Correlation of Neuroendocrine Differentiation with a Distinctively Suppressive Immune Microenvironment in Gastric Cancer

Abstract

Introduction

Neuroendocrine neoplasms (NENs) harbored significantly suppressive tumor immune microenvironments (TIMEs). However, the immunological effects of neuroendocrine differentiation (NED) on non-NENs, such as gastric cancer (GC), were unknown.

Methods

Between pure gastric cancer (PGC) and GC-NED, TIME features were scored based on expression data and validated on serial whole-tissue sections of surgical samples, with tertiary lymphoid structures (TLSs) and the extra-TLS zone evaluated independently using multi-marker immunohistochemical staining. Risk analyses of TIME features on tumor behaviors were performed in GC-NED. The universal immunological effects of NED were explored preliminarily in adenocarcinomas arising in other organs.

Results

Based on over 11,500 annotated TLSs and 2,700 extra-TLS zones, compared with PGC, GC-NED harbored a distinctively more suppressive TIME characterized by increased but immature TLSs, with higher naïve B-cell and follicular regulatory T-cell densities and downregulated TLS maturation-related cell ratios inside TLSs; increased naïve B-cell and regulatory T-cell densities; and a high proportion of exhausted T cells in the extra-TLS zone. The upregulated tumor PD-L1 expression and its close correlations with TLS formation and maturation were remarkable exclusively in GC-NED. TIME features, especially those regarding TLSs, were significantly correlated with tumor growth and invasion. The desynchrony between TLS formation and maturation and increased naïve or regulatory immune cell infiltration was observed in adenocarcinomas of the colorectum, pancreas, lung, and prostate.

Conclusion

NED highlighted a distinct GC entity with more suppressive TIME features correlated with tumor behaviors, indicating a cohort that would benefit more from immunotherapies.

Introduction

Gastric cancer with neuroendocrine differentiation (GC-NED) is a special type of gastric adenocarcinoma that expresses NE-related proteins, such as synaptophysin and/or chromogranin A, but without the NE morphology [1]. In clinicopathological practice, the positive immunohistochemical (IHC) staining for NE markers is relatively common in GC, and the prevalence of GC-NED ranged from 11.0% to 39.6% [2, 3].

Previous studies have demonstrated that, compared to pure gastric cancer (PGC), GC-NED possessed more aggressive tumor behaviors, including deeper invasion, larger tumor size, higher tumor stages, and more frequent node or distant metastasis [4, 5]. Nevertheless, there have been no specific treatment guidelines for GC-NED, most of which underwent chemotherapies using routine regimens against gastric adenocarcinomas, such as SOX, XELOX, or FOLFOX [6–8], when it is necessary. In this setting, the prognosis of GC-NED patients was significantly worse than that of PGC [2, 4, 9].

Recently, immunotherapy, especially the application of immune checkpoint inhibitors (ICIs), has attracted more and more attention in treating advanced gastric adenocarcinoma [10, 11], of which efficacy is closely associated with features of tumor immune microenvironments (TIMEs). TIMEs are complex immune networks where immune cells, structures, and molecules exert their functions against tumors. Tumor infiltration lymphocytes (TILs) are the most famous TIME component [12, 13]. Nevertheless, recent research reported that features of the “local home/school” for infiltrating immune cells – tertiary lymphoid structures (TLSs) – showed more profound effects on orchestrating anti-tumor immune responses and modulating the efficacy of immunotherapies [14–16].

Since TIMEs are surrounded by tumor cells and tissue fluid, their composition and function are susceptible to tumor-specific factors that help different tumors shape their distinctive TIMEs. This phenomenon has been found in various types and grades of neuroendocrine neoplasms (NENs) [17–20]. However, regarding non-NENs but where ectopic NED occurred, such as GC-NED, their TIMEs have not been touched, and whether GC-NED harbored a different TIME from that of PGC was unknown. This gap hindered the more profound understanding of GC-NED and confined the feasibility evaluation of immunotherapy or the selection of specific immunomodulatory drugs against GC-NED.

Hence, to fill the gap above, based on mRNA expression data and serial whole-tissue sections (SWTSs) of surgical samples, we comprehensively profiled and compared the TIMEs of GC-NED and PGC, from the aspects of two immune compartments, the TLS zone and the extra-TLS zone, using multi-marker IHC staining and image analysis. Furthermore, the significance of TIME features on GC-NED behaviors was evaluated, and the immunological effects of NED were also preliminarily validated in other adenocarcinomas.

