Integrative Analysis of TLS‐Associated Gene Signatures, Immune Infiltration and Drug Sensitivity in Pancreatic Cancer

ABSTRACT

Pancreatic adenocarcinoma (PAAD) remains highly lethal because of chemotherapy resistance and immunosuppressive microenvironments. Tertiary lymphoid structures (TLSs) were analysed in PAAD to develop personalised therapeutic strategies. Nine TLS‐related genes (CCR6, CD1d, CD79B, CETP, EIF1AY, LAT, PTGDS, RBP5 and SKAP1) were selected for integrative analysis of TLS status in relation to clinical outcomes, immune cell infiltration, tumour mutational burden (TMB) and drug resistance. High TLS scores (TLS_H) were associated with improved overall survival (OS) and progression‐free survival (PFS), independent of age or tumour grade. Twelve immune cell types differed across TLSs. Single‐cell RNA‐seq analysis revealed that the 9 TLS‐related genes were enriched in distinct immune cell populations. Combining TLS and TMB improved survival prediction. Notably, the TLS_H group demonstrated enhanced sensitivity to chemotherapeutics including AZD8055, axitinib, vorinostat, nilotinib, camptothecin and paclitaxel. Real‐time fluorescent quantitative PCR (RT‐qPCR) validation in Mia PaCa2 and Jurkat cells indicated that LAT, RBP5 and SKAP1 may play important roles in modulating sensitivity to these chemotherapeutics. These findings establish TLS as a potential biomarker for PAAD, enabling personalised chemotherapy selection by integrating immune contexture and genomic drivers to improve clinical outcomes.

Pancreatic adenocarcinoma (PAAD) has high lethality because of chemo‐resistance and an immunosuppressive microenvironment. The study selected nine TLS‐related genes and analysed the TLS status in patients with PAAD. High TLS scores were linked to better survival, enhanced immune cell infiltration and increased chemosensitivity. Findings suggest that using TLS as a biomarker for personalised PAAD treatment.

1. Background

Pancreatic adenocarcinoma (PAAD) is a digestive tract tumour with high malignancy and poor prognosis. The 5‐year survival rate is only 11.5% (from SEER database, 2014–2020). PAAD has an insidious onset; additionaly, the cancer cells proliferate rapidly to facilitate disease progression, and distal metastasis can occur at an early stage of tumour development [1]. However, less than 25% of patients are able to benefit from chemotherapy, and some patients who are relatively sensitive to chemotherapeutic agents also develop secondary resistance after several weeks of treatment, resulting in poor patient prognosis [2]. It is urgent to develop new therapeutic strategies to deliver benefit for patients with PAAD.

PAAD is characterised by a ‘cold’ tumour microenvironment (TME) and hard to benefit from immunotherapy [3]. Tertiary lymphoid structures (TLSs) are lymphoid‐like structures without encapsulation found in non‐lymphoid tissues, and are composed of immune cells and high endothelial venules [4]. Recent studies suggest that TLSs might modulate systemic immune responses against pancreatic ductal adenocarcinoma (PDAC) carcinogenesis, leading to favourable outcomes of immune checkpoint inhibitors (ICI) treatment in PDAC patients [5]. Similarly, in patients treated with adjuvant chemotherapy, the development of TLSs is usually associated with improved treatment response [5, 6]. However, the influence of the TLSs on chemotherapy resistance and immune checkpoint genes of PAAD has not yet been elucidated.

Tumour mutational burden (TMB) predicts the efficacy of immunotherapy in PDAC. TMB‐high cases (≥ 10 mut/Mb) displayed prolonged survival and strong anti‐tumour immune responses fuelled by T helper cell/DC‐mediated priming of the cytotoxic T cells [7]. The accumulation of tumour protein p53 (TP53), KRAS proto‐oncogene, GTPase (KRAS), cyclin dependent kinase inhibitor 2A (CDKN2A) and SMAD family member 4 (SMAD4) alterations has been major drivers of pancreatic carcinogenesis and poor prognosis [8]. The expression of oncogenic KRAS leads to the recruitment of immunosuppressive cells [9]. These results suggest the importance of developing optimal treatment strategies according to the TLSs and TMB.

The nine TLS‐related genes (CCR6, CD1d, CD79B, CETP, EIF1AY, LAT, PTGDS, RBP5 and SKAP1) have been used as characteristic markers for TLS‐related studies in metastatic melanoma tumours [10], muscle‐invasive bladder cancers [11], papillary thyroid cancer [12], oesophageal carcinoma [13], breast cancer [14], gastrointestinal cancers [15] and pan‐cancer analysis [16]. Ying et al. (2024) used corresponding least absolute shrinkage and selection operator (LASSO) regression coefficients based on the 9‐gene signature to determine TLS scores in cell clusters of single‐cell data from PDAC samples, followed by investigations into survival analysis and chemosensitivity. However, experimental validation in cells was not conducted [17].

