and Obstetrics.

Treatment Delivery, Modifications, and Discontinuations

Patients were treated with a median of 3.5 cycles of therapy (range, 1.5–26). Thirty-four treatment cycles of Bevacizumab were held for 15 patients (range, 1–6 weeks). Reasons included: hypertension, non-compliance, patient choice, and development of a GI fistula which subsequently closed on restarting Bevacizumab. Dose reductions of erlotinib were required in 14 patients: to 100 mg (n=7), to 75 mg (n=1), to 50 mg (n=4), and discontinued (n=2). Treatment was discontinued in 39 patients for: disease progression (n=30), adverse events (n=6), patient or physician decision to withdraw (n=2), and completion of protocol (maximum 24 months of treatment; n=1). Adverse events leading to treatment discontinuation were grade 3 or 4 nasal septal perforation (n=2), unconfirmed intracranial bleed (n=1), myocardial infarction (n=1), bowel obstruction (n=1), and GIP (n=1). One patient died while on study as a result of complications related to disease progression.

Safety

All patients were assessed for safety, and the majority of adverse events reported were grade 1 or 2 (Table 2). The most common treatment-related grade 3 events were skin rash (n=6, 15.0%), diarrhea (n=5, 12.5%), fatigue (n=4, 10.0%), and hypertension (n=3, 7.5%). Two patients (5.0%) had grade 4 treatment-related adverse events (myocardial infarction and nasal septal perforation, respectively). No treatment-related deaths occurred. There was only one grade 3 fistula and one grade 2 bowel perforation (microperforation found on computed tomography scan). In addition there was one grade 3 nasal septal perforation.

Table 2.

Treatment-Related Toxicity, by Grade (N=40)

	Patients With Event, no. (%)
Adverse event	Grade 1	Grade 2	Grade 3	Grade 4
Hematologic
Anemia	1 (2.5)	0 (0)	1 (2.5)	0 (0)
Neutropenia	1 (2.5)	0 (0)	0 (0)	0 (0)
Thrombocytopenia	1 (2.5)	0 (0)	0 (0)	0 (0)
Gastrointestinal
Nausea	8 (20.0)	1 (2.5)	0 (0)	0 (0)
Vomiting	3 (7.5)	3 (7.5)	0 (0)	0 (0)
Constipation	1 (2.5)	0 (0)	0 (0)	0 (0)
Diarrhea	23 (57.5)	3 (7.5)	5 (12.5)	0 (0)
Bowel perforation	0 (0)	1 (2.5)	0 (0)	0 (0)
Fistula	0 (0)	0 (0)	1 (2.5)	0 (0)
Dehydration	0 (0)	0 (0)	1 (2.5)	0 (0)
Cardiovascular
Hypertension	0 (0)	4 (10.0)	3 (7.5)	0 (0)
Embolism	0 (0)	0 (0)	0 (0)	0 (0)
Myocardial infarction	0 (0)	0 (0)	0 (0)	1 (2.5)
Thrombosis	0 (0)	0 (0)	0 (0)	0 (0)
Nonhematologic
Rash	15 (37.5)	13 (32.5)	6 (15.0)	0 (0)
Fatigue	1 (2.5)	7 (17.5	4 (10.0)	0 (0)
Proteinuria	3 (7.5)	1 (2.5)	0 (0)	0 (0)
Septal perforation	0 (0)	0 (0)	1 (2.5)	1 (2.5)

Efficacy

One of the 40 enrolled patients was not assessable for efficacy because of treatment discontinuation shortly after consent. In the 39 remaining patients, the ORR was 23.1% (n=9), with one patient (2.6%) experiencing a CR (Table 3). An additional 25.6% (n=10) of patients achieved SD for a disease control rate of 48.7%.

Table 3.

Efficacy Outcomes

Measure	Outcome (n=39)
Response
Best response, no. (%)
Complete response	1 (2.6)
Partial response	8 (20.5)
Stable disease	10 (25.6)
Progression	20 (51.3)
Objective response rate, %	23.1
Disease control rate, %	48.7
Duration of response, weeks (range)
Patients with a response (n=9)	36.1+ (5.6–83.0)
Patients with stable disease (n=10)	25.6+ (6.0–65.0)
Progression-free survival
6-month progression-free survival rate, %	30.8
Median progression-free survival, months (range)	4 (1–26.8)

The median duration of response was 36.1+ weeks (range, 5.6–83) for responders and 25.6+ weeks (range, 6–65) for those with SD. The median duration of response for responders and those with stable disease combined was 29.7+ weeks. Twelve (30.8%) patients were progression-free at 6 months. The median PFS was 4 months (range, 1–26.8) for the overall population (see Table 3 and Figure 1).

