Sexual dimorphism in gastric cancer: tumor-associated neutrophils predict patient outcome only for women

All executed procedures were in accordance with the ethical standards of the responsible committee on human experimentation (institutional and national) and with the Helsinki Declaration of 1964 and later versions. Ethical approval was obtained from the local ethical review board (D 453/10 and D 468/17). All patient data were pseudonymized prior to study inclusion. All experimental work complied with all mandatory laboratory health and safety procedures.

We retrospectively examined all patients who had undergone either total or partial gastrectomy for adenocarcinoma of the stomach or the gastroesophageal junction between 1997 and 2009. Specimens were obtained from the archive of the Institute of Pathology, University Hospital Kiel. The following patient characteristics were retrieved from the electronic database: tumor location, type of surgery, age at diagnosis, sex, tumor type, tumor grade, residual tumor status, tumor size, depth of invasion, number of lymph nodes resected and number of lymph nodes with metastases. Patients were included if a primary adenocarcinoma of the stomach or gastroesophageal junction was histologically confirmed. Patients were excluded if (1) histology identified a tumor type other than adenocarcinoma, (2) patients had undergone a perioperative chemo- or radiotherapy, (3) basic clinicopathological variables (i.e. sex, Laurén phenotype, T-, N-, and M-category, or UICC stage) or (4) more than two of the study compartments (i.e. mucosa, tumor surface, tumor center, and invasion front) were missing (Suppl. Figure 1). Date of patient death was obtained from the “Epidemiological Cancer Registry” of the state of Schleswig–Holstein, Germany. Follow-up data of those patients who were still alive were retrieved from hospital records and general practitioners.

Specimens had been fixed in formalin and embedded in paraffin (FFPE). Paraffin sections were stained with hematoxylin and eosin. Tumors were classified according to Laurén (Lauren 1965). The pathological tumor (T), node (N), distant metastasis (M) stage of all study patients was defined according to the 8th edition of the International Union Against Cancer guidelines (Brierley et al. 2016).

Immunostaining was carried out using the Bondmax automated slide staining system (Leica Microsystems, Wetzlar, Germany) and a rabbit polyclonal anti-human-MPO antibody (dilution 1:2000; Dako, Carpinteria, CA, USA) diluted with Bond Primary Antibody Diluent (Leica, Newcastle, GB). Immunostaining was visualized with the Bond Polymer Refine Detection Kit (Leica Biosystems, Newcastle, GB). Automated antigen retrieval was carried out using Bond Epitope Retrieval Solution (Leica Biosystems, Newcastle, GB) at 20 min. Omission of the primary antibody served as a negative control.

Digital images of immunohistochemically stained tissue sections were obtained using a Leica SCN400 microscopic whole-slide scanner (Leica Biosystems, Nussloch, Germany) at its maximum, nominally 40-times magnification. The pixel-to-pixel distance equals 0.25 µm in the virtual image. The scanned images were exported from the scanner system as Leica SCN format files. To detect MPO⁺ cells, semiautomatic image analysis with Definiens Tissue Studio (version 3.6.1, Definiens, München, Germany) was performed at 20-times magnification. Software settings were used to vary the programmed outputs and classify the cells wanted. Tissue-background separation distinguishes between tissue and background (auto threshold; multiple tissue pieces; 10,000 µm² minimum tissue size). Nuclei were detected (nucleus detection: 0.1 hematoxylin threshold; 50 µm² typical nucleus size), and virtual cell borders were designated (cell simulation: simulation mode = grow from nuclei; 1 µm maximum cell growth). Depending on the intensity of brown chromogen (general settings: stain combination = HC brown chromogen; IHC marker = cytoplasm), the designated cell was classified as a TAN (cell classification: selected feature = HC marker intensity; measurement in = cell; threshold none/low  = 0.5). Analyses were exported as CSV files, which had to be converted in order to be used. The viewer and painting program VMP was used to mark four tumor compartments (Fig. 1).

Four tumor compartments were identified: nonneoplastic peritumoral mucosa, tumor surface, tumor center and invasion front. The peritumoral mucosa consisted of the whole tumor-free area between the muscularis mucosa and the mucin layer. For the tumor surface compartment, we marked an area from the surface up to 500 μm into the tumor, avoiding the necrotic layers directly covering the luminal tumor surface. In the tumor center compartment, many parts of the tumor were captured. The marked invasion front was up to 250 μm wide and could include small parts of the surrounding stroma. All marked compartments avoided tissue-free space (e.g., artifacts generated during the cutting of the FFPE tissue samples), large necrotic areas or neoplastic glands wider than 200 μm filled with neutrophil debris and apoptotic bodies.