Here, we extracted data from study [11] from The Cancer Genome Atlas (TCGA) and Gene Expression Omnibus (GEO) to perform integrated analysis of TLS status in relation to clinical course, immune status, TMB status and chemotherapeutic drug resistance in patients with PAAD, and conducted validation studies of 9‐gene TLS signature in Mia PaCa2 and Jurkat cell lines treated with chemotherapeutic drugs. We hoped to provide more help for developing optimal treatment strategies according to the TLS status in patients with PAAD.

2. Materials and Methods

2.1. Data Acquisition

Transcriptome data of 182 PAAD samples sequenced using the Illumina HiSeq 2000 platform and normalised by log2(FPKM+1) were obtained from the University Of Cingifornia Sisha Cruz (UCSC) Xena (https://xenabrowser.net). After matching with the clinical information of pancreatic cancer samples downloaded at the same time, a total of 181 samples were included in the analysis (177 pancreatic cancer tumour samples and 4 normal control samples). This dataset was used as the training dataset in the analysis. R software (version 3.6.1) was used for subsequent analyses and the total workflow was shown in Figure 1.

FIGURE 1.

Flowchart of this study.

2.2. TLS Scores Assessment and Correlation With Clinical Characteristics

The limma package (https://bioconductor.org/packages/release/bioc/html/limma.html,Version 3.34.7) was used for analysis of expression level variation of nine TLS‐related genes between PAAD versus control samples (CTRL). TLS scores of each sample were assessed by gene set variation analysis (GSVA) Version 1.36.3 (http://www.bioconductor.org/packages/release/bioc/html/GSVA.html) [18], and the Kruskal–Wallis test was performed to compare the differences in the distribution of TLS scores between PAAD versus CTRL groups.

Next, all samples were divided into high and low TLS levels groups (TLS_H and TLS_L group) according to the median TLS score. The overall survival (OS) and progression free survival (PFS) were assessed by the Kaplan–Meier (KM) survival curve in the survival package Version2.41–1 (http://bioconductor.org/packages/survivalr/) [19]. Finally, the clinical independent prognostic correlation of TLS subgroup factors was examined by combining various clinical information of PAAD tumours using univariate and multifactorial Cox regression analysis [19].

2.3. The Correlation Between TLS Status and Tumour Immune Microenvironment

GSVA package (Version 1.36.3) in R software, based on the single‐sample gene set enrichment analysis (ssGSEA) algorithm was used to assess the proportion of individual immune cells for each sample included in the analysis [20]. The Kruskal–Wallis test was then used to compare the distribution of each immune cell in different TLS feature groupings. The immune checkpoint genes (MUC1, MUC4, ENO1, PVRIG, BTLA, C10orf54, CD274, CD276, CTLA4, HAVCR2, LAG3, PDCD1, PDCD1LG2 and TIGIT) were selected from the expression profile data and compared for their expression variability was compared among different TLS feature groupings.

2.4. Acquisition of Nine TLS Related Genes in Single Cell Data of PDAC

Single‐cell RNA‐seq data analysis was performed following the methodology described in Mo et al. (2022) with minor revisions [21]. For the PDAC GSE212966 dataset [22, 23], gene expression data underwent rigorous filtering and quality control: genes expressed in fewer than three cells per sample and cells with fewer than 200 expressed genes were excluded. Doublets were removed using the Scrublet (single‐cell remover of doublets) framework. Cells were filtered out if they exhibited a mitochondrial gene counts > 16.43%, a ribosomal gene counts > 41.71% or more than 6201.88 detected genes per cell. The data matrix was normalised for library size using Scanpy's ‘scanpy.pp.normalise_total’ function, and the log‐transformed normalised matrix was used for downstream analyses. The top 4000 highly variable genes were identified using ‘scanpy.pp.highly_variable_genes’. Effects of total counts per cell and mitochondrial/ribosomal gene expression percentages were regressed out via ‘scanpy.pp.regress_out’, and genes were scaled to unit variance with ‘scanpy.pp.scale’ (parameter: ‘max_value = 10’). Dimensionality reduction was initiated with principal component analysis (PCA), followed by batch effect removal using ‘sc.external.pp.harmony_integrate’ (parameter: ‘n_pcs = 42’). Uniform manifold approximation and projection (UMAP) was implemented via ‘scanpy.tl.umap’ to reduce the dimensionality of merged datasets, and cell clustering was performed using the Leiden algorithm on neighbourhood graphs.

2.5. The Correlation Between TLS Status and Drug Sensitivity

2.5.1. Drug Sensitivity Analysis and Pharmacogenomic Analysis

The sensitivity of each patient to drugs was based on TLS scores combined with the information in the Genomics of Drug Sensitivity in Cancer (GDSC) database (https://www.cancerrxgene.org/). pRRophetic was used to analyse the variability of IC50 values of various chemical drugs, which was then compared and presented between the TLS_H and TLS_L groups (http://127.0.0.1:22402/doc/html/Search?objects=1&port=22402).

