Tumour necrosis factor-alpha expression in tumour islets confers a survival advantage in non-small cell lung cancer

The study was approved by the Leicestershire Research Ethics Committee. The tissue specimens evaluated were from patients with NSCLC who had undergone resection with curative intent at the University Hospitals of Leicester National Health Service Trust (Leicester, United Kingdom). These patients had resections during two periods - one dating from 1991 to 1994 and the second from January to December 1999. This cohort of patients has been described previously [26]. Of note, due to exhaustion of tumour tissue, 133 patient samples with >60 day survival post surgery were available for analysis in this study. Of the 133 patients studied, 88 were men and average age at surgery was 65.8 years (standard deviation, 9.8; range, 33 to 82 years). Full clinicopathologic information was gathered before and after surgery, including patient characteristics, treatment, combined clinical and surgical staging results (preoperative staging by computed tomography scan, selective mediastinoscopy, and systematic lymph node sampling at operation), histologic subtype, tumour grade, and survival data. Patients were divided into two groups: above median survival (AMS) (mean ± SEM 84.0 ± 5.1 months) and below median survival (BMS) (mean ± SEM 10.8 ± 0.8 months). Macrophage-associated TNFα expression has been described previously in 26 of these patients [25]. Patient characteristics are shown in Table 1.

Specimens studied were formalin fixed and paraffin embedded. Only blocks containing the advancing edge of the primary tumour were evaluated. Tissue sections of 4 μm thickness were cut onto glass slides and then de-waxed in xylene and rehydrated through graded alcohols. Antigen retrieval was carried out using Trilogy Antigen Retrieval solution (Cell Marque, Hot Springs, United States of America) in a pressure cooker (heated to 117.5°C for 1 min and then cooled to 100°C for 30 seconds). TNFα mouse antihuman antibody was used (clone P/T2; Abcam, Cambridge, United Kingdom). Immunostaining was performed and TNFα was developed with peroxidase and 3,3'-diaminobenzidine tetrahydrochloride (brown reaction product). Sections were then counterstained with haematoxylin and mounted in an aqueous mounting medium (BDH Chemicals Ltd, Poole, United Kingdom). Appropriate isotype controls were performed where the primary antibodies were replaced by irrelevant mouse mAb of the same isotype and at the same concentration as the specific primary mAb. To assess whether TNFα + cells were localised to mast cells, a double-stain technique described previously by us was used [25, 26], with the mouse monoclonal antibody to human tryptase (clone AA1, Dako Cytomation, Ely, Cambridgeshire, United Kingdom) employed to identify mast cells.

Analysis was performed blind with respect to the clinical outcome. The ten most representative high-power fields (x400) per slide were manually selected using an Olympus BX50 microscope (Olympus, Southall, United Kingdom). The respective areas of stroma and tumour islets were then measured at x400 magnification using Analysis imaging software (Soft Imaging System GmbH). The number of nucleated cells with positive staining for TNFα in each area were then counted manually and expressed as cells/mm² of stroma or tumour islets. Analysis was repeated for 20 patients to assess repeatability and validity.

Statistical analyses were carried out using the GraphPad Prism software package (v. 4.02; GraphPad Prism Software Inc, San Diego, CA). For categoric analysis, the median value was used as a cut point to dichotomise the series. The χ² test was used to test for relationships between categoric variables, and the Mann-Whitney nonparametric test was used to compare categoric with continuous variables. The Kruskal-Wallis one-way analysis of variance test was used to compare multiple groups. Kaplan-Meier survival curves were used to look for correlations with survival and were compared with the use of the log-rank statistic. For the above comparisons, p < 0.05 was considered statistically significant. A multivariate Cox proportional hazards model was used to estimate adjusted hazard ratios, 95% CIs, and to identify which of the macrophage markers were independent prognostic factors using the Statistical Package for the Social Sciences (SPSS, v. 13.0; SPSS Inc, Chicago, IL). The validity of the proportional hazards assumption was assessed from log(-log [Survival]) curves.

Patent characteristics are shown in Table 1. Of the 133 patients studied, 110 had died at the time of analysis. Thirty-day mortality for the cohort was 4% and patients who died within the first 60 days of surgery were not included in this analysis. Sixty-eight tumours were squamous, 42 adenocarcinoma, 12 large cell, and 11 other. Sixty were stage I, 37 stage II, 32 stage IIIa, and four stage IIIb or IV. Two patients had additional postoperative chemotherapy and 20 had additional radiotherapy, 16 of whom had it for later palliation. Neither had any effect on survival. The overall 5-year survival was 31.3%.

