The clinical importance of the host anti-tumour reaction patterns in regional tumour draining lymph nodes in patients with locally advanced resectable gastric cancer: a systematic review and meta-analysis

A systematic search of the literature was performed using the Embase, MEDLINE, CINAHL, and Cochrane databases. Relevant studies published before April 2023 were collected using Medical Subject Headings (MeSH) terms as mentioned in Supplementary Table S2. Reference lists from relevant papers were screened to identify further relevant original publications.

As the initial search showed a relatively low yield when only including gastric cancer, we extended the search to include junctional and oesophageal cancer. The following criteria were used to identify studies eligible for inclusion in our analyses: (1) patients diagnosed with gastric, junctional or oesophageal cancer treated by surgical resection including lymph node dissection; (2) histological examination of morphological changes in the lymph node microarchitecture using H&E staining or other staining procedures; (3) acceptable study design: cohort studies, randomized controlled trials and case control studies; (4) full-text available in English, German, Dutch, Czech, or Japanese. Studies reporting results from case reports, cell culture-based studies, animal studies, diagnosis based on cytology, and letters to editors containing no additional information were excluded.

Study selection was performed independently by two authors (MK and EB) considering the pre-defined inclusion and exclusion criteria. The eligibility of the papers was evaluated by initial screening of title and abstract, followed by evaluation of the full text. Inter-evaluator disagreement regarding eligibility of a study to be included or not was resolved by discussion with a third researcher (HIG).

Initially, the following data items were collected: author, year of publication, country of the researcher’s affiliation, cancer type, histological type, number of patients, patient demographics (age, sex), number of lymph nodes (LNs) analyzed per patient including number of metastatic and non-metastatic LNs, criteria used by investigators to identify LN suitable for analyses, LN reactive pattern(s) analyzed in the study, histological stain used, evaluation method, result measures (quantitative versus qualitative), treatment groups if applicable, survival time, other prognostic positive/negative parameters and primary, and secondary endpoints reported in the study.

The main research outcome of interest was the relationship between LN microarchitecture feature and survival in patients with GaC or OeC treated with surgery with or without neoadjuvant or adjuvant treatment. Alternative study outcomes such as association of the LN reaction pattern with clinicopathological variables other than survival were also considered. To account for potential risk of bias in the studies, the Quality in Prognostic Studies (QUIPS) tool was applied to selected studies [33].

The meta-analysis was performed using R packages metafor (version 3.8–1) and ggforestplot (version 0.1.0) for visualization. In order to compare survival between patients who exhibited a reactive LN pattern against those without, pooled risk ratios (RR) and 95% confidence intervals were calculated using the DerSimonian and Laird algorithm [34]. The clinical heterogeneity of the studies was used to determine random or fixed model suitability, accounting for preoperative treatment, patient geography, mean age, sample size, and male/female ratio of the participants in the included studies.

Following the database search, 7597 unique records were retrieved for subsequent title- and abstract-based screening. The selection process is outlined in Fig. 1. A total of 7370 records were excluded as they did not fulfill the predefined inclusion criteria. From the remaining 227 articles, the full text was available for 170 studies. After reading the full text, 142 of the 170 studies did not meet the inclusion criteria leaving 28 studies for final analyses within this systematic review. Five of the 28 studies provided sufficient details on number of patients with reactive LN patterns and patient survival to be included in the meta-analysis. As individual patient data were not available from these studies (all published before 1992), the overall survival data was obtained from publication text, tables, or calculated from published survival curves. Observations reported by Kuriya et al. could not be included into the meta-analysis, as we were unable to calculate the required values from survival curves plotted in the paper [35]. The risk ratios were assessed using a random-effects model, since the study pool was not homogeneous with respect to the age, sex distribution of the investigated patients as well as the geographical location, as described in Table 1. Based on the QUIPS risk of bias score evaluation, 18 (64%) articles were assigned medium risk of bias, leaving the remaining 10 articles (36%) with low risk scores (see Supplementary Table S3).

The studies included in this review consisted of 17 studies in gastric cancer (GaC) patients and 11 studies in oesophageal cancer (OeC) patients including 5685 patients and 26,503 investigated LNs. On average, the LN reaction pattern was investigated in 4.66 LNs per patient, including tumour-positive and tumour-negative LN. Thirteen studies reported histopathological findings based on H&E-stained LNs, 16 studies used immunohistochemistry (IHC) or gene expression patterns (see Table 1). Studies investigating gene expression or specific immune cell type infiltration patterns were more recently published compared to studies reporting on LN compartment hyperplasia based on H&E-stained sections. Studies originated from several different geographical locations, see Table 1.