2.5.2. The Effect of Chemotherapy Drugs on TLS Genes Expression

The effect of drugs screened from 2.4.1 (AZD8055, axitinib, vorinostat, nilotinib, camptothecin, paclitaxel, 10 μM, Targetmol, China) on TLS related genes was tested using real‐time fluorescent quantitative PCR (RT‐qPCR) in Mia PaCa2 and Jurkat cells. The total RNA was extracted (TIANGEN, DP120‐01) and the cDNA was transcript (YEASEN, 11151ES60). PrimerBank ID of nine TLS genes were as follow: CCR6 (150417990c1), CD1d (110618228c1), CD79B (90193589c1), CETP (169636438c1), EIF1AY (33356162c1), LAT (62739158c1), RBP5 (141802281c1) and SKAP1 (115527075c1). PTGDS‐specific primers were designed as follows: forward (5′‐GCTTCACAGAGGATACCATT‐3′) and reverse (5′‐TTGCTTCCGGAGTTTATTGT‐3′). The RT‐qPCR programme was as follows: 95°C 3 min, 95°C 10 s→55°C 30 s, 40 cycles, 65°C 30 s→65°C 15 s and 60 cycles with increments of 0.5°C per cycle.

2.6. Association Analysis of TLS Features and Mutations

The TMB and mutation‐related mutation annotation format files of PAAD samples were calculated using maftools (https://bioconductor.org/packages/release/bioc/html/maftools.html, Version 2.6.05) and the correlation between TMB and TLS features was calculated using the cor function [24]. The patients were divided into TMB high and low groups (TMB_H and TMB_L) according to the median TMB value. Then, the TMB groupings were combined with the previous TLS scores and divided into the following four groups: TLS_H&TMB_H, TLS_H&TMB_L, TLS_L&TMB_H and TLS_L&TMB_L. The association between the four different subgroups and OS and PFS was evaluated by the KM survival curve method [19], and then the differential distribution of the expression levels of immune checkpoint genes in the four different subgroups was compared.

The top 30 mutated genes in different TLS feature groupings were counted using the maftools package Version 2.6.05 [24]. The top 4 genes with the highest mutation frequency were assessed in combination with survival prognosis using the KM survival curve method [19].

2.7. KEGG Enrichment Analysis Associated With Risk Grouping

Based on the genome‐wide expression levels, the KEGG pathways that were significantly associated with TLS signature grouping were screened through GSEA. The significantly enriched KEGG pathways were screened by combining three key statistical values: enrichment score, normalised enrichment score (NES) and p < 0.05.

2.8. Predictive Effect of Externally Validated TLS Features on PAAD Patients

The pancreatic cancer‐related expression dataset GSE79668 [25] was downloaded from the GEO database and used as the validation dataset [26]. A total of 51 PAAD tumour samples were included. The detection platform was GPL11154 Illumina HiSeq 2000 (Homo sapiens).

TLS scores were calculated in the GSE79668 dataset using genome‐wide expression profiles, after which patients were stratified into distinct groups based on their TLS score distributions. The association between the TLS characteristics grouping and survival prognosis OS (this dataset only has OS information, not PFS) was examined in combination with the clinical prognosis information of the samples. Finally, the clinical independent prognostic correlations of the TLS grouping factors were examined by using the univariate and multivariate Cox regression analysis in the survival package version 2.41–1.

2.9. Statistical Analysis

Pearson correlation was used to assess the relationship between two normally distributed variables. Differences between or among groups with normally distributed variables were analysed using Student's t‐test or two‐way ANOVA. A two‐sided p value or an FDR q‐value below 0.05 indicated statistical significance.

3. Results

3.1. Prognostic Significance of TLS in TCGA PAAD Patients

To evaluate the prognostic significance of TLS, the expression levels of the nine TLS genes were analysed between the PAAD and CTRL groups. CCR6, CD1d, CD79B, CETP, EIF1AY, LAT, PTGDS and RBP5 were significantly downregulated in the PAAD group, whereas SKAP1 expression was lower in the CTRL group than in the PAAD group (Figure 2A). Higher SKAP1 expression has also been previously reported in primary pancreatic neuroendocrine tumours [27], colorectal cancer, malignant fibrous histiocytoma and breast cancer when compared with normal tissues [28]. Subsequently, TLS scores of the samples were evaluated via the GSVA algorithm. The distribution of TLS values was significantly lower in the PAAD group compared with that in the CTRL group (Figure 2B). Patients with PAAD were stratified into high (TLS_H) and low (TLS_L) TLS subgroups based on the median TLS value. KM analysis revealed that a high TLS score was significantly associated with longer OS and PFS (Figure 2C and D).

FIGURE 2.

Correlation of the TLS signature with prognosis in patients with PAAD. (A) The expression levels of 9 TLS genes between PAAD and CTRL groups. (B) The distribution of TLS values between PAAD and CTRL groups. (C) KM plots of OS between TLS_H and TLS_L groups in patients with PAAD. (D) KM plots of PFS between TLS_H and TLS_L groups in patients with PAAD. *0.01 ≤ p < 0.05, **0.001 ≤ p < 0.01, ***p < 0.001.