Clear and distinguishable staining was evident for TNFα (Figure 1). Appropriate isotype controls were negative. In order to assess the validity of the method, area measurements and cell counts were repeated and intraclass correlation coefficients calculated. Good correlations were found for both: 0.997 (95% CI, 0.996 to 0.998, p < 0.001) and 0.994 (95% CI, 0.992 to 0.996, p < 0.001). This method of analysis has also been validated by our group previously [26].

There was increased expression of TNFα in the tumour islets of patients with AMS compared to those with BMS (median 27.2 versus 18.8 cells/mm² respectively, p = 0.006). There was no significant difference in the expression of stromal TNFα between the two groups (AMS 23.8 versus BMS 16.8 cells/mm², p = 0.31) (Figures 2A and 2B).

Scatter plots of the raw data of TNFα density versus survival are shown in Figure 3. Spearman's rank correlation coefficient was calculated to assess any potential relationship with survival. A direct relationship between tumour islet TNFα density and survival was noted (r_s = 0.213, p = 0.01). No significant relationship was seen between survival and stromal TNFα density (r_s = -0.01, p = 0.90).

For further analysis, the data were divided into two or three equal groups according to TNFα density. Kaplan-Meier survival curves were plotted and the log-rank statistic used to compare survival rates. When looking at two groups separated by the median, there was a non-significant trend for improved survival with above median TNFα expression in the tumour islets (p = 0.15, not shown). When divided into tertiles, there was significantly improved survival in the top tertile of TNFα expression in the tumour islets compared to the middle and lower tertiles (Figure 4A). Thus patients with high islet TNFα expression (upper tertile) were noted to have a significantly higher 5-year predicted survival as opposed to patients with low TNFα expression (lower tertile) (43% versus 22%, p = 0.01). There was no significant relationship between TNFα expression in the stroma and survival (Figure 4B). Similar differences in survival with respect to TNFα expression in the tumour islets were also evident within tumour stages (Figure 5), although the differences did not reach statistical significance due to the smaller numbers within each stage. Interestingly, patients with stage IIIa disease in the top tertile of islet TNFα expression had a 5 year survival of 25% compared to 26% survival for those patients with stage I disease in the lower tertile of islet TNFα expression.

A Multivariate Cox Proportional Hazards model was used to estimate adjusted hazard ratios, 95% CIs, and to assess whether TNFα islet or stromal counts were independent prognostic factors, using a forward stepwise method with a probability of 0.05 for entry and 0.10 for removal by likelihood ratio statistics. Results of the multivariate analysis are shown in Table 2. Continuous variables were used for islet and stromal TNFα. For every increase of 1 in the value of the variable, the hazard increases by the value of the hazard ratio. Expression of TNFα in the tumour islets emerged as a significant independent predictor of survival (hazard ratio 0.995, 95% CI 0.989 to 1.000, p = 0.048). Surprisingly, expression of TNFα in the tumour stroma emerged as a significant independent predictor of reduced survival (hazard ratio 1.007, 95% CI 1.002 to 1.011, p = 0.007). There was no evidence of violation of the proportional hazards assumption.

We have shown previously that in patients with extended survival, there is increased expression of macrophages in the tumour islets compared to patients with poor survival [26], and that in a separate study the majority of tumour islet macrophages expressed TNFα [25]. Macrophages accounted for approximately 60% of the TNFα + cells in the extended survival patients. While there was a marked difference in the number of TNFα + macrophages between extended survival and poor survival islets [25], it is evident from Figure 2 that the difference in total TNFα expression in this cohort of 133 patients is not so marked. Based on morphology, it was considered that a proportion of the TNFα + cells that were not macrophages may be mast cells. We therefore analysed the cellular distribution of TNFα expressed by mast cells (Figure 6) in the subset of samples previously stained for macrophage-TNFα (AMS [n = 20] and BMS [n = 20]) [25]. In keeping with our previous observations [26], mast cell numbers were significantly increased in the islets of the AMS compared to BMS patients (medians 22.6 versus 0.7 cells/mm², p < 0.001). Interestingly, mast cells accounted for 45.2% of TNFα expression in the islets of the AMS patients, but only 5.2% of the BMS patients (Table 3)(p < 0.001). Thus taking macrophage and mast cell-associated TNFα together, approximately 100% of islet TNFα was localized to mast cells or macrophages in the subset of AMS patients studied, while in the islets of the subset of BMS patients studied, only 28% of TNFα immunoreactivity was localized to these cells (Table 3) (p < 0.001). Similar results were evident in the tumour stroma (Table 3, p < 0.001). This is interesting because it demonstrates that in BMS patients, there is robust TNFα expression by cells other than macrophages and mast cells. These cells were predominantly mononuclear cells and rarely tumour epithelial cells. This suggests that TNFα is highly beneficial only when localized to macrophages and mast cells in tumour islets, and not when expressed by other cell types.