Materials and Methods

Data Availability and mRNA-Based TIME Profiling

mRNA expression, demographic, and clinicopathological data of stomach adenocarcinoma (STAD, TCGA, PanCancer Atlas), colorectal adenocarcinoma (COAD, TCGA, PanCancer Atlas), lung adenocarcinoma (LUAD, TCGA, PanCancer Atlas), pancreatic adenocarcinoma (PAAD, TCGA, PanCancer Atlas), and prostate adenocarcinoma (PRAD, TCGA, PanCancer Atlas) were obtained from cBioPortal (https://www.cbioportal.org/). For each tumor type, according to their expression levels of NE markers (SYP and CHGA), cases were divided into the NED group, whose expression levels were higher than the upper tertile, and the non-NED group, whose expression levels were lower than the lower tertile.

The immune cell infiltration was estimated with MCPCOUNTER, XCELL, QUANTISEQ, CIBERSORT-ABS, EPIC, and TIMER algorithms on the TIMER2.0 platform [21–23]. TLS formation and maturity scores were calculated with single-sample Gene Set Enrichment Analysis (ssGSEA) based on the corresponding gene sets reported previously [24–26] (online suppl. Table 1; for all online suppl. material, see https://doi.org/10.1159/000534427) with GenePattern 3.9.11 [27].

Patient Selection and Propensity-Score Matching

Surgical pathology reports of GC patients who underwent radical gastrectomy between 2015 and 2020 in the Second Affiliated Hospital of Zhejiang University School of Medicine were reviewed. Patients who met all the following criteria were enrolled as candidates for the subsequent propensity-score matching: (i) no neoadjuvant chemo- and (or) radiotherapy was performed; (ii) the pathological diagnosis was primary gastric adenocarcinoma; and (iii) the IHC staining for Syn and CgA was performed (tumors with ≥10% cells positive for either marker were defined as GC-NED). To minimize the influence of demographic and clinicopathological confounding factors on TIME comparisons, PMS was applied between the PGC and GC-NED groups according to the variables of sex, age, tumor location, histological differentiation, and pathological stages (pStages) by the matching ratio of 1:1 using the algorithm of Nearest Neighbors with calipers of width equal to 0.02 in R software v4.1.0 (The R Project for Statistical Computing, Vienna, Austria).

Immunohistochemical Staining

By reviewing the H&E slides, a representative tissue block that contained the tumor and adjacent normal tissues was chosen for IHC staining in each case. On the SWTSs from the blocks above, IHC staining for CD20 (OTI4B4, ZSGB-Bio, Ready-to-use), IgD (EPR6146, Abcam, 1:3,500), CD21 (EP3093, Abcam, 1:500), CD23 (EPR3617, Abcam, 1:400), CD8 (SP16, ZSGB-Bio, Ready-to-use), FOXP3 (EPR22102-37, Abcam, 1:250), PD-1 (EPR4877(2), Abcam, 1:500), EOMES (EPR21950-241, Abcam, 1:1,000), NCR1 (EPR22403-57, Abcam, 1:1,000), CD163 (EPR19518, Abcam, 1:500), and PD-L1 (E1L3N, Cell Signaling Technology, 1:200) was performed with the two-step polymer-based detection system (PV-8000, ZSGB-Bio). IHC slides passing the quality control were scanned into a computer with an Aperio Digital Pathology Slide Scanner (Aperio Technologies, Vista, CA, USA) at ×200 magnification for subsequent analyses.

SWTS-Based TIME Profiling

The TIME of each tumor was divided into two immune compartments, the TLS zone and the extra-TLS zone, which were assessed independently, as described previously [28]. Briefly, TLS locations were classified into intratumoral (those surrounded by tumor cells), stromal (those surrounded by tumor stromal cells), and peritumoral (those located between tumor and normal tissues) types. The number and area of every TLS were analyzed with QuPath v0.3.0 software (University of Edinburgh, UK) by one-by-one annotating, according to which the TLS area to tumor area ratio (TLS/tumor) and the average single-TLS size (the mean area of all TLSs in a tumor) were calculated for each case. Based on CD21 and CD23 staining, the maturity of TLSs was classified into early (CD21-CD23-), primary (CD21+CD23-), or secondary (CD21+CD23+), and then the fractions and densities of TLSs at each stage were calculated for each case, respectively. The fractions and densities per mm² TLS or per mm² tumor of intra-TLS immune cells (TLSICs), including CD20+ pan B cells (tB cell-pan), IgD+ naïve B cells (tB cell-naïve), CD8+ cytotoxic T lymphocytes (tCTLs), FOXP3+ follicular regulatory T (tTfr) cells, PD-1+ follicular T helper (tTfh) cells, and EOMES+ exhausted T (tTeom) cells, were measured with QuPath v0.3.0 software (University of Edinburgh, UK). The immune ratios, including tB cell-naïve/tB cell-pan, tTfh/tB cell-naïve, tTfh/tB cell-pan, tTfr/tB cell-naïve, tTfr/tB cell-pan, and tTfr/tTfh, were also calculated.