Figure. 1.

Progression-free survival for all evaluable patients (n = 39). Censored observations are designated with a “+” at the time of censoring.

Biomarkers

Toxicity as a marker of clinical outcome was studied. No significant association was found between toxicity (diarrhea, rash, or hypertension) and median PFS, 6-month PFS rate, and progression rate (Supplementary Table 1), as tested using the log-rank test and Fisher’s Exact Test, respectively. The potential role of urinary VEGF-A as a marker of toxicity or efficacy was also studied; however, no significance was found between urinary VEGF-A and toxicity grade, or the efficacy parameters (not shown).

Tissue biomarkers were examined as potential indicators of toxicity or efficacy. No association was seen between tumor VEGF-A or VEGFR-1, and toxicity grade (Supplementary Table 2), as tested using linear regression. There was no association between VEGFR-1 expression and any of the clinical parameters. However, when exploring tissue VEGF-A’s association with clinical efficacy end points, an interesting and significant association between tumor VEGF-A level and progression (P=0.03; Table 4) was observed, as tested using logistic regression. For every 100-unit increase in VEGF-A score, there was a 3.7-fold increase in the odds of progression (95% CI: 1.1–16.6). The mean VEGF-A score (95% CI) for 20 patients who progressed versus the 10 stable patients versus the 9 patients with response were 129 (94, 163), 86 (40, 132), 77 (35, 119), respectively. In contrast, there was no statistically significant association (P=0.16) between tumor VEGFR-1 and odds of progression (Table 4). The mean VEGFR-1 score (95% CI) for those who progressed versus those with stable disease versus responders were 119 (86, 151), 158 (115, 202), 137 (97, 177), respectively.

Table 4.

Association Between Tissue VEGFR-1 and VEGF-A Expression Level and Progression Rate

Biomarker	P	50 Unit	100 Unit
		Odds ratio (95% CI)	Odds ratio (95% CI)
VEGFR-1	0.16	0.636 (0.293–1.190)	0.404 (0.086–1.416)
VEGF-A	0.03	1.918 (1.049–4.068)	3.680 (1.101–16.550)

VEGF-A, vascular endothelial growth factor; VEGFR-1, VEGF receptor 1.

Progression Rate, percent with progression as best response.

Figure 2 depicts IHC staining of VEGF-A in tumor epithelium from a representative case with stable as compared to progressive disease. There was no significant association of tissue VEGF-A levels with PFS or 6-month PFS (not shown), as tested using a Cox proportional hazards model and logistic regression model, respectively. These associations were not statistically significant, likely due to the small sample sizes.

Figure. 2.

(A) Immunohistochemical staining for VEGF-A (score 70) and (B) H&E staining in a patient with stable disease following study treatment. (C) VEGF-A staining (score 140) and (D) H&E staining in a patient with disease progression following study treatment. All photographs are 10×.

DISCUSSION

In this nonrandomized phase II trial in heavily pretreated patients with platinum- and taxane-resistant ovarian and primary peritoneal cancer, the efficacy of bevacizumab and erlotinib (ORR, 23.1%; duration of response, >8 months; median PFS, 4 months; 6-month PFS, 31%) was comparable to that previously reported for single-agent bevacizumab in similar patients with refractory ovarian cancer (24, 25) , and on the low end when compared to bevacizumab combined with chemotherapy (13, 26, 43–45). A recently reported phase II trial by Nimeiri and colleagues that similarly evaluated bevacizumab and erlotinib in patients with recurrent ovarian cancer reported responses in two of 13 patients (15%) (14).

Given the comparability of the efficacy results with those reported with single-agent bevacizumab in a similar population of patients, it is unlikely that erlotinib added substantially to the clinical benefit of bevacizumab in our study. Erlotinib did not add to the efficacy of bevacizumab in renal cell cancer (46), and a recent report in colon cancer found the addition of the EGFR inhibitor cetuximab to the combination of bevacizumab and chemotherapy resulted in a significantly worse PFS and a decrease in quality of life compared with the combination of bevacizumab and chemotherapy alone (47). In our study, the use of erlotinib was associated with increased toxicity. The most common grade ≥3 adverse events were rash (15%, all grade 3) and diarrhea (12.5%, all grade 3), both of which were attributed to erlotinib therapy and were reversed by discontinuing or reducing the dose of erlotinib. Similar rates of grade ≥3 diarrhea (15%) and nausea (15%) were observed in the Nimeiri study (14). In our study, toxicities of rash, diarrhea, or hypertension did not correlate with clinical outcome. This is in contrast to the report of bevacizumab and erlotinib in non-small cell lung cancer that showed rash correlated with enhanced survival (48). Our limited search for urinary and tissue biomarkers of toxicity in this study found no correlation.