Moreover, age (HR = 1.022; 95% CI, 1.001–1.045; p < 0.05), neoplasm histologic grade (HR = 1.389; 95% CI, 1.028–1.876; p < 0.05) and TLS signature (HR = 0.611; 95% CI, 0.400–0.932; p < 0.05) were identified as independent predictors of OS. Likewise, neoplasm histologic grade (HR = 1.685; 95% CI, 1.230–2.308; p < 0.05) and TLS signature (HR = 0.564; 95% CI, 0.351–0.906; p < 0.05) were found to be significant independent predictors of PFS. These results indicate that a higher TLS score is correlated with a better prognosis (Table 1).

TABLE 1.

Univariate and multivariate cox analyses of prognosis of patients with PAAD in the TCGA cohort.

Variables	Univariate analysis			Multivariate analysis
	HR	95% CI	p value	HR	95% CI	p value
OS
Age (mean ± sd)	1.029	1.008–1.051	7.34E‐03	1.022	1.001–1.045	4.50E‐02
Gender (male/female)	0.809	0.537–1.219	3.10E‐01	—	—	—
Pathologic stage (I/II/III/IV)	1.197	0.831–1.725	3.35E‐01	—	—	—
Neoplasm histologic grade (G1/G2/G3)	1.453	1.091–1.934	1.03E‐02	1.389	1.028–1.876	3.25E‐02
Chronic pancreatitis history (Yes/No)	1.177	0.562–2.465	6.65E‐01	—	—	—
Diabetes history (yes/no)	0.927	0.532–1.615	7.89E‐01	—	—	—
Alcohol history (yes/no)	1.147	0.738–1.783	5.42E‐01	—	—	—
Tobacco history (never/reform/current)	1.092	0.892–1.337	3.92E‐01	—	—	—
TLS signature (high/low)	0.562	0.369–0.855	6.34E‐03	0.611	0.400–0.932	2.23E‐02
PFS
Age (mean ± sd)	1.019	0.996–1.043	1.11E‐01	—	—	—
Gender (male/female)	0.907	0.575–1.432	6.76E‐01	—	—	—
Pathologic stage (I/II/III/IV)	1.434	0.967–2.125	8.39E‐02	—	—	—
Neoplasm histologic grade (G1/G2/G3)	1.163	1.198–2.172	1.52E‐03	1.685	1.230–2.308	1.15E‐03
Chronic pancreatitis history (yes/no)	0.818	0.296–2.258	6.97E‐01	—	—	—
Diabetes history (yes/no)	0.918	0.501–1.683	7.81E‐01	—	—	—
Alcohol history (yes/no)	1.216	0.742–1.993	4.37E‐01	—	—	—
Tobacco history (never/reform/current)	1.098	0.883–1.365	4.00E‐01	—	—	—
TLS signature (high/low)	0.606	0.381–0.964	3.25E‐02	0.564	0.351–0.906	1.79E‐02

Note: The p values less than 0.05 are marked in bold black.

3.2. Correlation Between TLS Signature and Immune Microenvironment

TLS is comprised of B cells, enveloped by T cells and additionally encompasses T follicular helper (TFH) cells, dendritic cells and other cellular components [29]. A total of 12 immune cells showed significantly different distributions among the various TLS characteristic subgroups. Consistent with studies linking TLS to cytotoxic T cell priming [10], we found that the abundance of activated B cells, immature B cells, eosinophils, activated CD8⁺ T cells and T follicular helper cells was lower in patients with TLS_H. In contrast, monocytes, gamma delta T cells, central memory CD8⁺ T cells, activated dendritic cells, type 17 T helper cells (Th17), type 2 T helper cells (Th2) and neutrophils infiltrated more in patients with TLS_H (Figure 3A). This immune cell profile reflects a coordinated response to tumours, involving both immediate defence (neutrophils and monocytes) and sustained adaptive immunity (central memory CD8 T cells and Th17/Th2 cells) [30, 31]. Beyond mechanisms that solely promoting TLS formation, the TLS_H group exhibited a memory immune cycle in the tumour microenvironment [32].

FIGURE 3.

Correlation between TLS signature and immune microenvironment. (A) Heat map displaying of the distribution of various immune cells in patients with TLS_H and TLS_L; (B) Distribution of the expression levels of immune checkpoint genes in patients with TLS_H and TLS_L. *0.01 ≤ p < 0.05, **0.001 ≤ p < 0.01 and ***p < 0.001.

Meanwhile, the expression of immune checkpoint genes MUC1, MUC4 and ENO1 was lower in the TLS_H group, whereas BTLA, C10orf54, CTLA4, LAG3, PDCD1 and TIGIT showed a countervailing trend (Figure 3B). High expression of MUC1 and/or MUC4 correlated with decreased OS of patients with PDAC [33]. ENO1 promotes immunosuppression and tumour growth in pancreatic cancer [34]. In line with previous reports, the lower abundance of MUC1, MUC4 and ENO1 in the TLS_H group might be beneficial for prognosis, as infiltrated TIGIT‐B cells were suggested to be indicative of poor prognosis in gastric cancers [35]. This suggests a complex immune response involving both innate and adaptive immunity in the TLS formation.