In the extra-TLS zone, five regions were selected randomly as the representatives at ×100 magnification for each case; the densities per mm² tumor of extra-TLS immune cells (ETICs), including CD20+ pan B cells (eB cell-pan), IgD+ naïve B cells (eB cell-naïve), CD8+ cytotoxic T lymphocytes (eCTLs), FOXP3+ regulatory T (eTreg) cells, PD-1+ exhausted T (eTpd-1) cells, EOMES+ exhausted T (eTeom) cells, NCR1+ natural killer (eNK) cells, and CD163+ tumor-associated macrophages (TAMs), were measured with QuPath v0.3.0 software (University of Edinburgh, UK). The immune ratios were also calculated, including eB cell-naïve/eB cell-pan, eTreg/eCTL, eTpd-1/eCTL, and eTeom/eCTL. In addition, according to the distribution of CD8+ eCTLs, the extra-TLS immune microenvironments were classified into immune-excluded (IEP), inflamed (INP), or immune desert phenotypes (IDP) [29]. The combined positive score (CPS), the density, and the proportion of PD-L1+ cells were used to evaluate PD-L1 expression [30].

Statistical Analysis

The demographic, clinicopathological features, TLS presence and location, and tumor immune phenotypes were compared using the χ² test or Fisher’s exact test. Quantitative variables following abnormal distributions were presented as medians (range) and compared using the U or Kruskal-Wallis test. Uniform Manifold Approximation and Projection (UMAP) analysis was performed on the SangerBox 3.0 platform [31]. Correlation analyses were performed with the Spearman test. The significance of immune features on clinicopathological features was evaluated with binary logistic regression. A p value <0.05 was considered statistically significant. Statistical analyses were performed with GraphPad Prism 9 (GraphPad Software LLC., San Diego, CA, USA) and SPSS 26.0 (SPSS Inc., Chicago, IL, USA).

Results

The Baseline of Patients for mRNA-Based TIME Profiling

Four hundred and seventeen STAD patients with complete demographic, clinicopathological, and mRNA sequencing data were obtained, of which 139 patients expressing high-level (>the upper tertile) NE markers were classified into the GC-NED group and 139 patients expressing low-level (<the lower tertile) NE markers were classified as the PGC group (shown in Fig. 1a). No significant difference was found between the two groups in baseline features, such as sex, diagnosis age, and pStage (shown in Fig. 1b).

Fig. 1.

mRNA-based TIME profiling and comparison. Comparison of SYP and CHGA expression (a); the baseline of patients for mRNA-based TIME profiling and comparison (b); differential immune cell scores with statistical significance (dot sizes indicate significantly higher or lower scores for each cell type) (c); comparison of TLS signature gene expression (d); comparison of TLS-related scores (e); correlations between TLS formation and maturity (f). *p < 0.05, **p < 0.001.

Higher Naïve and Resting Immune Cell Scores in GC-NED

According to corresponding signature gene sets, the immune cell scores, including T-cell, NK-cell, B-cell, and mononuclear phagocyte lineages, were estimated using multiple algorithms based on different calculation models. Overall, more immune cells, notably naïve and resting cells, scored significantly higher in GC-NED than in PGC.

In the T- and NK-cell lineages, scores of total CD4+ T cells, naïve CD4+ T cells, CD4+ resting memory T cells, total CD8+ T cells, and Treg cells significantly increased in GC-NED; in contrast, CD4+ Th1 cells, CD4+ effector T cells, CD4+ activated memory T cells, CD8+ effector memory T cells, gamma delta T cells, and all NK cells scored significantly lower in GC-NED. In the B-cell lineage, except for scores calculated by the algorithm of TIMER, all B-cell subtypes, such as total and naïve B cells, got significantly higher scores in GC-NED. In the mononuclear phagocyte lineage, GC-NED showed prominently higher monocyte scores, while the M0 and M1 macrophage scores decreased significantly. The M2 macrophage scores were discrepant between different algorithms but with a rising trend in GC-NED (shown in Fig. 1c).

Higher Formation but Lower Maturity Scores of TLS in GC-NED

To evaluate TLS formation and maturity, the expression of TLS signature genes and TLS formation or maturity score were compared using ssGSEA between PGC and GC-NED. The expression of TLS signature genes (shown in Fig. 1d) and TLS formation scores were significantly higher in GC-NED (shown in Fig. 1e); however, TLS maturity scores were markedly lower in GC-NED than those in PGC (shown in Fig. 1e). Meanwhile, there was a significantly positive correlation between TLS formation and maturity scores in PGC, which was absent in GC-NED (shown in Fig. 1f).

