Computer-Assisted Analysis of Biopsy Specimen Microvessels Predicts the Outcome of Esophageal Cancers Treated with Chemoradiotherapy

Materials and Methods

Patients. A total of 348 patients with esophageal cancers were diagnosed and treated at the National Cancer Center Hospital East between 1992 and 1999. Surgery was done as the main treatment for 139 patients, and 209 received definitive chemoradiotherapy. Among the 209 patients, 51 met the following criteria to be included in this study: (a) the tumor was histologically diagnosed as squamous cell carcinoma; (b) patient age was ≤75 years; (c) sufficient biopsy specimens (tumor area >0.6 mm², which is about thrice of ×400 magnification field) were available before treatment; (d) performance status on the Eastern Cooperative Oncology Group scale ≤2; and (e) stage T_2-3, any N, M₀ on the International Union against Cancer tumor-node-metastasis classification. The patients' characteristics are listed in Table 1 .

Treatment protocol. Chemotherapy consisted of continuous infusion of 5-fluorouracil (400 mg/m²/d, on days 1-5 and 8-12) and a weekly infusion of cisplatin (40 mg/m², days 1 and 8). Concurrent radiotherapy was given at 2 Gy/d for 5 d/wk with a 2-week break after a dose of 30 Gy, and restarted on day 36 along with the same schedule of chemotherapy as before. The total radiation dose was 60 Gy. After concurrent chemoradiotherapy, two additional courses of chemotherapy (5-fluorouracil, 800 mg/m²/d, on days 1-5 and 29-33; cisplatin, 80 mg/m², on days 1 and 29) were basically administered if patients responded to treatment without serious side effect. Further additional courses were optional although they were limited to a total of four courses.

Definition of tumor response. The first evaluation was done ∼1 month after treatment. Patients then received computed tomography scanning and esophagoscopy every 2 or 3 months during the first year and every 6 months thereafter. Tumor recurrences were all proved histologically by biopsy.

Response at the primary site was evaluated by endoscopic examination. The criteria for evaluation were as follows: complete remission was defined as disappearance of tumor lesion and ulceration for ≧4 weeks with negative biopsy results; partial remission was determined when primary tumor was observed on esophagography as being reduced in area by ≧50%. Responses of metastatic lymph nodes were assessed by computed tomography scanning according to the WHO criteria for measurable disease.

Immunohistochemical staining of blood microvessels and computer-assisted image analysis. All biopsy specimens were taken at the time of diagnosis. Immunohistochemical staining of blood microvessels was done with the standard avidin-biotin complex technique using diaminobenzidine as a chromogen and hematoxylin as counterstain. Antigen was retrieved by treating with 0.05% pepsin in 0.01 N HCl for 5 minutes at room temperature. A mouse antibody for CD31 was used as primary antibody (4°C, overnight, 1:50 dilution; DAKO, Glostrup, Denmark). After washing, sections were incubated with an avidin-biotin complex reagent (DAKO). Color reactions were developed for 5 minutes in diaminobenzidine-Tris buffer (pH 7.6) containing 0.3% hydrogen peroxide.

Digitized images of immunohistochemically stained sections of whole specimen at ×100 magnification (10× objective and 10× ocular) were obtained using a KS 300 image analysis system (capture resolution 768 × 580, Karl Zeiss Vision K.K., Jena, Germany). Vessels with lumens located around or inside tumor nest were traced by one of the authors (H.S.). The process of image analysis has been described elsewhere (9). Briefly, traced microvessels and the outline of the total specimen were converted to binary images and TN, TP, and TA were calculated. Data were recorded as TN/TA, TP/TA, and TP/TA ratios. As the oxygen diffusion distance from blood vessels is ∼150 μm (10), the hypoxic ratio was calculated as the ratio of tumor area >150 μm from blood vessels to the TA (Fig. 1A ).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 1.