3.3. Distribution of the 9 TLS‐Related Genes in Single Cells of PDAC

Subsequently, we interrogated the distribution of the 9 TLS‐related genes using single‐cell RNA‐seq data from the GSE212966 dataset [22, 23]. CCR6 was predominantly enriched in Treg cells, CD4⁺ T cells and B cells (Figure 4A), all of which are critical for TLS formation [4, 36]. Notably, CCR6 exhibited significantly higher expression in Treg and CD4⁺ T cells within PDAC samples compared to adjacent noncancerous pancreatic tissue (ADJ) (Figure 4K). CD1d was mainly expressed in macrophages (Figure 4B), mediating the presentation of glycolipid antigens to rapidly activate these macrophages and initiate critical immune crosstalk [37], but its expression was reduced in PDAC macrophages (Figure 4K). CD79B and EIF1AY were primarily localised in B cells (Figure 4C and E). CD79B, as a component of the B cell receptor, facilitates B cell activation and proliferation [38], which are critical for forming B cell follicles in TLS [36]. EIF1AY was found to be oncogenic and more related to glioblastoma occurrence [38], and may be involved in gender predominance of some immune diseases via its downregulation in healthy female B cells [39], though its function in PDAC requires further investigation. CETP and RBP5 were mainly enriched in endothelial cells (Figure 4D and H), with CETP silencing reducing endothelial oxidative stress, tumour necrosis factor (TNF) α levels, intercellular cell adhesion molecule‐1 (ICAM‐1) and vascular cell adhesion molecule‐1 (VCAM‐1) expression to diminish monocyte adhesion [40], whereas RBP5's function in endothelial cells needs further study. LAT was mainly expressed in mast cells, essential for mast cell stability [41] and also clustered in NK cells, Treg, CD4⁺ T and CD8⁺ T cells, indicating its important role in TLS formation (Figure 4F). PTGDS, analogous to its involvement in hepatic stellate cell activation [42], was predominantly expressed in pancreatic stellate cells but decreased in PDAC pancreatic stellate cells compared with the ADJ group (Figure 4G and K), whereas SKAP1 was mainly found in Treg cells, NK cells, CD8⁺ T cells, CD4⁺ T cells and mast cells (Figure 4I and K), reflecting its complex function in TLS and the cancer microenvironment. Fifteen cell clusters in PDAC were depicted in Figure 4J. This analysis further validated the cellular localization of the 9‐gene TLS signature across different cell lineages in the tumour microenvironment, providing mechanistic insights into their roles in TLS formation and function.

FIGURE 4.

Distribution of the 9 TLS‐related genes in single cells of PDAC. (A) Expression distribution of CCR6. (B) Expression distribution of CD1d. (C) Expression distribution of CD79B. (D) Expression distribution of CETP. (E) Expression distribution of EIF1AY. (F) Expression distribution of LAT. (G) Expression distribution of PTGDS. (H) Expression distribution of RBP5. (I) Expression distribution of SKAP1.(J) Fifteen cell clusters of PDAC. (K) Dot plot showing cell‐type‐specific expression of 9 TLS‐related genes in adjacent noncancerous pancreatic tissue (ADJ) and PDAC.

3.4. Correlation Analysis of TLS Features and TMB

TMB may lead to new tumour antigens, which could be beneficial for patients treated with ICIs [43]. The samples were divided into TMB_H and TMB_L groups according to the median value of TMB. The TLS_H&TMB_H subgroup was associated with improved OS (Figure 5A) and PFS (Figure 5B). The expression levels of six immune checkpoint genes (BTLA, C10orf54, CTLA4, LAG3, PDCD1 and TIGIT) were significantly lower in the TLS_L&TMB_H and TLS_L&TMB_L groups compared with those in the TLS_H&TMB_L group. Notably, only TIGIT was expressed at higher levels in the TLS_H&TMB_L group than in the TLS_H&TMB_H group (Figure 6). This indicates that the combined assessment of TLS and TMB is of great significance for predicting patient prognosis and immunotherapy response.

FIGURE 5.

Kaplan–Meier plots comparing the survival of TLS_H&TMB_H, TLS_H&TMB_L, TLS_L&TMB_H and TLS_L&TMB_L groups. (A) Correlation of TLS and TMB status with OS. (B) Correlation of TLS and TMB status with PFS.

FIGURE 6.

Immune checkpoint genes with differential distributions across subgroups of diverse TLS and TMB statuses.