The Baseline of Patients for SWTS-Based TIME Profiling

A total of 460 PGC and 192 GC-NED patients who met the selection criteria were enrolled in the present study. Significant differences were found in tumor location, histological differentiation, and lymphovascular invasion (Table 1). To balance the patient baseline, propensity-score matching was performed between PGC and GC-NED patients. Thirty pairs of matched patients were obtained, with all statistical differences in baseline features removed (Table 1). The matched patients were aged from 39 to 87 years old, with a mean of 62.83 ± 9.78, of which 70% were male and 30% were female. Tumor sizes ranged from 1.0 to 13.0 cm, with a mean of 4.02 ± 2.15; 85% of the tumors occurred in the middle and lower 1/3 of the stomach, and the distribution of pStage was relatively even.

Table 1.

Patient baseline before and after PSM

Factor	Before matching			After matching
	PGC (n = 460)	GC-NED (n = 192)	p value	PGC (n = 30)	GC-NED (n = 30)	p value
Age, years, mean ± SD	62.54±11.34	61.77±10.74	0.421	62.78±7.90	62.88±11.49	0.968
Sex, n (%)
Male	304 (66.1)	133 (69.3)	0.465	20 (66.7)	22 (73.3)	0.779
Female	156 (33.9)	59 (30.7)		10 (33.3)	8 (26.7)
Tumor location, n (%)
Upper 1/3	49 (10.7)	35 (18.2)	0.006	5 (16.7)	4 (13.3)	0.683
Middle 1/3	131 (28.5)	62 (32.3)		9 (30.0)	13 (43.3)
Lower 1/3	280 (60.9)	95 (49.5)		16 (53.3)	13 (43.3)
Tumor size, cm, mean ± SD	4.37±2.41	4.17±2.45	0.355	4.09±2.46	3.94±1.83	0.790
Differentiation, n (%)
Well	19 (4.1)	22 (11.5)	0.001	3 (10.0)	1 (3.3)	0.721
Moderate	224 (48.7)	75 (39.1)		12 (40.0)	13 (43.3)
Poor	217 (47.2)	95 (49.5)		15 (50.0)	16 (53.3)
Depth of invasion, n (%)
T1	99 (21.5)	46 (24.0)	0.723	6 (20.0)	6 (20.0)	1.000
T2	69 (15.0)	26 (13.5)		8 (26.7)	8 (26.7)
T3	174 (37.8)	77 (40.1)		11 (36.7)	11 (36.7)
T4	118 (25.7)	43 (22.4)		5 (16.7)	5 (16.7)
LVI, n (%)
Negative	217 (47.2)	113 (58.9)	0.008	15 (50.0)	16 (53.3)	1.000
Positive	243 (52.8)	79 (41.1)		15 (50.0)	14 (46.7)
N, n (%)
Negative	180 (39.1)	88 (45.8)	0.117	13 (43.3)	13 (43.3)	0.807
Positive	280 (60.9)	104 (54.2)		17 (56.7)	17 (56.7)
M, n (%)
Negative	445 (96.7)	186 (96.9)	1.000	30 (100.0)	30 (100.0)	–
Positive	15 (3.3)	6 (3.1)		0	0
pStage, n (%)
I	132 (28.7)	62 (32.3)	0.697	9 (30.0)	9 (30.0)	0.946
II	131 (28.5)	57 (29.7)		11 (36.7)	13 (43.3)
III	182 (39.6)	67 (34.9)		10 (33.3)	8 (26.7)
IV	15 (3.3)	6 (3.1)		0	0

Bold: variables for PSM.

Increased TLS Formation but with Lower Maturity in GC-NED

Based on SWTSs and the IHC image analyses (shown in Fig. 2a), detailed and independent comparisons of the TIME compartments, including the TLS zone and the extra-TLS zone (shown in Fig. 2b), were conducted between PGC and GC-NED. More than 11,500 TLSs and 2,700 extra-TLS zones were annotated and measured.

Fig. 2.

Workflow of SWTS-based TIME profiling and comparison of TLS basic features. Schematic outline of SWTS-based TIME profiling and comparison (a); diagram of TIME compartments (b); morphology of TLSs on H&E and IHC staining for CD20 (arrows: germinal centers) (c); comparison of formation and maturity (d). ^#p < 0.05.

TLSs were present in 93.3% of tumors, most located at tumor stroma and peritumoral areas. The average count of TLSs per tumor was 19.17 ± 15.94, with a median size of 6.53*10 ^4 μm². On H&E slides, TLSs were present as nests of lymphocytes with small and dense nuclei, in some of which germinal centers indicated higher maturity. IHC staining for CD20 confirmed and highlighted TLSs at a higher resolution (shown in Fig. 2c).