Overall, the 28 studies could be grouped into three different methodological approaches: (1) Thirteen (46%) studies investigated LN compartment specific reactions in GaC and OeC, namely sinus histiocytosis, paracortical hyperplasia, and follicular hyperplasia (SH, PH, FH, respectively) in H&E-stained tissue sections (see supplement Figure S1). Five of these studies contained detailed information on the relationship between LN reaction pattern and overall survival and were therefore deemed eligible for meta-analysis. (2) Twelve (43%) studies used immunohistochemistry to characterize specific immune cell distribution patterns in LNs (Table 1). (3) Three (11%) studies investigated RNA expression to identify differential gene expression signatures in LNs.

Thirteen studies measured volumetric changes in regional LNs of GaC and OeC patients, all of which were based on H&E staining. Although the studies shared the same LN reaction pattern terminology, namely sinus histiocytosis, follicular hyperplasia, and paracortical hyperplasia, the morphological evaluation itself was inconsistent between studies due to the use of different grading systems. The purpose of these grading systems was to semi-quantify the findings of the LN reaction patterns. Six different grading systems were used for sinus histiocytosis, four different grading systems for follicular hyperplasia, and three different grading systems for paracortical hyperplasia, see Table 2. It is noteworthy that the WHO LN reaction pattern classification by Cottier in 1972 [27] included evaluation of the LN medulla for medullary plasmacytosis, which was not investigated in any of the included gastro-oesophageal cancer studies.

We identified ten studies investigating sinus histiocytosis (SH) in GaC and OeC using multiple different grading systems. Black et al. proposed a semi-quantitative four stage grading for SH evaluation, based on manually counting of the number of histiocytes across the sinus [36] (see Table 2). Oka et al. measured the proportion of the lymph node area covered by sinuses [30].

Five studies investigating SH reported a significant association between the presence of SH and favorable outcomes. Oka reported favorable outcomes with increased incidence of SH irrespective of preoperative radiotherapy treatment in OeC patients and GaC patients [30, 37]. Presence of SH in regional LNs was also related to better survival in GaC by Riegrova and Bedikian [38, 39]. Kuriya et al. associated SH with better prognosis in oesophageal squamous carcinoma patients using the SH grading system proposed by Black et al. [35]. Black et al. reported a relationship of SH with improved OS which did not reach significance probably due to low incidence of SH in their GaC study cohort (LNs from only 30 (5%) out of 592 patients showed SH) [31]. Similarly, Lu et al. used a qualitative adaptation of Black’s SH grading and observed no association of SH with clinical outcome in a large oesophageal squamous cell carcinoma patient cohort (186 (18%) out of 1,033 patients with SH positive LN) [40]. Following Black’s SH evaluation strategy, Eriguchi observed higher SH incidence in LNs from Bacillus Calmette–Guérin (BCG)-treated GaC patients compared to control [41].

Okamura and Kodama have used the SH grading system (none versus slight versus obvious) proposed by Cottier [27]. Both studies reported a relatively low SH incidence in GaC cohorts (21 (15%) out of 141 patients in Kodama’s study, 26 (14%) out of 189 patients in Okamura’s) and no relationship between presence of SH and survival [24, 42].

A potential clinical value of follicular hyperplasia (FH) was suggested by seven studies both in GaC and OeC patients, for details, see Table 1. Similar to reports on SH, the lack of an universally accepted grading system is reflected by a variety of FH grading approaches (Table 2). Black et al. proposed a binary classifier based on the presence or absence of secondary follicles in the LN cortex defined as a germinal center (GermC) surrounded by a mantle zone [31], whereas Oka et al. proposed to account for varying extent of FH in the LN by distinguishing GermC presence in the cortex versus medullary area [30]. LNs exhibiting both, FH and paracortical hyperplasia (PH) have been associated with a longer 5-year survival in a GaC study by Kodama et al. Okamura’s results suggest that presence of concurrent PH and FH is related to an increased number of tumour-infiltrating lymphocytes in primary GaC [24, 42]. Black et al. found an association between presence of FH and favorable survival [31].

A relationship between LN area occupied by GermCs in tumour-negative LNs of OeC patients and survival was recently confirmed by Kloft et al. [43]. In the same study, it was suggested that LN compartment hyperplasias might contribute to an increase in LNneg diameter, which was found to be an independent favorable prognostic marker.