H. pylori (Warneke et al. 2013a, b), EBV (Warneke et al. 2013a, b), HER2 (Warneke et al. 2013a, b), MET (Metzger et al. 2016), and MSI (Mathiak et al. 2017) statuses were assessed as previously described.

Whole tissue sections from GCs were stained with an antibody directed against MPO. The density of MPO⁺TANs within the tumor compartments, the tumor surface, tumor center, invasion front, and peritumoral area was calculated. The results were correlated with clinicopathological patient characteristics and survival. The study was carried out following REMARK criteria (Altman et al. 2012).

Statistical analyses were performed using SPSS 24.0 and 25.0 (IBM Corporation, New York, USA). TAN densities were first examined as raw score values by the Wilcoxon signed-rank test, then dichotomized at the median and divided into four groups by splitting the densities into quartiles (Q1, Q2, Q3 and Q4). Subsequently, quartiles were grouped into a TAN-low (Q1) and a TAN-high (Q2, Q3, Q4) group. The significance of the correlations between clinicopathological variables was tested using Fisher’s exact test. For ordinal scale variables, we used Kendall’s tau test for the calculation instead. A p-value was considered statistically significant if p ≤ 0.05. To compensate false discovery rate within the correlations we applied the Simes (Benjamini-Hochberg) procedure (multiple testing correction). Overall (OS) and tumor-specific survival (TSS) were computed using the Kaplan–Meier method and compared by log-rank test to determine the significance of differences between the survival curves. The cohort was also separated by sex, and raw score values, OS, and TSS were calculated again. To estimate their impact, clinicopathological features and genetic alterations were correlated by sex and their impact on survival. Additionally, a multivariate Cox regression model was carried out by counting all factors with p ≤ 0.10 in the univariate analysis. Sample size calculation for survival analysis was done according to Schoenfeld (1983). We assumed a type II error rate β = 0.2.

First, we examined the distribution of TANs in four different compartments, i.e., the peritumoral nonneoplastic mucosa, tumor surface, tumor center, and invasion front (Fig. 1). The median density (n/mm²) of TANs differed significantly between the four different compartments (p < 0.001). The highest median density was found at the tumor surface (870.2 TANs/mm²) and the lowest in the nonneoplastic mucosa (57.9 TANs/mm²; Table 1). Interestingly, the median number of TANs was lower in the tumor center (130.0 TANs/mm²) than in the tumor surface (870.2 TANs/mm²) and invasion front (226.8 TANs/mm²; Table 1). These data provide evidence that TANs in GC are not simply related to an unspecific inflammatory response, e.g., loss of mucosal barrier function, but are specifically spatially enriched at the invasion front.

Next, we correlated TAN densities with the histological phenotype according to Laurén. With regard to the individual tumor type (i.e. intestinal, diffuse, mixed, and unclassified), again, median TAN densities varied between the four different compartments (Table 1). In connection with the individual compartments, no significant differences were found in the mucosa and tumor surface. However, TAN densities in the tumor center and invasion front differed significantly between diffuse and non-diffuse type, with GCs having the lowest median number of TANs in diffuse type GC (tumor center: p < 0.001; invasion front: p < 0.001). These data provide evidence that diffuse type GC and non-diffuse type GCs may differ in their ability to recruit TANs into the tumor center and invasion front.

Thereafter, we correlated TAN density with sex. Again, differences were found in the overall distribution of TANs in the four different compartments (Table 1, Fig. 2). However, it was interesting to note that the increase of TAN density from the tumor center to the invasion front was greater in men (128.2 vs. 238.9 TANs/mm²) compared with women (129.5 vs. 161.5 TANs/mm²). TANs may contribute to GC biology in both men and women but show subtle spatially related differences in median densities. Clinicopathological features closely linked to sex were: patients’ age, localization, Laurén phenotype, EBV- and MET status (Suppl. Table 1).