Although TLS presence, location, and single-TLS size were comparable between PGC and GC-NED (online suppl. Table 2), significant differences were found in TLS variables involving tumor parameters: both the TLS density per mm² tumor and the TLS area to tumor area ratio (TLS/tumor) were markedly higher in GC-NED than in PGC (shown in Fig. 2d). Further, the maturation of TLSs in GC-NED was significantly delayed than those in PGC: the early TLS density per mm² tumor and the early TLS fraction were all prominently increased in GC-NED (shown in Fig. 2d).

More Naïve/Regulatory Immune Cells and a Higher Proportion of Exhausted T Cells in GC-NED TIMEs

Benefiting from the one-by-one annotations and measurements of TLSs, differences in TLS immune cells (TLSICs), including B-cell and T-cell lineages, were revealed between PGC and GC-NED. TLSICs were mainly composed of B cells and a small number of tCTL, tTeom, tTpd-1, and tTreg cells. In GC-NED, tB cell-pan, tB cell-naïve, and tTreg cell densities per mm² TLS or fractions were significantly higher, while tCTL, tTpd-1, and tTeom cell densities per mm² TLS or fractions were comparable to those of PGC (shown in Fig. 3a, b). Similarly, tB cell-pan, tB cell-naïve, and tTreg cell densities per mm² tumor, as well as tCTL, tTeom, and tTpd-1 cell densities per mm² tumor, were also increased markedly in GC-NED (shown in Fig. 3c). Immune ratios indicated the downregulation of TLS maturity in GC-NED, including higher tB cell-naïve/tB cell-pan, tTreg/tB cell-naïve, tTreg/tB cell-pan, and tTreg/tTpd-1, but lower tTpd-1/tB cell-naïve (shown in Fig. 3d).

Fig. 3.

SWTS-based TIME profiling and comparison. Comparison of TLS features (a–d); comparison of extra-TLS features and tumor immunophenotypes (e–g); comparison of tumor PD-L1 expression (h). *, ^#p < 0.05, **p < 0.001.

In the extra-TLS zone, extra-TLS immune cells (ETICs) mainly comprised TAMs with eCTL, eTemo, eTpd-1, eTreg, eB cell-pan, eB cell-naïve, and eNK cells. Among them, eB cell-naïve and eTreg cell densities per mm² tumor were significantly higher in GC-NED, while other immune cell densities were comparable between the two groups (shown in Fig. 3e). Meanwhile, the proportions of exhausted T cells (eTeom/eCTL), naïve B cells (eB cell-naïve/eB cell-pan), and negatively regulating T cells (eTreg/eCTL) increased prominently in GC-NED (shown in Fig. 3f). According to the infiltration patterns of eCTL, the immune phenotypes of TIME were determined. The main immune phenotype of TIME was IEP (47.6%) in GC-NED and INP (41.2%) in PGC, but there was no statistical significance (shown in Fig. 3g).

Higher PD-L1 Expression and Closer Interactions with TLS Formation and Maturation in GC-NED

To comprehensively compare PD-L1 expression between PGC and GC-NED, CPS, the PD-L1+ cell fraction, and density were analyzed in the present study. 90.0% of tumors showed positivity for PD-L1 to varying degrees. CPS, the PD-L1+ cell fraction, and density were all significantly higher in GC-NED than in PGC (shown in Fig. 3h). According to PD-L1 expression and other significantly differential TIME features (shown in Fig. 4a), GC-NED cases could be distinguished from PGC and clustered well in UMAP analysis (shown in Fig. 4b).

Fig. 4.

Pooled analysis of TIME features. Heatmap of significantly differential immune parameters between PGC and GC-NED (a); dimensionality reduction and clustering based on significantly differential immune parameters (b); correlations among immune parameters (c); significant risk factors on tumor growth and invasion of GC-NED (d). *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Correlation analyses showed that, in GC-NED, PD-L1 expression had markedly more correlations with TLS formation and maturation: PD-L1 expression was negatively correlated with the TLS density, TLS/tumor, the early TLS density, and the secondary TLS density. Intriguingly, these significantly negative correlations above only existed in GC-NED, while the correlation between PD-L1 expression and the TAM density was exclusive to PGC (shown in Fig. 4c).