3.5. Correlation Analysis of TLS Characteristics and Diver Gene Mutations

The mutation frequencies of driver genes between the two TLS signature groups were analysed. TP53 acted as the most frequently mutated gene in the TLS signature high cohort, with the mutation frequency up to 51% (Figure S1A, right panel). In contrast, KRAS was observed to be frequently mutated in the TLS signature low samples (Figure S1A, left panel). The genes with the top four mutation frequencies were TP53, KRAS, CDKN2A and SMAD4, which are classic oncogenes mutated in pancreatic cancer (Figure S1B). The OS of patients with wild‐type TP53, KRAS, CDKN2A and SMAD4 in the TLS signature high group was significantly better than that of patients with the same wildtype gene (p = 0.016, p = 0.002, p = 0.0001 and p = 0.018, respectively). Wild‐type TP53 may enhance immunogenicity, whereas mutant TP53 could disrupt TLS formation via stromal remodelling [8]. KRAS mutations, prevalent in patients with TLS_L, likely drive immunosuppression through myeloid cell recruitment [9], highlighting KRAS‐targeted therapies as potential therapeutic option in TLS_L cohorts.

Conversely, no significant differences in OS were found between the two TLS signature groups of patients with mutated TP53 (Figure 7A), KRAS (Figure 7B), CDKN2A (Figure 7C) and SMAD4 (Figure 7D). This finding provides support for the stratification of patients, underscoring the need for more detailed characterisation of mutational landscapes in PAAD.

FIGURE 7.

Plots of clinical prognostic correlation between top four high‐frequency mutations and unmutated samples. (A) TP53, (B) KRAS, (C) CDKN2A and (D) SMAD4.

3.6. KEGG Pathway Significantly Associated With TLS Feature Grouping

The screening for KEGG signalling pathways significantly associated with TLS characteristic subgroups identified enrichment in 11 pathways. Only the KEGG_ABC_TRANSPORTERS pathway was significantly associated with TLS feature groupings in pancreatic cancer, as identified using GSEA (p = 0.0475, Figure S2).

3.7. Susceptibility Analysis to Drugs

Chemotherapy is a safe choice for treating pancreatic cancer. To screen for effective chemotherapeutic agents for patients with different TLS statuses, the IC50 values of these agents were analysed using R 3.6.1 and the pRRophetic package. A total of 12 drug molecules were identified (Figure 8). A lower IC50 was observed for AZD8055, vinblastine, gefitinib, axitinib, vorinostat, nilotinib and camptothecin in the TLS‐High group. Meanwhile, a higher IC50 was observed for dasatinib, erlotinib, bortezomib, vinorelbine and paclitaxel in the TLS‐High group. A higher TLS signature appears to improve the sensitivity of patients with pancreatic cancer to AZD8055, vorinostat, gefitinib, axitinib, nilotinib and camptothecin.

FIGURE 8.

The IC50 values of 12 chemotherapeutic agents across different TLS status groups. *0.01 ≤ p < 0.05, **0.001 ≤ p < 0.01 and ***p < 0.001.

Next, we examined the expression of TLS genes in Mia PaCa2 and Jurkat cells treated with 6 chemotherapeutic agents using RT‐qPCR. CCR6, CD1d, LAT and RBP5 were expressed in higher level in Mia PaCa2 cells treated with AZD8055 (Figure 9). Only AZD8055 prompt E1F1AY expression among 6 chemotherapeutic agents in Mia PaCa2 cells. CCR6, CD1d, CD79B, RBP5 and SKAP1 expressed higher in Mia PaCa2 cells treated with axitinib. Vorinostat only improved the expression of CCR6 in Mia PaCa2 cells, but had variable effects on TLSs genes in Jurkat cells. CD1d, CD79B and LAT were inhibited by vorinota treatment, and PTGDS and RBP5 were expressed at higher levels under the same situation. Nilotinib upregulated RBP5 level only in Mia PaCa2 cells. It was notable that LAT showed variable expression in cells treated with different drugs. Erlotinib elevated LAT expression in Mia PaCa2, but LAT and CD79B exhibited diametrically opposed responses in Jurkat, indicating the distinct role of erlotinib in pancreatic cancer cells and immune cells. The camptothecin analogue SN38 repressed the expression of CD79B in Jurkat, whereas it upregulated the expression of SKAP1 in Mia PaCa2. However, extremely low or barely detectable expression of CETP was observed (data not shown). Although divergent TLS genes were differentially expressed in Mia PaCa2 and Jurkat cells with different drugs, AZD8055 and axitinib showed a tight association with TLSs.

FIGURE 9.

The expression of TLS genes in Mia PaCa2 and Jurkat cells treated with chemotherapeutic agents; *0.01 ≤ p < 0.05, **0.001 ≤ p < 0.01, ***p < 0.001 and ****p < 0.0001.

3.8. External Validation of the Prognostic Significance of TLS in PAAD Patients

External validation is a crucial step towards the reproducibility, generalisability and clinical implementation of a prediction model [44]. To externally validate the prognostic effect of TLS characteristics in pancreatic cancer, the TLS scores were calculated based on the genome‐wide expression levels in the GSE79668 dataset. The high TLS features showed a significant positive correlation with survival prognostic OS (Figure S3). Whereas, the age and TLS features were independent prognostic risk factors for patients with pancreatic cancer this dataset (Table 2).

