Introduction
Gastric cancer (GC) is a major cause of cancer-related mortality globally (Sung et al.
2021), with exceptionally high disease burden in East Asia (Etemadi
2017). GC is featured by high heterogeneity at histological, cellular, (epi)genomic, and proteomic levels, accompanied by distinct clinical outcomes (Wadhwa et al.
2013; Ge et al.
2018). Despite good response to treatment in early-stage GC, advanced GC is highly aggressive, with a median overall survival (OS) time of around 10 months (Smyth et al.
2020). In recent years, immunotherapy has revolutionized the treatment landscape of advanced GC and significantly extended patients’ survival (Li et al.
2021a). However, only a limited population benefited from immunotherapy, and a wide variation of response rates was reported, calling for better biomarkers for stratified treatment (Janjigian et al.
2021; Shitara et al.
2020).
Proteomic markers such as PD-L1 expression (Kim et al.
2018) and molecular markers such as microsatellite instability (MSI) (Kwon et al.
2021) have been established as potential biomarkers of prognosis and responses to immune checkpoint inhibitors (ICIs), whereas findings between different clinical trials were inconsistent. Recently, components in the tumor microenvironment (TME) of GC, such as the contents (Ren et al.
2021) and spatial location (Chen et al.
2022) of different cell types, and specialized structures like tertiary lymphoid structure (TLS) (Yu et al.
2022), have received increasing attention for they could affect prognosis and immunotherapy efficiency.
Composed of cellular aggregates in non-lymphoid organs under inflammatory conditions like infection and tumor, TLSs show analogical functional and structural features with lymph nodes (Schumacher and Thommen
2022). Mature TLS is characterized by the B-cell zone that involves the germinal center and is surrounded by the follicular helper T cells. The T-cell zone containing dendritic cells and high endothelial venules is also crucial to TLS (Sautès-Fridman et al.
2019). TLSs signify privileged regions for immune cell maturation and antigen presentation, serving as the crucial milieu for anti-tumor immunity. Emerging evidence indicated that TLS presence strongly correlated with higher immunoreactivity and better clinical outcomes of GC. For instance, Li and colleagues reported that TLS presence was indicative of favorable OS based on a cohort containing 63 GC cases (Li et al.
2020), and Yin et al. further showed that TLS was a promising predictor for longer survival of Epstein-Barr Virus (EBV)-associated GC (Yin et al.
2022). Moreover, Jiang et al. proposed that TLS positively correlated with superior response of ICIs based on a cohort containing 13 GC samples (Jiang et al.
2022). However, some studies were limited by small sample sizes or specific GC subtypes, thus may lack generalizability to some extent. Moreover, associations between TLS and prognosis and therapeutic sensitivity of GC are controversial, probably because samples from different ethnicities and different detecting approaches for TLS were applied.
Consequently, we conducted a meta-analysis to clarify the prognostic and predictive values of TLSs in GC. Simultaneously, we performed bioinformatic validation to capture the biological underpinnings by TLS-related gene signature in external cohorts. Our comprehensive analyses provided the latest evidence for the relationships between TLS and GC, probably conveying a powerful biomarker for clinical practice.
Methods
Guidance and protocol
The present study was conducted based on Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) (Liberati et al.
2009), and the protocol was registered at the Prospective Register of Systematic Reviews (PROSPERO ID CRD42023413227).
Search strategy
Two authors (H.X.P. and X.R.W.) independently searched publicly available databases, including the Cochrane Library, Embase, PubMed, and Web of Science, to retrieve suitable studies before April 2, 2023. The references of identified articles were also reviewed to seek potential research. The search strings are presented in Supplementary Table 1.
Included and excluded criteria
Studies were regarded eligible if: (1) focusing on GC populations; (2) examining TLSs in situ of tumor samples by immunohistochemistry (IHC) or hematoxylin–eosin (H&E) staining; (3) evaluating the associations between TLSs and survival or therapeutic response of GC; (4) publishing in English with available full-text. Studies were excluded if: (1) sample size < 10; (2) comments, conference abstracts, or letters to the editor; (3) outcome data could not be obtained or estimated.