A, image analysis of microvessels. I, the lumen of each microvessel was traced by an observer and counted by the computer. The TP was then calculated using the computer. II, the tumor region was outlined by the observer and the tumor area was calculated using the computer. III, the area of the tumor located >150 μm from the nearest vessel was calculated by the computer (yellow area) and then hypoxia ratio was calculated. B, microvessel hotspot counted by computer. I, microvessels were labeled by an observer. Bar, 100 μm. II, labeled vessels were converted to binary images. III, the scanning circle is 500 μm in diameter and 0.196 mm² in area, which equals the area of a ×400 power field. The overlay of two neighboring circles is 375 μm. IV, the binary image was scanned. V, a schematic image of microvessel distribution derived from computer analysis. VI, areas with the highest microvessel number were identified as microvessel hotspots.

Two of the authors (Z.S. and H.S.) counted hotspot microvessel numbers independently without knowledge of patient outcomes. The immunohistochemically stained specimens were first scanned at low magnification (×10-×100); three high-magnification (×400) fields with plentiful vascular tumor areas were then selected and counted as hotspots. The mean number of vessels from three fields was recorded as the hotspot MVD.

As an alternative for manual counting, a computer-assisted method was used to identify hotspot and count vessels in specimens. The previously traced vessels were converted to binary images and were scanned consecutively. The scanning circle was 500 μm in diameter and 0.196 mm² in area, which was the same as a ×400 magnification field. Microvessels within each circle were counted by computer. The overlay of adjacent circles in both the X and Y axes was set arbitrarily at 375 μm (three fourths of the diameter). The mean number of MVDs in the three circles containing the highest MVD count was recorded as the hotspot MVD of the corresponding specimen (Fig. 1B).

Statistical analysis. Correlations between different factors were assessed using Pearson's correlation coefficient. A Cox proportional hazard analysis was used to evaluate clinicopathologic and microvessel factors in the prediction of treatment outcome. Survival curves were generated using the Kaplan-Meier method and statistical differences between curves were calculated using the log-rank test. For evaluation of continuous variables in survival analysis, patients were divided into two groups based on an optimal cutoff derived from receiver operating characteristic analysis. GraphPad Prism software (GraphPad Software, San Diego, CA) was used for receiver operating characteristic analysis and Statistica (StatSoft, Tulsa, OK) was used for all other analysis.

Results

Treatment outcome. All 51 patients completed the concurrent chemoradiotherapy with a total radiation dose of 60 Gy. Seven patients (14%) received one additional course of chemotherapy, and 25 patients (49%), 2 patients (4%), and 2 patients (4%) received an additional two, three, and four courses, respectively. The median patient follow-up time was 32 months (range, 5.9-121.5 months). Thirty-nine patients (76%) achieved clinical complete remission and 12 patients (24%) achieved partial remission at primary site. Among the 39 with clinical complete remission at primary site, 12 patients developed recurrence in the primary area, 3 in a distant area (a different site in esophagus), and 4 had distant metastases during follow-up (3 in the liver and 1 in the lung). Up to January 1, 2005, 23 patients had died of esophageal cancer, 8 patients died of other disease or accidents, and the other 20 cases were alive at the last follow-up time point.

Image analysis of microvessels in biopsy specimens. Areas of inflammation, sclerotic tumor, and adjacent benign tissue were identified and excluded by observer when calculating the TA, which ranged from 0.99 to 3.37 mm², with a median of 1.92 mm². The TN varied from 3 to 204, with a median of 100. The TP ranged from 0.35 to 41.26 mm, with a median of 11.81 mm. Tumor hypoxic ratio ranged from 0.11% to 78.63%, with a median of 8.34%.

Comparison of manual MVD counting and computer-assisted MVD counting. The results of two manual MVD counts and the computer-assisted counts are shown in Table 2 .

Although hotspot MVD counts for a given patient varied between different observers and the computer, analysis showed that they were correlated linearly (Table 2).

Receiver operating characteristic analysis. To evaluate the ability for the prediction of survival, we evaluated the accuracy of prediction of death of esophageal cancer at 2 years for each microvessel factor. This interval of 2 years was selected because most of the complete cases happened in this interval (19 in 23 cases) and only three patients (all of whom died of other diseases and were excluded from the receiver operating characteristic analysis) censored before the end of 2 years. The predictive power was estimated by calculating the area under receiver operating characteristic curves (11, 12). All factors, including TN/TA, TP/TA, and hypoxic ratios and observer-counted and computer-derived hotspot MVDs, were found to be predictors of 2-year survival (Table 3 ); the observer-counted MVD showed the weakest power in predicting death 2 years after treatment.