TABLE 2.

External validation of the prognostic significance of TLS in GSE79668 dataset.

Variables	Univariate analysis			Multivariate analysis
	HR	95% CI	p value	HR	95% CI	p value
Age (years)	1.026	1.003–1.053	4.72E‐02	1.021	0.999–1.053	5.67E‐02
Gender (male/female)	1.238	0.665–2.304	5.00E‐01	—	—	—
Pathologic_M (M1/M0)	1.357	0.184–10.03	7.64E‐01	—	—	—
Pathologic_N (N1/N0)	1.442	0.724–2.869	2.98E‐01	—	—	—
Pathologic_T (T4/T3/T2/T1)	1.351	0.919–1.982	1.25E‐01	—	—	—
Diabetes (yes/no)	0.765	0.419–1.395	3.81E‐01	—	—	—
TLS signature (high/low)	0.536	0.291–0.986	4.17E‐02	0.523	0.291–0.988	4.57E‐02

Note: The p values less than 0.05 are marked in bold black.

4. Discussion

In the tumour environment, TLSs promote immune cell infiltration into solid tumours, and thus the development of TLS is significantly associated with survival in untreated patients with PAAD [45, 46]. To expand the application of TLSs in the clinical treatment of pancreatic cancer, we calculated TLS scores based on 9 TLS‐related genes (CCR6, CD1d, CD79B, CETP, EIF1AY, LAT, PTGDS, RBP5 and SKAP1) signature in the TCGA database, and then analysed the correlation between TLS scores and clinical prognosis, immune microenvironment, TMB status, and chemotherapeutic drug resistance, respectively. Consistent with previous reports [17], a high TLS score was significantly associated with better OS and PFS in PAAD patients (Figure 2C and D and Table 1).

Notably, SKAP1 was highly expressed in the low TLS score group and was positively associated with tumour development [47]. As a T‐cell adaptor protein, SKAP1 regulates T‐cell function [48]. Mechanistically, SKAP1 promotes the formation of neutrophil extracellular traps (NETs) by upregulating C‐X‐C motif chemokine ligand 8 (CXCL8) via nuclear factor of activated T cells c1 (NFATc1) [49], a process that leads to neutrophil depletion [50]. This mechanism aligns with our observations of increased activated T cells and decreased neutrophil counts in the TLS_L group (Figure 3A). Notably, inhibiting SKAP1‐induced NETs has been shown to enhance the antitumour efficacy of adoptive natural killer (NK) cell therapy in colon tumour models [49].However, the effect of SKAP1 on TLS formation has not been clearly reported, and further investigation is warranted.

The spatiotemporal expression of the nine TLS‐related genes enables multi‐level regulation of TLS: (1) Initial immune cell homing via chemokine receptors [4, 36]; (2) Germinal center maturation through B/T cell co‐stimulation [38]; (3) Structural stabilisation via endothelial–stromal crosstalk [37, 51]. The differential expression patterns in PDAC versus normal tissue suggest that TLS dysfunction may arise from disrupted gene‐cell interactions (Figure 4K), providing mechanistic insight into why TLS_H correlates with enhanced antitumour immunity [52, 53]. These findings establish the nine‐gene signature as both functional drivers and molecular markers of TLSs, paving the way for targeted modulation of TLS formation to improve immunotherapeutic responses.

Only 10%–20% of patients with PAAD are eligible for surgical resection [54]. More than 90% of patients without adjuvant postoperative treatment experience recurrence and die [55]. First‐line adjuvant therapy for PAAD is based on FOLFIRINOX (folic acid, fluorouracil, irinotecan and oxaliplatin) or gemcitabine chemotherapy regimens in combination with albumin‐bound paclitaxel. Patients with a high TLS score showed greater susceptibility to AZD8055, vinblastine, gefitinib, axitinib, vorinostat, nilotinib and SN38.

Alterations in the MYC pathway that trigger changes in the tumour microenvironment may represent a potential mechanism by which TLSs influence tumor drug resistance. MYC pathway activation drives resistance to mTOR inhibitors such as AZD8055 in breast cancer [56]. Vorinostat in combination with anti‐PD‐1 therapy reduces tumour growth by suppressing c‐Myc [57]. Concurrently, the biological process of ribosome biogenesis and the MYC_TARGET_V2 hallmark pathway are enriched in vorinostat‐resistant tumours [58]. It was reported that the inhibition of the aspartate β‐hydroxylase/MYC axis enhances chemokine (C‐X‐C motif) ligand 13 (CXCL13) secretion, thereby promoting TLS formation [59, 60]. These interconnected mechanisms collectively underlie the heightened sensitivity to AZD8055 and vorinostat observed in TLS_H tumors (Figure 8). Our GSEA analysis further revealed a trend toward RIBOSOME pathway enrichment in TLS_H samples, despite lacking statistical significance, this suggests subtle perturbations in ribosomal function that may potentiate vorinostat sensitivity (Figure S2). Nilotinib exhibits remarkable interaction with the Asp420‐Lys426 region of MYC [61], but there is limited direct or indirect evidence linking it to TLS. Similarly, axitinib and camptothecin also lack substantial evidence connecting them to TLS, thus warranting further investigation into the mechanisms of action of these three drugs. Gefitinib is a first‐generation EGFR inhibitor, and KRAS mutations are one of the key drivers of gefitinib resistance [17]. The TLS_H subgroup exhibited fewer KRAS mutations (Figure S1), which may explain their enhanced sensitivity to gefitinib.