Study selection
Independent authors H.X.P. and X.R.W. screened titles and abstracts to obtain eligible studies, the full texts of which were further reviewed. Disagreements were addressed via discussion with senior investigators (Y.C.).
Researchers (H.X.P. and X.R.W.) independently utilized standardized forms to collect data, such as sample size, clinicopathologic characteristics, and TLS location and detecting approaches. Outcome measures were also extracted, including hazard ratio (HR) with corresponding 95% confidence interval (CI) and number of responders/non-responders to ICIs.
Quality assessment
The Newcastle–Ottawa Scale (NOS), with scores ranging from 0 to 9, was exploited to estimate the quality of the included study. Two independent authors (H.X.P. and X.R.W.) performed the workflow, and differing opinions were resolved by consensus. A study with a NOS score greater than 6 was determined high-quality.
Evaluating correlations between TLSs and clinicopathologic parameters of GC
Clinicopathologic data of GC samples, including age, tumor size, T stage, N stage, pTNM stage, and differentiation status, were extracted and re-classified into a high and low group of each parameter. Specifically, age of 50, tumor size of 5 cm, T1 + T2/T3 + T4 stage, N0/N1 stage, I + II/III + IV pTNM stage, and poor/moderate or well differentiation were used as cut-off values. Then, chi-square and Fisher’s exact tests were exploited to compare differences between groups with or without TLSs.
Data synthesis
Statistical analyses were performed on Stata (version 15) and R (version 4.3.1) software. Considering the between-study differences in detecting and quantifying TLS, TLS levels were utilized uniformly to report the findings. For outcome measures of prognosis, HRs and corresponding CIs were pooled. For ICIs response, outcome measures were expressed and pooled as odds ratio (ORs) and 95%CIs. Inter-study heterogeneity was estimated by
I2 statistic, and it was considered notable if
I2 ≥ 50% (Higgins and Thompson
2002). The random-effects model was adopted if substantial heterogeneity was observed. Otherwise, the fixed-effects model was utilized. Statistical significance was defined at
P < 0.05.
Publication bias and sensitivity analyses
The funnel plot test, Begg’s test, and Egger’s test were employed to assess publication bias, which was further tested and adjusted by the trim-and-fill method (Duval and Tweedie
2000). Sensitivity analysis was performed by removing each study one by one.
Biological validation of TLS signature
The genomic profiles, mRNA expression, T-cell receptor (TCR) and B-cell receptor (BCR) repertoire, and clinical characteristics data of GC samples (
n = 443) from the TCGA-STAD cohort were downloaded from the UCSC Xena (
https://xena.ucsc.edu/) database. The single-sample Gene Set Enrichment Analysis (ssGSEA) approach was adopted to calculate the enrichment scores of TLS by the nine-gene signature (CCL19, CCR7, CETP, CXCL10, CXCL11, CXCL13, CXCL9, LAMP3, SELL) as previously reported (Cabrita et al.
2020; Hou et al.
2022).
As for immune infiltration estimation, enrichment scores of 29 immune signatures were computed via the ssGSEA method (He et al.
2018), and the abundance of 22 immune cell lineages was quantified through the CIBERSORT algorithm (Newman et al.
2015) based on bulk RNA-seq data.
Predicting ICIs efficacy of the TLS signature
The clinical and transcriptomic data were collected from the PRJEB25780 cohort, in which metastatic GC patients (
n = 61) were treated with pembrolizumab monotherapy as later-line therapy (Kim et al.
2018). Forty cases (65.6%) had more than two sites of metastasis and nearly half of them had previously undergone second-line therapy. Twenty-eight (45.9%) and twelve (19.7%) patients had PD-L1 combined positive score of more than 1 and 5, respectively. Moreover, 6 patients were tested to be EBV positivity, and 7 patients held MSI-H status. The TLS-score represented by the nine-gene signature was calculated for each sample in the same manner to discover the predictive effects of TLS-score on ICIs efficacy.
