Angiogenesis-related protein expression in bevacizumab-treated metastatic colorectal cancer: NOTCH1 detrimental to overall survival

The medical data from CRC metastatic patients treated with a combination of bevacizumab and chemotherapy were retrospectively analyzed. All data from May 2006 to November 2009 were obtained from chemotherapy registries from A.C. Camargo Cancer Center - Fundação Antônio Prudente. A total of 151 patients were found. Paraffin blocks were available for 117 patients, but in 11 of them the data or material were not suitable for analysis, therefore, a total of 105 patients were eligible for further study (Fig. 1). The study was approved by the Ethics Committee of A.C. Camargo Cancer Center - Fundação Antônio Prudente (number 1134/08). The need of an informed consent was waived by Ethics Committee.

Demographic and clinical data were collected and included relevant medical history, disease stage, tumor pathology at the initial diagnosis, details of chemotherapy regimens used with bevacizumab, data concerning the primary surgery, and metastasectomy. Tumors were staged according to the 7th edition of the TNM classification of malignant tumors [19]. Data for the response criteria and survival outcomes were based on the chart review. Response criteria were evaluated according to the Response Evaluation Criteria in Solid Tumors version 1.1 (RECIST v1.1) [20]. Cases treated before the publication of RECIST v1.1 were classified accordingly based on data extracted from image reports. The PFS was defined as the time from the beginning of bevacizumab treatment until the first observation of disease progression. The OS was defined as the time from the beginning of bevacizumab treatment until the date of either the last contact alive or death from any cause. Hypertension was defined when blood pressure was either persistently elevated (>24 h) or repeated blood arterial pressure measurements were above 140 × 90 mmHg. The Common Terminology Criteria for Adverse Events (CTCAE) version 4.03 was used to grade hypertension [21]. Blood pressure values recorded by physicians and nurses during screening consultations, pre-chemotherapy, and prior to initiation of the antihypertensive treatment were analyzed. Archival pathological specimens were collected, and the expression of the angiogenesis-related proteins was first verified by immunohistochemistry and then quantified.

Archival formalin-fixed, paraffin-embedded specimens were obtained from patients who received bevacizumab. The chosen samples were from primary tumor and/or metastatic tissue collected as close as possible to the start of bevacizumab treatment. The paraffin blocks underwent tissue microarray construction before immunostaining. In brief, a fresh section stained with hematoxylin and eosin was cut from each block. The representative tumor areas were carefully selected and marked. Using a tissue microarrayer (Beecher Instruments, Silver Spring, MD, USA), 1 mm cylindrical cores were removed from each donor paraffin block and transferred to premolded recipient paraffin blocks, in duplicates. Sections 5 μm in thickness were placed on glass slides. In the recipient block, cores were arrayed according to the defined x-y coordinate position. Normal placenta tissue cores were used as a position marker. Slides were then incubated with the primary antibodies according to the manufacturer’s protocol. The polyclonal antibodies used in this study were: PlGF (1:20, R&D systems, Minneapolis, MN, USA), VEGFR2 (1:50, Neomarkers, Freemont, CA, USA), VEGFR3 (1:400, LabVision, Freemont, CA, USA), and DLL4 (1:200, Abcam, Cambridge, UK). The monoclonal antibodies used were VEGFR1 (1:50, clone Y103, Abcam, Cambridge, UK) and NOTCH1 (1:50, Thermo Scientific, clone A6, Rockford, IL, USA). Antibody detection was performed using a streptavidin-biotin system (Biotinylated Link Universal, LSAB+, Carpinteria, CA, USA) for PlGF and a biotin-free polymeric visualization system (Novolink Max Polymer, Carpinteria, CA, USA) for all the other antibodies, according to the manufacturer’s protocol. Glass slides were digitalized using the Aperio Scan-Scope XT Slide Scanner (Aperio Technologies, Vista, CA, USA) at 20x magnification. All the tumoral areas in the tissue microarray (spots) were evaluated and scored independently by the pathologist (M.M.P) and the oncologist (T.F.P.J.), without previous knowledge of the clinicopathological outcomes of the patients. The evaluation of the immunostaining was as follows: VEGFR1 (membrane and cytoplasm), VEGFR2 (membrane and cytoplasm), VEGFR3 (membrane and cytoplasm), PlGF in the cytoplasm, and DLL4 and NOTCH1 in the membrane. A membrane staining algorithm (Membrane v1, Aperio, Vista, CA, USA) was used to determine the intensity and extent of cell membrane staining. Tumor cells with weak or partial membrane staining were scored 1+; tumor cells with moderate and complete membrane staining were considered 2+; tumor cells with intense and complete membrane staining were classified as 3+. For each TMA core, the percentage of cells with score 0, +1, +2, +3 was registered. A positive staining was considered for cells with scores 2+ and 3+, except for DLL4, where a score of 1+, 2+ and 3+ were considered positive. The percentage of cells with positive staining in each TMA core was summed up. The mean value per replicate was used for the statistical analysis. A sample was considered non-representative when there were <500 analyzed cells. For the quantification of stain in the cytoplasm, the Positive Pixel Count Algorithm (Aperio, Vista, CA, USA) was used to sum the strongly and moderately positive pixels in each core. The analyses included the classification of staining as strongly, moderately and weakly positive, the number of negative cells, the analyzed area, and the ratio of the number of positive/total number of cells. The mean value per replicate was used for statistical analyses. A sample was considered non-representative when there was an area <0.08 μm² (10 % of the total core area).