Univariate analysis of survival. Univariate Cox proportional hazard analysis was done to evaluate the relation between clinicopathologic factors and overall survival. The result is presented in Table 4 . No clinicopathologic factor correlated with overall survival.

From the receiver operating characteristic curves, the optimal cutoffs for varied microvessel factors were determined to stratify patients into two groups, and univariate Cox proportional analysis revealed that patients with low TN/TA (P = 0.023) or low TP/TA (P = 0.037) had a higher risk of dying of esophageal cancer after chemoradiotherapy (Table 5 ). Patients with a low ratio of tumor hypoxic area tended to survive longer after treatment but this was not statistically significant (P = 0.329). The hotspot MVDs counted by the two observers had no relation with overall survival (P = 0.203 and 0.119, respectively) whereas hotspot MVDs counted by the computer showed a significant association with survival (P = 0.036).

Multivariate analysis for survival. In the multivariate Cox proportional hazard analysis, computer-derived hotspot MVD counts and the TN/TA and TP/TA ratios were analyzed combined with T and N stage, which showed the highest significance among clinicopathologic factors by univariate analysis. All three microvessel factors proved to be independent predictors for overall survival (P = 0.019 for hotspot MVD, P = 0.018 for TN/TA, and P = 0.044 for TP/TA; Table 6 ).

Kaplan-Meier survival analysis.Figure 2 shows the survival curves generated using the Kaplan-Meier method. Patients with high MVD and high TN/TA and TP/TA ratios had 5-year survival rates of 73%, 79%, and 73%, respectively, whereas the group of patients with low such factors had 5-year survival rates of 46%, 45%, and 41%, respectively. Log-rank test showed that these differences were statistically significant (P = 0.025 for hotspot MVD, P = 0.008 for TN/TA, P = 0.031 for TP/TA).

• Download figure
• Open in new tab
• Download powerpoint

Fig. 2.

Kaplan-Meier overall survival curves for patients with T_2-3 M₀ esophageal cancer treated with chemoradiotherapy.

Because hotspot MVD count and the TN/TA and TP/TA ratios all proved to be predictive factors for the outcome of patients treated with chemoradiotherapy, we investigated whether combinations of these factors would provide more powerful and more precise predictors. Hotspot MVD showed a strongly positive correlation with TN/TA (Pearson test, r = 0.843, P < 0.000001) whereas TP/TA was independent of hotspot MVD (r = 0.023) and was relatively weakly correlated with TN/TA (r = 0.318, P = 0.022). We therefore selected the combinations of MVD and TP/TA, TN/TA and TP/TA ratios as new factors and investigated if they could predict survival of patients. The high hotspot MVD plus high TP/TA group included eight patients and the high TN/TA plus high TP/TA included 11 patients (including all the eight in the high MVD plus high TP/TA group). Surprisingly, none of the patients with both high MVD and high TP/TA ratio died of esophageal cancer during follow-up and only one patient died of esophageal cancer in the high TN/TA plus high TP/TA group. The Kaplan-Meier survival curve of the high TN/TA plus high TP/TA group was presented in Fig. 3 .

• Download figure
• Open in new tab
• Download powerpoint

Fig. 3.

Kaplan-Meier overall survival curves for patients with high TN/TA and high TP/TA.

Discussion

We previously reported that hotspot MVD and the TN/TA, TP/TA, and hypoxic ratios in biopsy specimen are prognostic factors for laryngeal cancer patients treated by radiotherapy (9). Here we found that, for patients with T_2-3 esophageal cancers, hotspot MVD, TN/TA, and TP/TA were favorable predictors for overall survival. The combinations of hotspot MVD with TP/TA, or TN/TA with TP/TA, may provide more powerful predictors for predicting the outcome of such patients scheduled to undergo chemoradiotherapy.