LAT responds actively to these drugs, with the exception of axitinib and SN38 (Figure 9). Genetic defects for LAT were reported to cause severe immunodeficiencies and defective T‐cell signalling [62]. It was reported that LAT inhibitors may enhance the efficacy of gemcitabine in malignant pancreatic cancers through the mTORC1 and GAAC signalling pathways [63]. TLS_H patients exhibited greater sensitivity to mTOR inhibitors (AZD8055) and tyrosine kinase inhibitors (axitinib), potentially because of LAT upregulation, a transporter linked to nutrient uptake in proliferating cells [63]. Conversely, resistance to paclitaxel in TLS_H patients may reflect stromal barriers in ‘cold’ tumours [3]. These findings emphasise the need for TLS‐stratified therapies that combine immunomodulators with targeted agents.

The current study relies on bulk RNA‐seq with GSVA to assess TLS status and the Scrublet algorithm for single‐cell clustering of the nine TLS‐related genes, approaches that have inherent limitations. The bulk RNA‐seq method may obscure cell‐type specific expression patterns, while single‐modal analysis lacks integration of multi‐omic datasets. This limitation hinders the full characterisation of TLS formation mechanisms across diverse biological contexts. Advanced algorithms such as scCross and the Dynamic Hypergraph Hyperbolic Neural Network (DHHNN) could uncover fine‐grained cellular heterogeneity and gene regulatory networks [64, 65]. The deep learning models scHiCyclePred [66] and scHiClassifier [67], specifically developed for high‐throughput chromosome conformation capture (Hi‐C) sequencing analysis, enable in‐depth exploration of TMB by deciphering genome‐wide chromatin interactions. Additionally, the study focused only on a restricted set of TLS‐related genes and immune cell types. A more comprehensive analysis of TLS‐associated molecular networks and cell populations is essential to decipher their role in PAAD, which would facilitate the development of targeted immunotherapies.

5. Conclusion

In summary, our findings reveal that a high TLS score, derived from nine TLS‐related genes (CCR6, CD1d, CD79B, CETP, EIF1AY, LAT, PTGDS, RBP5 and SKAP1), correlates with improved prognosis in PAAD patients. Patients with high TLS scores exhibited enhanced immune cell infiltration and reduced immunosuppressive markers. Moreover, the combined analysis of TLS status and TMB revealed a synergistic effect, with the TLS_H&TMB_H subgroup showing the most favourable prognosis. Additionally, the spatiotemporal expression of the nine TLS‐related genes may facilitate TLS formation and enhance sensitivity to targeted agents such as AZD8055 and axitinib, thereby expanding the treatment options available for these patients. Overall, our findings support TLS as a biomarker for personalised PAAD therapy. Specifically, integrating immune contexture and genomic drivers can expand the selection of therapeutic agents, thereby improving clinical outcomes.

Author Contributions

Mengzhou Gao: data curation, formal analysis, investigation, writing – original draft. Guohui Li: data curation, investigation, validation. Xin Wang: investigation, validation. Xueyun Wang: data curation, formal analysis, writing – original draft. Danning Tang: investigation, validation. Xiang Ao: conceptualization, methodology. An Luo: project administration. Zhenguo Wen: methodology, writing – review and editing. Teng Wang: conceptualization, supervision, writing – review and editing. Zhaojun Jia: conceptualization, funding acquisition, methodology, resources, software, writing – review and editing.

Ethics Statement

The authors have nothing to report.

Consent

The authors have nothing to report.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting Information S1

Figure S1: Mutation frequency of driver genes between two TLS signature groups. (A) Differences in the landscape of driver gene mutations in PAAD patients with differentclinical characters. (B) Differences in mutational frequencies of the top 30 driver genes between the high and low TLS signature groups.

Figure S2: The plot of KEGG signaling pathways associated with TLS feature groupings.

Figure S3: The KM plot of TLS feature grouping versus OS correlation in the validation dataset GSE79668 shows significant results.

Gao, Mengzhou , Li Guohui, Wang Xin, et al. 2025. “Integrative Analysis of TLS‐Associated Gene Signatures, Immune Infiltration and Drug Sensitivity in Pancreatic Cancer.” IET Systems Biology: e70038. 10.1049/syb2.70038.

Data Availability Statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.