TLS Features Were Closely Associated with GC-NED Behaviors

In regression models, TIME features, especially the TLS-related parameters, profoundly affected tumor behaviors. The tTpd-1 and tB cell-naïve densities per mm² TLS, the tB cell-naïve, tTpd-1, and eNK-cell densities per mm² tumor, the ratios of tTpd-1/tB cell-pan and tB cell-naïve/tB cell-pan were risk factors on tumor size, tumor invasion, lymphovascular invasion, and node metastasis. In contrast, the tCTL and tB cell-pan densities per mm² TLS, the tCTL, tTreg, TAM, and eB cell-naïve densities per mm² tumor, the ratios of tTpd-1/tB cell-naïve and tTreg/tB cell-pan were protective factors on tumor size, tumor invasion, and lymphovascular invasion (shown in Fig. 4d; online suppl. Tables 3–7).

NED Was Also Correlated with Suppressive TIMEs in Adenocarcinomas of Other Organs

To explore whether NED was associated with more suppressive TIMEs in other adenocarcinomas with NED as GC-NED, based on their expression data, we preliminarily assessed the immune cell infiltration, TLS formation, and maturity in adenocarcinomas where NED was commonly found in clinical practice, such as colorectal adenocarcinoma (COAD), pancreatic adenocarcinoma (PAAD), lung adenocarcinoma (LUAD), and prostate adenocarcinoma (PRAD).

According to the expression of NE markers, the tumors above were divided into non-NED and NED groups. A total of 197 COAD, 197 COAD-NED, 59 PAAD, 59 PAAD-NED, 170 LUAD, 170 LUAD-NED, 164 PRAD, and 164 PRAD-NED were enrolled in the TIME analyses. Dramatically, compared to their corresponding non-NED groups, TLS formation scores were significantly higher in all NED groups with significantly lower maturity scores. Meanwhile, the enrichment of inactivated or negatively regulating immune cells, such as naïve lymphocytes, Treg cells, and M2 macrophages, was also remarkable in NED groups (shown in Fig. 5a). In addition, NED was correlated with more differential immune cell scores in tumors occurring in hollow organs that communicate with the outside directly and harbor naturally active immune backgrounds, such as STAD, COAD, and LUAD (shown in Fig. 5b).

Fig. 5.

TIME comparison between non-NED and NED adenocarcinomas arising in other organs. a Significantly differential immune cell scores and TLS-related scores between non-NED and NED groups. b Relationships among immune cell sets with significantly differential scores between non-NED and NED groups of different tumors. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001.

Discussion

To the best of our knowledge, this is the first study focusing on the immunological effects of NED in gastric adenocarcinoma from the aspects of the two main TIME compartments, the TLS zone and the extra-TLS zone. In the present study, based on mRNA expression data and SWTSs of surgical samples, we first reported that the ectopic NED in non-NENs was significantly associated with a distinctive TIME. Compared to its non-neuroendocrine counterpart, GC-NED harbored a more suppressive TIME characterized by a prominent delayed TLS maturation, higher infiltration of naïve B cells, regulatory and exhausted T cells, and higher tumor PD-L1 expression.

It has been demonstrated that the TIMEs of NENs were significantly suppressive. Zhou et al. [20] compared the TIMEs of differentiated, poorly differentiated GC and gastric neuroendocrine carcinoma (NEC), showing that gastric NECs or the NEC component in mixed adenoneuroendocrine carcinoma had a higher proportion of malignant cells, increased INHBA+ TAM, but reduced CD8+ T-cell infiltration, which was accompanied by the downregulated interferon-related pathways. In colorectal NECs, Chen et al. [32] also reported lower scores of CD8+ T cells, B cells, myeloid dendritic cells, and macrophages/monocytes. Tannoet al. [18] compared the immune infiltration among pancreatic/ileal NETs, adjacent normal tissues, and non-small cell lung cancer, revealing that the CD8+ T-cell density was significantly lower in NETs than that of normal tissues or lung cancer. Meanwhile, Ferrata et al. [33] also found an immune ignorant TIME containing scarce T-cell infiltration in gastroenteropancreatic NETs with high proliferation. However, Takahashi et al. [34] showed significantly increased B cells, pan, and exhausted T cells in pancreatic NECs than in pancreatic adenocarcinomas or NETs. Inside NETs, B cells, CD204+ macrophages, and exhausted T cells rather than pan-T cells increased along with tumor grades.

In the extra-digestive system, Tian et al. [19] demonstrated a suppressive immune landscape in lung NECs, featured by dysfunctional and exhausted T cells, immunosuppressive TAMs, and dendritic cells expressing high-level IDO1. More interestingly, they found the heterogeneity of tumor cells and their corresponding TIMEs inside NECs: cell clusters with the NE phenotype had a colder TIME, characterized by lower immune scores and fewer enrichments of immune response-related pathways than those with non-NED phenotype, which emphasized the close correlation between the NE phenotype and TIME features, even inside NENs. In lung NETs (carcinoids), Bischoff et al. [17] found a comparable lymphoid microenvironment as normal tissues, while the noninflammatory monocyte-derived myeloid cells increased prominently.