The assessment of the median values of TANs showed that densities vary, and we next sought to test the hypothesis that TAN density in each compartment (tumor surface, tumor center and invasion front) may correlate with clinicopathological patient characteristics. Using an explorative approach, we first categorized TAN density into “low” and “high” by splitting at the median (Suppl. Table 2). This categorization showed that TAN density correlated with different clinicopathological patient characteristics, e.g., tumor type according to Laurén and MSI status (Suppl. Table 2). However, given the large range of TAN density in GC (0–6711.0 TANs/mm²), we considered that dichotomization of the densities at the median may underestimate the effect of “lower” TAN counts. Therefore, we categorized TAN densities into four quartiles (Q1, Q2, Q3, Q4). This largely showed similar results (Suppl. Table 3), and we then hypothesized that, similar to reference values in whole blood counts, TAN density may have “a (near) physiological range” corresponding to the lowest quartile and a “pathological range” corresponding to the remaining quartiles. Hence, we next dichotomized the densities into a TAN-low group and a TAN-high group by comparing Q1 with Q2-Q4, respectively (Table 2). This showed that TAN density in the tumor center is related to Laurén phenotype (significant after multiple testing correction), tumor grade (significant after multiple testing correction), and EBV and MSI status. At the invasion front, TAN density is related to patient sex, Laurén phenotype (significant after multiple testing correction), tumor grade, T-category, EBV, MSI (significant after multiple testing correction), and HER2 status and patient survival (Table 2). Collectively, these data show that TANs are biologically relevant to tumors in GC and that the effect can already be demonstrated when the cutoff is set below the median values (i.e., Q1 vs. Q2–Q4).

Finally, we were interested in exploring the correlation between TAN density and patient survival (Table 2; Suppl. Tables 2 and 3). Dichotomization of the patient cohort at the median (Suppl. Table 2) and at the four quartiles (Suppl. Table 3) showed no differences in tumor-specific survival. However, when TAN density was dichotomized into Q1 (low) and Q2–4 (high), patients with high TAN densities at the invasion front had a much more favorable overall survival (OS; median survival 13.6 ± 2.2 months TAN-low vs. 16.7 ± 1.9 months TAN-high; p = 0.017) and tumor-specific survival (TSS; median survival 14.7 ± 2.9 months TAN-low vs. 19.6 ± 3.8 months TAN-high; p = 0.013) compared to patients with low TAN densities. However, after multiple testing correction these results were no longer significant.

Since we noticed sex-specific differences in the TAN density at the invasion front (Table 2), we next explored patient survival separately for men and women. This finally showed that TAN density had no impact on patient survival in men, while it was highly significantly correlated with patient outcome in women. Women with high TAN densities (Q2–Q4) in the tumor center had a better OS (median survival 12.9 ± 1.6 months TAN-low vs. 17.9 ± 2.4 months TAN-high; p = 0.079) and a significantly better TSS (median survival 12.8 ± 1.5 months TAN-low vs. 18.8 ± 3.7 months TAN-high; p < 0.026) compared with women with low TAN densities. Women with high TAN densities (Q2–Q4) at the invasion front had also a significantly better OS (median survival 12.1 ± 2.2 months TAN-low vs. 26.3 ± 7.1 months TAN-high; p < 0.001) and TSS (median survival 12.1 ± 1.7 months TAN-low vs. 36.1 ± 14.0 months TAN-high; p < 0.001) compared with women with low TAN densities (Q1; Fig. 3; Suppl. Fig. 2). This effect was not related to tumor type according to Laurén (Fig. 4) and there was also no general difference of TSS between men and women in our cohort (Suppl. Fig. 3).

The entire GC collective showed a median OS of 15.0 months and a median TSS of 16.7 months. Patient prognosis significantly depended on the Laurén phenotype, T-, N-, M-, L-, V- , and R-category, UICC stage, grade, lymph node ratio, MSI status, MET status (data not shown), and TAN density at the invasion front (p = 0.013). TSS of women depended on Laurén phenotype, T-, N-, M-, L-, V-, and R-category, UICC stage, grade, lymph node ratio, MET status, and TAN density at tumor center and invasion front. On multivariate analysis, TAN density at the invasion front (HR 2.77; 95% CI 1.64–4.77), R-, L-, and N-category turned out to be significant and independent prognosticators of women’s tumor-specific survival (Table 3; Suppl. Fig. 1).

None of our patients received perioperative or neoadjuvant chemotherapy, and no comment can be made on the effect of chemotherapy on TANs in GC. We also did not study the suitability of endoscopic biopsies to asses TAN densities. However, since the invasion front is rarely captured by biopsies, the assessment of TAN status is prone to a sampling error. Due to the retrospective and observational character of our study using only formalin-fixed and paraffin-embedded tissue samples, we were unable to provide any functional data. However, our extensive morphological and statistical analyses may help to generate hypotheses for future experimental studies on the tumor-biological mechanisms of TANs in GC.