Statistical analysis
Mann–Whitney U test and chi-square test were applied to compare categorical and continuous data, respectively. Correlation analysis was conducted via Spearman’s test. Visualization of survival differences was generated by Kaplan–Meier curves and tested through the log-rank test. The threshold of statistical significance was set as P < 0.05.
Discussion
Through a comprehensive meta-analysis of 11 studies containing 4,224 GC cases, we pinpointed that TLS correlated with favorable prognosis and ICIs sensitivity of GC. Biological validation in the TCGA-STAD and PRJEB25780 cohorts further corroborated that TLS presence signified higher immunoactivity in TME.
The prognosis of GC is known to be affected by tumor- and host-correlated characteristics, such as age, pTNM stage, and histologic subtypes. Consequently, we first interrogated the relationships between TLS levels and clinicopathologic features. Results showed that higher TLS levels correlated with smaller tumor size and earlier T and N stages, consistent with the findings in breast cancer (BC) (Wang et al.
2022) and lung cancer (Rakaee et al.
2021). Interestingly, higher TLS levels were discovered in diffuse and mix than intestinal subtypes of GC. Since TLS is distinguished by B-cell enriched regions, a recent single-cell atlas also documented significantly higher B-cell infiltration as a salient feature of DGC (Kumar et al.
2022). Wang et al. also reported that TLS presence predicted higher tumor-infiltrating lymphocyte (TIL) levels of BC by meta-analysis (Wang et al.
2022). Nonetheless, we could not assess whether TLS levels correlate with higher infiltration of TILs in meta-analysis due to the nature of the data.
Systematical meta-analysis indicated that higher TLS levels strongly predicted favorable OS of postoperative GC, with HR of 0.36 and 0.55 of univariate and multivariate Cox analysis, respectively. Sensitivity analyses further validated the stability and robustness of our findings. Noteworthy, high heterogeneity was found between the included studies. Thus, subgroup analyses stratified by the TLS detecting methods, sample size, and median age of the included cases were carried out to interrogate potential heterogeneity. Strikingly, the prognostic value of TLS remained significant across subgroups and was more notable in younger than elder GC cases. The heterogeneity decreased as expected, whereas it remained modest. Moreover, meta-analysis implicated that GC patients with high TLS levels significantly benefited from anti-PD-1 inhibitors as later-line therapy than those with low TLS levels. However, considering the limited included cases and retrospective design of studies, the predictive effects of TLS on ICIs sensitivity need to be interpreted cautiously.
The maturation degree of TLS is postulated to affect its clinical significance by recent work. For instance, mature TLS with GC predicted significantly longer survival than total TLSs, whereas such prognostic effect attenuated when the formation of GC was damaged (He et al.
2023; Ling et al.
2022). Meanwhile, the prognostic value of TLS varied by its location in the tumor tissue. Intratumor TLS seemingly demonstrated a stronger prognostic effect than peritumor TLS, whereas findings were inconsistent between different cancer types (Sofopoulos et al.
2019; Li et al.
2021b). There is evidence that TLS levels remarkably varied from early to advanced stage of cancer (Sautès-Fridman et al.
2019). Additionally, studies showed that TLS levels significantly attenuate in the metastatic sites compared to the primary tumors, and can even be absent (Lee et al.
2019). And patients who hold TLS both in the primary tumor and metastatic site exhibit superior prognosis (Cipponi et al.
2012). However, we could not perform comprehensive subgroup analyses on the maturation degree or spatial location of TLS due to the nature of data. Therefore, future studies characterizing TLS with compositional, spatial, and functional details are imperative and encouraged.
Considering the unique puzzles in identifying and quantifying TLSs via traditional methods like H&E staining and IHC, we further validated the findings from meta-analysis and discovered the underlying biological underpinnings via TLS-relevant gene signature, which has been verified in multiple cancer types (Cabrita et al.
2020; Hou et al.
2022; Feng et al.