Compared with trials in other tumor types, higher rates of bevacizumab-related GIPs were originally reported in several ovarian cancer trials (14, 24, 28, 29). In contrast, in our trial population, which included many patients who received long-term treatment, we observed only one grade 3 bowel fistula and one grade 2 bowel perforation. The rate of GIPs in our study (5%) is comparable to that observed in a recent retrospective cohort study of patients with recurrent ovarian cancer, which found no increased risk for GIPs or fistulas in patients receiving bevacizumab compared with those receiving standard chemotherapy (7.2% versus 6.5%, respectively) (30). Our finding is also notable in that erlotinib was recently reported in a May 2009 FDA alert to increase the risk of bowel perforations.

Individualizing cancer treatment by using molecular markers to aid in the selection of a chemotherapy agent is gaining wide acceptance. However, it is increasingly apparent that prognostic biomarkers are not always predictive of response to the corresponding targeted therapy. Furthermore, expression of a single molecule or those representing a single molecular pathway is unlikely to be a reliable predictor for treatment efficacy.

In the present study, no significant associations between efficacy outcomes (median PFS, 6-month PFS rate, or progression rate) and tissue VEGFR-1 expression, or urinary VEGF-A expression were detected. Unexpectedly, we found a significant association between elevated tissue VEGF-A levels and higher odds of progression. This is the first study to raise the intriguing possibility that elevated tissue VEGF-A levels in the tumor microenvironment of ovarian cancer are associated with resistance to bevacizumab-containing regimens. Our analysis is exploratory and the finding limited by small numbers, thus this needs to be validated in larger studies with sufficient power to detect a difference in PFS. The large majority of specimens expressed VEGF-A, at varying levels. These data suggest that the presence of VEGF-A in the ovarian cancers alone is not a good biomarker for prediction of response to bevacizumab. It is highly likely that tumors overexpressing VEGF-A were also resistant to erlotinib (38). In line with our finding, in a trial of bevacizumab and cytotoxic chemotherapy in advanced breast cancer, low plasma VEGF levels predicted prolonged time to progression (49). While blood VEGF levels derive contributions from different tissue compartments not restricted to the cancer cell, this datum in addition to ours suggests that the biology of VEGF signaling is complex, reflecting both paracrine and autocrine interactions. It is probable that high VEGF-A reflects activation of upstream factors, like hypoxia-inducible factor 1α, that contribute to aggressive tumor growth by a variety of mechanisms in addition to VEGF-A. To date, there are no validated markers for response to antiangiogenic therapy in cancer (50). In a mouse model of ovarian cancer, the addition of low-dose paclitaxel to a VEGFR-2 inhibitor yielded additive effects only with tumors expressing low levels of VEGF. In contrast in high-VEGF-expressing tumors, there was antagonism observed between the two agents (51). We interpret the results from the present study to mean that ovarian tumors that express high levels of VEGF-A may be resistant to bevacizumab and erlotinib, at least in the doses and manner given in this study. On the other hand, when the tumors express a more moderate level of VEGF-A, this regimen can result in either stabilization of growth or tumor response.

In conclusion, this phase II study of bevacizumab and erlotinib in patients with drug-resistant ovarian cancer demonstrated that response rates, duration of response, and PFS for this regimen are comparable to previous trials of single-agent bevacizumab in patients with recurrent ovarian cancer. Despite the majority of patients being heavily pretreated, there were only two bowel perforations or fistulas, and neither of these events were grade 4. The majority of the common toxicities resulted from erlotinib, which did not appear to add to the efficacy of this regimen. This study suggests for the first time that high tumor VEGF-A levels in ovarian cancer may correlate with enhanced odds of progression. This intriguing but preliminary finding emphasizes the complexity of the tumor microenvironment and the interacting networks that govern tumor resistance to targeted therapy. While these results need to be validated, caution should be exercised when using tumor VEGF-A expression alone in ovarian cancer as a surrogate for likelihood of response to VEGF inhibitor therapy.

Statement of Translational Relevance

This article demonstrates that the combination of bevacizumab and erlotinib in platinum-resistant ovarian cancer patients is well tolerated and has efficacy similar to bevacizumab alone. Exploratory analyses, which require validation, suggest that elevated tumor cell VEGF-A expression levels may be associated with an enhanced odds of progression. This intriguing but preliminary finding emphasizes the complexity of the tumor microenvironment and the interacting networks, which govern tumor resistance to target therapy. Caution should be exercised when using tumor VEGF-A expression alone in ovarian cancer as a surrogate for likelihood of response to VEGF inhibitor therapy.