In the present study, although as non-NENs, GC-NED indeed shared some suppressive TIME features as NENs. Like the research by Takahashi et al. [34] and Tian et al. [19], the densities and ratios of exhausted T cells increased significantly in both TLSs and extra-TLS zones; however, the decrease in CD8+ CTLs was not found. Furthermore, although the CD163+ TAM density increased in the GC-NED extra-TLS zone, it slightly missed the margin of significance (p = 0.056). In addition, an overall increase of Treg infiltration was observed in GC-NED, including their proportions, densities, and related immune ratios, which was also reported in cervical NECs with high PD-L1 expression [35].

TLSs are ectopic lymphoid structures, presenting as aggregates of lymphoid cells in or surrounding tumor tissues. Their features, including the presence, location, maturity, and cellular composition, are associated with anti-tumor immunity and patient survival in various tumors [36]. However, few studies focused on the TLS features in NENs or tumors with neuroendocrine phenotypes. A study by Zhang et al. [37] explored the infiltrating patterns and prognostic significance of TLSs in nonfunctional pancreatic NETs and found TLSs were present in more than one-third of tumors, mainly composed of CD20+ B cells and CD4+ T cells. And patients with TLSs showed significantly better prognoses than those without TLSs. A study by Cives et al. [38] also observed TLSs in one-fifth of small bowel NETs, of which presence was associated with high tumor PD-L1 expression. In the present study, TLSs were seen in 93.3% of tumors, most at stromal or peritumoral sites. There was no significant difference in TLS presence and location between PGC and GC-NED; however, benefitting from more detailed annotation and analyses, remarkably increased densities of TLSs, tB cell-naive, and tTreg, and lower TLS maturity were revealed in GC-NED, accompanied by disordered immune ratios regarding the TLS development. Further, the desynchrony between TLS formation and maturation above was also seen in other adenocarcinomas where NED was common in clinicopathological practice.

TLS formation follows a developing pathway similar to their counterpart, secondary lymphoid organs (SLOs), with complex interactions among cytokines, stromal cells, and immune cells involved. Briefly, the RORγt+ ID2+ lymphoid tissue inducer (LTi) cells firstly release lymphotoxin LTα1β2 to activate LTβR + lymphoid tissue organizer (LTo) cells. Activated LTo cells upregulate the expression of cell adhesion molecules and secrete the cytokines CXCL13, CCL19, and CCL21, which recruit CXCR5+ B cells and CCR7+ T cells to form lymphoid aggregation at peripheral sites [36, 39, 40]. Compared to SLOs, conditions for TLS formation are not so strict, and they can form themselves even in the absence of LTi cells but with the help of various types of cells that express CXCL13, CCL19, or CCL21 [41–43]. Hence, given the significantly elevated expression of CXCL13, CCL19, and CCL21 in GC-NED and the findings that neuroendocrine cells produced CXCL13 [44], an attractive microenvironment for B and T cells probably formed by high-level LTo cell-related cytokines, which might contribute to the B- and T-cell accumulation in GC-NED, providing a solid base for the marked TLS increase. However, the identification and confirmation of LTo cells and their functions are needed in the future.

Maturity is another critical feature of TLSs, which is determined by B-cell activation and differentiation and symbolized by the presence of germinal centers (GMCs). In some previous studies, TLS maturity was even more significant than TLS presence or density on tumor recurrence [45] and patient survival [46]. In peripheral tissues, after the recruitment by CXCL13, B cells recognize antigens and upregulate CCR7 expression, which helps them move to the T-cell zone rich in CCL19 and CCL21 and get costimulatory signals for the subsequent proliferation, differentiation, and germinal center (GMC) formation [47]. In this process, serial factors participate in the regulation of GMC formation, such as the cytokine gradient and the balance between follicular helper T (Tfh) cells and follicular regulatory T (Tfr) cells [47, 48]. In the present study, we found the delayed TLS maturation in GC-NED, accompanied by an increased proportion of tTreg cells, which served as Tfr cells, but a decreased proportion of tTpd-1 cells, which served as Tfh cells, inside TLSs. Combined with the high cytokine expression in GC-NED, we speculated that (i) high-level cytokines in TIMEs led to intense background noise, lowering the cytokine gradient difference between B-cell and T-cell zones and hindering the B-cell migration to the T-cell zone from getting costimulatory signals; (ii) the increased Tfr cells directly suppressed the Tfh cell function and B-cell development through CTLA-4/CD80/CD86 signaling pathways [48]; (iii) the increased Treg cells in the extra-TLS zone suppressed the T-cell activation stimulated by dendritic cells [48], further reducing the costimulatory signals for B-cell activation. Besides, the increased PD-L1 expression in GC-NED could inhibit the proliferation of Tfh and their migration to lymphoid follicles through the PD-1/PD-L1 pathway [49, 50], which tipped the scales in favor of Tfr further.