All analyses were performed either with the R statistical software, version 2.3.0 (www.​r-project.​org) or SPPS v.17 statistical package (Chicago, IL, USA).

Median age at diagnosis was 56 years (range, 28–80 years); 58 % of patients were male, and 51 % of patients presented one or more comorbidities. The most common CRC localization was descending/sigmoid colon. Although all tumors were adenocarcinomas, 9.6 % presented mucinous histology (mucinous and mucin-secreting) and 82.6 % were moderately differentiated. By the time of diagnosis, 72.8 % of patients had stage IV (synchronous metastases), the remaining patients developed metachronous metastases (27.2 %), the liver being the most common site of metastasis. Ninety-five percent of patients underwent resection of the primary tumor, and 58 % underwent metastasis resection at some point. Tumor characteristics are presented in Table 1.

Although most of the patients had an advanced stage of the disease at diagnosis, 96.9 % presented an ECOG (Eastern Cooperative Oncology Group) performance status of either 0 or 1. Bevacizumab was administered as the first, second, third, and even the fourth line of therapy, with the majority of cases (70 %) using it as the first-line. Irinotecan plus 5-fluorouracil and leucovorin was the most common chemotherapy drug regimen associated with bevacizumab (53.3 %). Hypertension is a common side effect of bevacizumab therapy. In our cohort, 25.8 % of patients experienced chemotherapy-related hypertension grade 2 or 3.

In the total cohort of 105 patients, the median PFS was 10.2 months. There was a significant association between PFS and ECOG 0 or 1 (p = 0.002), and PFS and histological differentiation grades 1+ and 2+ (p = 0.027), shown in Fig. 2a and b. Concerning the presence of metastasis, a longer PFS was significantly associated with the presence of liver metastasis (p = 0.003, Fig. 2c), absence of lymph node metastasis (p = 0.014, Fig. 2d), absence of peritoneal metastasis (p = 0.023, Fig. 2e), and with patients undergoing metastasis resection (p < 0.001, Fig. 2f).

The median OS of the total cohort was 24.4 months (Fig. 3a). There was a significant association between better OS and ECOG 0 or 1 (p = 0.01, Fig. 3b). Improved OS was also associated with histological differentiation grades 1+ or 2+ (p = 0.006, Fig. 3c), presence of liver metastasis (p < 0.001, Fig. 3d), absence of lymph node metastasis (p = 0.004, Fig. 3e), and with patients undergoing metastasis resection (p < 0.0001, Fig. 3f). Patients who underwent metastasis resection at some point in the course of their disease had better OS (5-year survival rate of 40 %) as well as better PFS, with a10-year survival rate of 18 %.

Response rates were analyzed in 95 patients since nine patients did not present detectable disease prior to bevacizumab treatment and one patient had no data regarding response to treatment. Association with clinical variables is shown in Table 2. Histological differentiation grades 1+ and 2+, non-mucinous histology, liver metastasis and first-line therapy were significantly associated with better objective response rates. In a multivariate logistic regression model for objective response the included variables were line of bevacizumab treatment (first vs. others), combined histology (mucin-secreting vs. non-mucin-secreting), histological differentiation grade (1, 2 vs. 3) and the presence of liver metastasis (yes vs. no). Bevacizumab, used in the first-line of treatment, and the presence of liver metastasis were independently associated with objective response rate (Table 3).

We also evaluated ORR, PFS and OS from patients that received bevacizumab as a first-line (N = 73) and second-line treatment (N = 29) separately. Demographics and clinical characteristics of patients treated with bevacizumab as a first-line therapy are shown in (Additional file 1: Table S1). ORR, median PFS and median OS was 59.7 %, 10.6 months and 28.9 months in first-line and 36.0 %, 8.2 months and 19.9 months in second-line, respectively (Additional file 1: Table S2).

Angiogenesis-related proteins were investigated by immunohistochemistry in patients with the disease detected prior to bevacizumab treatment (Additional file 2: Figure S1). The proteins expressed most abundantly were VEGFR1 and NOTCH1. The expression values for VEGFRs, PlGF, DLL4 and NOTCH1 are presented in Table 4. All of the VEGF receptors 1, 2 and 3 had significant positive correlations among each other for both membrane and cytoplasmic expression. All other significant correlations involving PlGF, DLL4 and NOTCH1 are presented in Table 5.

The clinical characteristics significantly associated with angiogenesis-related proteins are described in Table 6. DLL4 did not show any significant association with clinical characteristics. There was no significant association between objective response rate and the angiogenesis-related proteins investigated.

There was a trend toward an association between better PFS with lower NOTCH1 expression (p = 0.06, Fig. 4a). Better OS was significantly associated with lower NOTCH1 expression (p = 0.01, Fig. 4b). To determine the variables independently associated with OS, ECOG (0, 1 vs. 2, 3), metastasis resection (yes vs. no), lymph node metastasis (yes vs. no), peritoneal metastasis (yes vs. no), liver metastasis (yes vs. no), histological grade (1, 2 vs. 3), and NOTCH1 expression (lower vs. higher) were included in a multivariate analysis shown in Table 7.