B3GNT3 overexpression promotes tumor progression and inhibits infiltration of CD8⁺ T cells in pancreatic cancer

B3GNT3 overexpression predicts poor prognosis in PC

B3GNT3 was overexpressed in tumor tissues (P < 0.0001) according to GSE62452 dataset and GSE60979 dataset (Figure 1A, 1B). The Oncomine database (https://www.oncomine.org/resource/main.html) contained seven studies about PC, among which two studies revealed B3GNT3 expression in tumor tissues was significantly higher compared with that in normal pancreatic tissues (P < 0.05) (Figure 1C, 1D). The results of the other five studies in the Oncomine database are also shown in Supplementary Figure 1 (P > 0.05). Moreover, the overexpression of B3GNT3 in PC was also observed in the Gene Expression Profiling Interactive Analysis (GEPIA) database (https://gepia.cancer-pku.cn/) (Figure 1E). Moreover, in GSE28735 dataset, B3GNT3 expression in tumor tissues was significantly upregulated compared with that in the adjacent non-tumor tissues (P < 0.0001) (Figure 1F). The human protein atlas (HPA) database (http://proteinatlas.org/) demonstrated that the protein expression of B3GNT3 was notably upregulated in tumor tissues (Figure 2A, 2B), while the expression of B3GNT3 in CAPAN-2 cells was obviously higher than that in various other cancer cells (Figure 2C). In addition, B3GNT3 expression was mainly located in the cytoplasm of PC cells (Figure 2B). Of note, KM survival analysis according to the TCGA PC dataset (cutoff point = 41.5), GSE62452 dataset (cutoff point = 4.70), and GSE79668 dataset (cutoff point = 2860) demonstrated that patients with lower levels of B3GNT3 expression had a superior OS than those with higher levels of B3GNT3 expression (P < 0.05) (Figure 3A–3C). Taken together, our study suggests B3GNT3 as a critical but unfavorable prognostic factor for PC.

Figure 1. Multiple databases demonstrated that B3GNT3 was overexpressed in PC. (A–B) B3GNT3 expression in GSE62452 and GSE60979 datasets. (C–D) B3GNT3 expression in the Oncomine database. (E) B3GNT3 expression in GEPIA database. (F) B3GNT3 expression in GSE28735 dataset. PC, pancreatic cancer; N: normal; T: tumor; GEPIA: Gene Expression Profiling Interactive Analysis. (*P value < 0.05; **P value < 0.01;***P value < 0.001; ****P value < 0.0001).

Figure 2. The human protein atlas database analysis. (A, B) The translational differences of B3GNT3 between pancreatic cancer tissues and normal pancreatic tissues. (C) The expression of B3GNT3 in CAPAN-2 cells was much higher than that in various other cancer cells.

Figure 3. KM survival analysis for B3GNT3 through the TCGA, GSE62425, and GSE79668 datasets. (A–C) High expression of B3GNT3 was significantly associated with poor survival in PC (P value < 0.05). PC, pancreatic cancer; KM, Kaplan-Meier.

B3GNT3 expression and the correlation with clinicopathological characteristics

The correlation between the expression of B3GNT3 and clinicopathological characteristics in the TCGA PC dataset and GSE62452 dataset are shown in Tables 1, 2. B3GNT3 expression was significantly correlated with tumor size (P = 0.017) (Table 1) and histologic grade (P = 0.001) (Table 2). Patients with larger tumors (≥ 4 cm) and higher histologic grades had higher levels of B3GNT3 expression (P < 0.05) (Figure 4). These results further suggest that B3GNT3 overexpression may promote tumor progression in PC.

Figure 4. The different B3GNT3 expressions based on clinicopathological characteristics in the boxplot. Boxplots showing the expression differences of B3GNT3 according to tumor size (A) and tumor histologic grade (B, C). PC, pancreatic cancer; G1: grade 1; G2: grade 2; G3: grade 3. (*P value < 0.05; **P value < 0.01;***P value < 0.001; ****P value < 0.0001).

B3GNT3 expression and correlation with somatic mutation

Previous studies have demonstrated that KRAS, TP53, SMAD4, and CDKN2A mutations are four of the most common genetic hallmarks of PC [20–22]. Our study demonstrated that KRAS, TP53, CDKN2A, and SMAD4 mutation were significantly associated with higher levels of B3GNT3 expression (Figure 5A). Pearson correlation analyses (|Cor| > 0.3 and P < 0.05) were conducted in the TCGA PC, GSE62452, GSE60979, and GSE79668 datasets. We excluded the GSE28735 dataset from the correlation analysis because GSE28735 was actually part of the GSE62452 dataset. We found that B3GNT3 expression in the TCGA PC dataset was positively correlated with KRAS expression (Cor = 0.44, P < 0.0001), while it was negatively correlated with SMAD4 expression (Cor = ‒0.42, P < 0.0001) (Figure 5B). Similar results were also obtained in GSE62452 (Cor = 0.53, P < 0.0001 for KRAS; Cor = ‒0.70, P < 0.0001 for SMAD4) and GSE60979 datasets (Cor = 0.64, P < 0.0001 for KRAS; Cor = ‒0.51, P < 0.0001 for SMAD4) (Figure 5C, 5D). In addition, the associations among B3GNT3, KRAS, and SMAD4 in PC were also validated using the cBioportal database, the results of which were consistent with our findings (Supplementary Figure 2A, 2B). These findings suggest that B3GNT3 overexpression is correlated with the mutation and expression patterns of driver genes, especially for KRAS and SMAD4.