High tumor PD-L1 expression has been reported in NENs previously. A study by Yang et al. [51] found a high positive rate of 48.8% in 43 gastric NECs. A study by Robert et al. [52] observed positive PD-L1 expression in 32.4% of NECs of the digestive system and concluded that PD-L1 expression is frequent in NECs. In gastroenteropancreatic NETs, Rösner et al. [53] reported a positive rate of 73% in an IHC study with 457 cases enrolled and found increased PD-L1 expression in tumors with higher grades or proliferation. All these findings and ours revealed a close correlation between neuroendocrine phenotype and PD-L1 expression, suggesting that tumors positive for NE markers were a cohort that might probably benefit from immunotherapies with ICIs.

Meanwhile, we also noticed some distinctive correlations regarding PD-L1 expression in GC-NED or PGC. In GC-NED, PD-L1 expression was negatively correlated with TLS formation and maturation. Nevertheless, in PGC, its expression was positively correlated with TLIC and ETIC densities, such as TAMs, instead. In TIMEs of GC and NENs, tumor cells and TAMs are the primary sources of PD-L1 [53, 54], which were hard to distinguish due to their semblable cell morphology in IHC staining [54]. However, the correlations between PD-L1 expression and the TAM density above indicated that the main component of cells expressing PD-L1 was macrophages in PGC but tumor cells in GC-NED. More importantly, other exclusive correlations to GC-NED or PGC further demonstrated the different immunological effects of PD-L1 from different cells on TIMEs, which might help explain the differential immunotherapeutic efficacy in patients with the same PD-L1 expression, implying a more damaging role of PD-L1 in GC-NED and potentially more benefits from ICI therapies in this cohort.

There are some limitations to this study. First, to keep the spatial and morphological information about immune cells and TLSs, SWTSs were applied in our research. Still, tiny differences in cell location or TLS composition among sections could not be eradicated. Second, although the accepted specific markers and spatial distribution were used to identify immune cells, co-expression among different cells existed to varying extents, which might affect quantitative cell analyses. Third, one-by-one annotating TLSs manually ensured the measurement accuracy, but much human work and time were required, significantly limiting the acceptable sample capacity.

In summary, the present study first reported the immunological effects of NED on the non-NEN of the stomach. It revealed a more suppressive TIME in GC-NED, featured by desynchronized TLS formation and maturation, higher infiltration of naïve, regulatory, and exhausted immune cells, and higher PD-L1 expression, which was significantly correlated with tumor growth and invasion. These findings emphasized the distinctiveness of GC-NED from the aspect of the immune environment and indicated the potential benefits of immunotherapies in this cohort.

Acknowledgments

Grateful acknowledgment is made to the support of the Department of Pathology, Department of Medical Oncology, Second Affiliated Hospital Zhejiang University School of Medicine and Cancer Center, Zhejiang University. The results in silico here are based upon data generated by the TCGA Research Network: https://www.cancer.gov/tcga. Illustrations in Figure 2 were created using BioRender.

Statement of Ethics

This study was conducted ethically in accordance with the World Medical Association Declaration of Helsinki and approved by the Institutional Review Board of the Second Affiliated Hospital of Zhejiang University (2023-ERR-0357). Patient consent was waived by the institutional review boards as this study was retrospective and patients’ information was protected by a blind method.

Conflict of Interest Statement

The authors have no conflicts of interest to declare.

Funding Sources

Ying Yuan received funding from the Provincial Key R&D Program of Zhejiang Province (2021C03125), and Chunpeng Zhu received funding from the Medical and Health Science Program of Zhejiang Province (2022RC176).

Author Contributions

Conceptualization: Yi Zou, Ying Yuan, and Chunpeng Zhu; methodology: Yi Zou, Dan Li, Xiaoyan Yu, and Chenqi Zhou; software: Dan Li and Yi Zou; validation: Dan Li and Ying Yuan; formal analysis: Yi Zou and Dan Li; investigation: Yi Zou and Dan Li; resources, data curation, writing – reviewing and editing, supervision, project administration, and funding acquisition: Ying Yuan and Chunpeng Zhu; writing – original draft preparation and visualization: Yi Zou. All authors reviewed the manuscript.

Funding Statement

Ying Yuan received funding from the Provincial Key R&D Program of Zhejiang Province (2021C03125), and Chunpeng Zhu received funding from the Medical and Health Science Program of Zhejiang Province (2022RC176).

Data Availability Statement

Data are not publicly available due to ethical reasons. Further inquiries can be directed to the corresponding author.