The expansion of the paracortical area, termed paracortical hyperplasia (PH), is histologically visible as an increase in lymphocytes located between the cortex and medulla of the LN. The presence of high endothelial venules (HEVs) in this region has been proposed as a criterion for PH by some investigators [27, 44]. In the study by Syrjanen et al., PH together with high cuboidal shape of endothelial cells in HEV located in the paracortex was associated with better overall survival (OS) and lower number of regional LN metastasis in GaC patients. Findings of an experimental, translational approach by Kojima et al. revealed a higher lymphocyte reactivity upon antigen stimulation in cell cultures if lymphocytes were derived from GaC patients LNs exhibiting paracortical activation [44, 45]. Likewise, Riegrova et al. found that increased PH was associated with increased survival in GaC patients [38].

The immune cell signature of tumour draining LNs was investigated by immunohistochemistry (IHC) in twelve studies. The density of dendritic cells (DCs), known for their antigen-presenting capacity and T cell response induction, was measured as IHC-positive cell density per area or as total cell count and shown to be associated with lower number of LN metastases and lower grade of tumour differentiation in regional LNs in GaC and OeC studies, see Supplementary Tables S4 and S5 [46‐48]. No significant differences in the extent of DC density were observed between tumour-positive and tumour-negative sentinel LNs in GaC studies by Ishigami et al. and Lee et al., see Supplementary Table S4 [48, 49]. Ishigami et al. reported no significant cell density differences in natural killer (NK) cells between sentinel and non-sentinel LNs in GaC neither in tumour-negative nor in tumour-positive LNs [49]. A significant association was reported between the presence of LN metastasis and the density of FOXP3+ T-regulatory cells (Tregs) in the LNs of GaC patient by Lee et al. [50]. CD169+ macrophages in LNs are functionally dependent on DCs, and higher numbers of CD169+ macrophages in LNs have been associated with favorable outcome in OeC patients [51, 52]. Takeya showed an association between LN resident CD169+ sinus macrophages (LySMs) and higher tumour infiltrating lymphocyte (TIL) counts in patients after neoadjuvant therapy, suggesting host immune reactivity induced by cancer cell phagocytosis by LySMs [51]. Besides cancer cells, immune-suppressive macrophages from the tumour microenvironment (TME) are also well-established modulators of LN reactions. The studies by Go et al., Hiramatsu et al., and Tokumoto et al. showed that the presence of tumour associated CD163+ macrophages (TAMs) and CD15+ neutrophils (TANs) in tumour draining LNs was related to poor survival in GaC patients [53‐55].

Otto et al. has reported a significant down-regulation of DKK1 in the subcapsular sinuses and in  endothelial cells of pre-metastatic regional LNs in OeC patients [56]. Overexpression of the IDO gene, coupled with downregulation of tumour suppressing BIN1 in LNs was associated with poor prognosis in OeC patients in a study by Jia et al. [57]. O’Sullivan et al. reported a significant suppression of the proliferation of LN-derived lymphocytes in oesophageal squamous cell carcinoma compared to oesophageal adenocarcinoma cases, suggesting cancer subtype-specific immune response mechanisms in LNs [58].

Five GaC studies with well-defined LN reactive patterns and sufficiently large number of patients were used for meta-analysis (Black et al. 1971, Kodama et al. 1976, Oka et al. 1981, Riegrova et al. 1982, Oka et al. 1992) [24, 30, 31, 37].

Our meta-analysis showed a statistically significant association between the presence of SH and favorable outcome with pooled risk ratios (RR) and 95% confidence intervals (CI) of 2.73 [1.23–6.05] and 1.36 [1.01–1.82] for 1- and 5-year survival, respectively, as well as 1.78 [1.19–2.66] for 5-year OS when analyzing results from tumour-negative LNs (Fig. 3). Forest plots showing the study-specific weights and p values are provided in Supplementary Figure S2A-O. Furthermore, the meta-analysis showed a significant association between the presence of FH and favorable 1-year OS with RR [95% CI] of 1.09 [1.03–1.16] and 1.85 [1.32–2.59] in tumour-negative and all investigated LNs, respectively. A significant association with favorable 5-year OS was detected when pooling results from tumour-negative and tumour-positive LNs exhibiting PH with RR [95% CI] of 1.43 [1.09–1.88]. The studies exhibited high heterogeneity with respect to age, sex distribution, and geography of selected patients, rendering the data eligible for random-effects model.

To compare the incidence of LN reactive patterns between tumour-negative and tumour-positive LNs, the incidence of reactive LN patterns was pooled from all available GaC studies. Our analysis showed that tumour-negative regional LNs had a twice higher incidence of reactive pattern compared to tumour-positive regional LNs (see Table 3). Although a similar analysis was not feasible in OeC studies due to missing data per LN metastatic state, the pooled SH and FH frequencies from three OeC studies showed FH in 54.9% LNs compared to SH in 19.6% LNs.