Figure 5. Association between B3GNT3 expression and somatic mutations. (A) KRAS, TP53, CDKN2A, and SMAD4 mutation status were significantly associated with higher expression of B3GNT3. (B–D) Correlation analysis of KRAS, SMAD4, and B3GNT3. PC, pancreatic cancer; TCGA, the Cancer Genome Atlas; WT: wild type; MUT: mutation. (*P value < 0.05; **P value < 0.01;***P value < 0.001; ****P value < 0.0001).

Functional enrichment analysis for B3GNT3

First, we used gene set enrichment analysis (GSEA) to determine whether these pathways and their associated genes differ between the high B3GNT3 and low B3GNT3 expression groups. We found that the T cell receptor signaling pathway-related gene set (NES = ‒1.75, P < 0.05, FDR < 25%) was enriched in the low B3GNT3 expression group (Figure 6A). Then, we conducted co-expression analysis (|Pearson correlation coefficient| > 0.75, P < 0.05) for B3GNT3 based on the TCGA dataset. As a result, 19 co-expressed genes for B3GNT3 were subsequently uploaded into ConsensuspathDB for pathway enrichment analysis (P < 0.05) (Figure 6B) [23]. B3GNT3 was found to be potentially involved in glycosphingolipid biosynthesis, leukocyte transendothelial migration, TGF-β signaling pathway, EMT, and TGF-β signaling pathway in EMT (Figure 6C). These findings indicate that B3GNT3 overexpression provides the necessary support for tumor growth and immune regulation of PC.

Figure 6. Functional enrichment analysis for B3GNT3. (A) GSEA revealed that T cell receptor signaling pathway related gene set (NES = -1.75, P < 0.05, and FDR < 25%) was enriched in the low B3GNT3 expression group. (B) Co-expression analysis for B3GNT3 in the TCGA PC dataset using Pearson correlation coefficients (Pearson correlated coefficient (Cor) > 0.75 or < -0.75, P value < 0.05). (C) Pathway enrichment analysis for B3GNT3. GSEA: gene set enrichment analysis; EMT: epithelial to mesenchymal transition; TCGA, the Cancer Genome Atlas; PC: pancreatic cancer.

B3GNT3 knockdown inhibits the proliferation, invasion, and EMT of PC cells

To further evaluate the functional role of B3GNT3 in tumor cell proliferation and invasion, we transfected PANC-1 and AsPC-1 cells with sh-B3GNT3. Abrogated levels of B3GNT3 in these two cells were validated using RT-qPCR analysis (P < 0.001) (Figure 7A). Then, the subclones KD1 and KD3 were selected for further experimental studies. The MTT assay revealed that cell proliferation was significantly inhibited in B3GNT3-depleted PANC-1 and AsPC-1 cells (P < 0.0001) (Figure 7B). Moreover, the transwell assay demonstrated that the number of invading cells was significantly decreased in the sh-B3GNT3 groups (KD1 and KD3) compared with those in the NC group (P < 0.0001) (Figure 7C). Western blot assay demonstrated that the expression of E-cadherin was increased, while the expression of N-cadherin and vimentin protein was decreased in the sh-B3GNT3 groups (KD1 and KD3) (Figure 7D). These findings suggest that the B3GNT3 knockdown significantly inhibits the proliferation, invasion, and EMT of PC cells.

Figure 7. The knockdown of B3GNT3 suppresses the proliferation, invasion, and EMT of PC cells. (A) RT-PCR analysis validated the knockdown of B3GNT3 in PANC-1 and AsPC-1 cells transfected with sh-B3GNT3. (B) Assessment of cell proliferation using the MTT assay. (C) Transwell assay was performed to determine the invasive capacity of PANC-1 and AsPC-1 cells transfected with sh-B3GNT3. (D) WB analysis to investigate the association between B3GNT3 and EMT in PC. PC: pancreatic cancer; EMT: epithelia-mesenchymal transition; WB: western blot. (*p<0.05; **p<0.01;***p<0.001; ****p<0.0001).