It is likely that a low density of microvessels will lead to a decrease in oxygen transport and drug delivery into local tumor environments. Bhattacharya et al. (13) investigated avascular regions of human head and neck cancer xenografts and found that cells in these areas were hypoxic and chemotherapy resistant. Hypoxia-related factors, such as hypoxia-induced factor 1α and carbonic anhydrase IX, and hypoxic area imaging by pimonidazole or misonidazole binding have been discussed as prognostic factors for radiotherapy of cancers (14–17). Although there are substantial data implying that poorly vascularized tumors are resistant to chemotherapy and/or radiotherapy, no definitive conclusion has been drawn at present on the clinical usefulness of MVD as a marker for prognosis. There are some studies that did not show a relationship between MVD and survival (18, 19) and a reverse relationship between MVD and survival has been reported by some groups (20, 21).

These contradictory conclusions might be explained by the difficulty in evaluating MVD accurately. Manual hotspot MVD counting has been the predominant method for analysis. The observer first scans a section at low magnification (×10-×100). High angiogenesis areas can be recognized as hotspots and a higher magnification (×200-×400) is then selected to count the number of microvessels in these areas (22). All procedures, including screening hotspot area and counting vessels, are done subjectively and intraobserver or interobserver variability is almost inevitable. Our study presents one resolution of this problem. Because pathologic section is converted into digitized image, the observer only needs to trace the outlines of the microvessels and the following work is all accomplished by the computer with minimum variability and perfect reproducibility. When two observers counted the microvessels of the same patients in our study using the manual hotspot counting method, the results differed and both failed to predict the survival of patients.

Using a computerized system to evaluate tumor microvessels has been reported by some groups (23–26). The present method has two novel features. The first is observer intervention in microvessel tracing. Although completely automated analysis will undoubtedly be developed, the nonspecific staining and the varying threshold for positive endothelial staining are likely to comprise this. Manual tracing ensures the accuracy of microvessel recognition and therefore maintains a high specificity of this analysis. The second feature is the use of full section scanning to determine microvessel hotspots, which provides the most objective information on microvessel distribution within any biopsy sample.

The tumor hypoxic ratio counted by computer has been shown as a predictor for radiosensitivity (9). In the present study, there was a tendency for a low hypoxic ratio to show good patient prognosis, but this was not statistically significant. The predictive power of the hypoxic ratio for the outcome of chemoradiotherapy thus needs further investigation.

A surprising finding in our study is that in the group of patients with both high hotspot MVD and high TP/TA ratios, none died of esophageal cancer, and in the group with high TN/TA and TP/TA ratios, only one patient died of esophageal cancer during follow-up. Six of eight patients (75%) in the high hotspot MVD plus high TP/TA ratio group and 8 of 11 patients (73%) in the high TN/TA plus high TP/TA ratio group survived longer than 3 years, compared with the average level of 45% (23 of 51) for all patients. These data suggested that well-vascularized tumor may be more sensitive to chemoradiotherapy.

The subjects of this study were patients who received definitive chemoradiotherapy. Because induction chemoradiotherapy before surgery is one of the most adopted multimodal treatments for esophageal cancer, whether the computer-assisted microvessel analysis is useful in predicting the outcome in these patients remains to be determined. After induction chemoradiotherapy, only 15% to 56% of patients achieved pathologic complete remission (27, 28). This method may be extremely helpful for treatment selection in residual disease after induction chemoradiotherapy. Similarly, the usefulness of this method in assessing treatment advantage of postoperative chemoradiotherapy is also an interesting topic needing further investigation.

Esophageal adenocarcinomas were not included in this study because most esophageal cancers in Japan (>90%) histologically are squamous cell carcinoma (29). However, in western countries, adenocarcinomas account for >60% of all esophageal cancers (5, 30). An independent study is needed to investigate whether microvessel analysis is suitable for adenocarcinoma when this method is expected to be used in these countries.

High MVD was reported to correlate with distant metastasis and short survival in solid tumors (31, 32). It should be noted that surgery was the main treatment modality in most of these prognosis analysis studies. For esophageal cancer, if it is proved that patients with a high MVD have short survival after surgery, chemoradiotherapy should then be more strongly recommended for these patients.

In conclusion, using a computer-assisted image analysis system for biopsy specimens, we found that hotspot MVD and the TN/TA and TP/TA ratios were powerful predictors for the outcome of patients with esophageal cancer treated with chemoradiotherapy. Compared with manual microvessel counting, this computer-assisted method produced lower variability and higher reproducibility for the evaluation of tumor vasculature.