2021). Strikingly, higher TLS levels were found in DGC than in other subtypes, in congruence with the meta-analysis. Despite without significance, a trend of better prognosis was also observed in DGC individuals with higher TLS levels. Moreover, TLS level was proven to be an independent prognostic factor of intestinal GC.
The TME landscapes concerning TLS levels were subsequently depicted. Higher immune infiltration of major immune effector cells, such as T and B lymphocytes and natural killer cells, were observed in the high TLS-level group than in low ones, indicative of an “immune-hot” TME. Future studies integrating TLS levels and immune infiltration features within TME may offer a more comprehensive and robust prognosticator. Interestingly, higher mast cell (MC) infiltrates were found in the TLS-low group. Evidence showed that MCs could stimulate regulatory T cells to facilitate GC progression (Lv et al.
2023). Studies also reported that inhibition of the degranulation of MC attenuated the development of GC, signifying a potential target (Gunjigake et al.
2021). We also parsed the relationships between the TLS levels and the diversity and richness of the immune repertoire, which represents the strength and breadth of immune responses and acts a paramount role in anti-tumor immunity (Jiang et al.
2019). Higher diversity and richness of TCR and BCR were discovered in the TLS-high group, representing higher antigen presentation function.
Higher TLS levels also correlated with upregulated immune checkpoint genes expression. This may also partly explain why ICIs boost strong antitumor immunity in cancers with enriched TLSs (Petitprez et al.
2020). Intriguingly, ICIs could also instigate TLS formation. For instance, Sarah et al. reported the accumulation of TLS-correlated B cells in responders after neoadjuvant ICIs of melanoma (Helmink et al.
2020). Moreover, the genetic portraits significantly differed between different TLS-level groups. Elevated mutational frequencies of several genes that correlated with immune infiltration were observed in the TLS-high group. For example, a higher mutation rate of ARID1A, a tumor suppressor gene that is relevant to the MSI feature of cancers (Mullen et al.
2021), was found in the TLS-high group. ARID1A-mutated GC held higher TMB and PDL1 levels and favored higher immune cell infiltrates (Li et al.
2019). In brief, high TLS levels represent high immunogenicity and immunoactivity, possibly driving benefits from immunotherapy.
Eventually, we interrogated whether TLS levels could predict ICIs response in the PRJEB25780 cohort, in which GC patients received later-line pembrolizumab monotherapy. Intriguingly, more responders were identified in the TLS-high group than the low ones. Moderate accuracy in predicting benefit from ICIs of TLS-level was presented (AUC > 0.75), higher than the MSI and EBV-status, which are established biomarkers indicative of immunotherapy response (Bai et al.
2022; Yu et al.
2022).
The present work firstly and comprehensively offered substantial evidence for the clinical significance of TLS in GC by meta-analysis and biological validation. Meanwhile, several limitations should be noted. First, different scoring approaches and thresholds in evaluating high/low TLS levels were utilized in different studies. However, we could not perform corresponding subgroup analyses due to unavailable data, probably leading to bias. Second, the pooled sample size for discovering and validating the predictive effect of TLS on ICIs response was limited, thus may lack robustness. Third, high heterogeneity among studies caused potential publication bias in the meta-analysis. Additionally, all the included researches were retrospectively investigated and may risk intrinsic structural biases. Moreover, findings concerning the biological underpinnings of TLS were still at the speculative and analytic stage based on gene signature, without in vivo and in vitro functional validation.
Future studies should focus on establishing a common standard for identifying and quantifying TLS and future prospectively validating it in randomized trials for better clinical applications. Second, pinpointing dynamic changes of cellular components and location within TLS during immunotherapy leads to better comprehending its biological implications. Moreover, inducing the formation of TLS, like by intratumoral injection of vital cytokines such as CXCL13 (Delvecchio et al.
2021), administration of engineered cells (GeurtsvanKessel et al.
2009) and tumor vaccines (Zhang et al.
2021), may provide a neoteric perspective for synergistic immunotherapeutic method.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.