B3GNT3 overexpression inhibits the infiltration of CD8⁺ T cell in PC

The ESTIMATE algorithm showed that higher B3GNT3 expression was significantly correlated with higher tumor purity (Cor = 0.50, P < 0.0001), but lower immune score (Cor = ‒0.39, P < 0.0001) (Figure 8A), a finding that was also observed in the GSE62452 dataset (Cor = 0.33, P = 0.0052 for tumor purity; Cor = ‒0.29, P = 0.016 for immune score) (Figure 8D). Using the CIBERSORT algorithm, we found that B3GNT3 overexpression significantly correlated with lower CD8⁺ T cells infiltration within the TME of PC (Figure 8B). In the TCGA PC dataset, similarly, ssGSEA analysis indicated that B3GNT3 overexpression was notably correlated with lower infiltrations of CD8⁺ T cells (Cor = ‒0.47, P < 0.0001), TILs (Cor = ‒0.46, P < 0.0001), cytolytic activity (Cor = ‒0.41, P < 0.0001), T cell co-stimulation (Cor = ‒0.37, P < 0.0001), and Th1 cells (Cor = ‒0.37, P < 0.0001), but higher Th2/Th1 (Cor = 0.33, P < 0.0001) (Figure 8C), which was entirely consistent with those we observed in GSE62452 dataset (Cor = ‒0.34, P = 0.004 for CD8⁺ T cells; Cor = ‒0.29, P = 0.015 for TILs; Cor = ‒0.24, P = 0.043 for cytolytic activity; Cor = ‒0.37, P = 0.0018 for T cell co-stimulation; Cor = ‒0.36, P = 0.0026 for Th1 cells; Cor = 0.50, P < 0.0001 for Th2/Th1) (Figure 8D). Furthermore, similar results were also observed in the GSE79668 dataset, which showed that B3GNT3 overexpression was significantly correlated with a low infiltration of TILs (Cor = ‒0.28, P = 0.045), T cell co-stimulation (Cor = ‒0.37, P = 0.0083), and Th1 cells (Cor = ‒0.29, P = 0.042), whereas it was correlated with high Th2/Th1 (Cor = 0.32, P = 0.023) (Figure 8E). Furthermore, B3GNT3 expression was also negatively correlated with gene markers from CD8⁺ T cells (CD8A and CD8B), T cells (general) (CD3D, CD3E, and CD2), and Th1 cells (STAT1 and TBX21), but positively correlated with gene markers from Th2 cells (STAT6 and GATA3) (Figure 9). Similar results were observed in the GEPIA database (Table 3).

Figure 8. Association between B3GTN3 expression and the immune infiltration within tumor. (A) ESTIMATE algorithm to analyze the correlation between B3GNT3 expression and immune infiltration in PC. (B) CIBERSORT algorithm to analyze the correlation between B3GNT3 expression and immune infiltration in PC. (C) ssGSEA analysis to analyze the correlation between B3GNT3 expression and the immune infiltration in PC. (D) ESTIMATE algorithm and ssGSEA to analyze the correlation between B3GNT3 expression and immune infiltration within tumors in GSE62452 dataset. (E) ssGSEA analysis to analyze the correlation between B3GNT3 expression and immune infiltration within tumors in GSE79668 dataset. TCGA, the Cancer Genome Atlas; PC: pancreatic cancer; ssGSEA: single-sample gene set enrichment analysis; TILs: tumor-infiltrating lymphocytes; Th1 cells: Type-1 T helper cells; Th2/Th1: Type-2 T helper cells/ Type-1 T helper cells.

Figure 9. Correlation analysis of B3GNT3 expression and the expression of marker genes of CD8⁺ T cells (CD8A and CD8B), T cells (general) (CD2, CD3D, and CD3E), Th1 cells (STAT4 and TBX21), and Th2 cells (STAT6 and GATA3). Th1 cells: Type-1 T helper cells; Th2 cells: Type-2 T helper cells.

To validate our findings from the TCGA and GEO datasets, we investigated the correlation between the expression pattern of B3GNT3 and the infiltration of CD8⁺ T cell by multi-color immunofluorescence in an independent dataset of 50 PC cases. Cases with less than 5% tumor or loss of tissues were excluded for the quantification of B3GNT3. Representative tumor cores were available from 46 patients. The high B3GNT3 expression group displayed a significantly increased CD8⁺ T cell infiltration compared to the low B3GNT3 expression group (Figure 10A–10C). Moreover, we further proved that B3GNT3 overexpression positively correlated with a worse OS in PC patients, while a higher CD8+ T cells infiltration in TME positively correlated with a better OS of PC patients (Figure 10D, 10E). Taken together, these findings suggest that B3GNT3 overexpression may inhibit TILs in PC, especially CD8⁺ T cells.

Figure 10. Multi-color immunofluorescence to investigate the relationship between B3GNT3 expression and CD8⁺ T cells infiltration in PC. (A) High B3GNT3 expression in multi-color immunofluorescence. (B) Low B3GNT3 expression in multi-color immunofluorescence. (C) High B3GNT3 expression group displayed a significant increased CD8⁺ T cells infiltration compared to low B3GNT3 expression group. (D) Patients with higher B3GNT3 expression had significant shorter OS that those with lower B3GNT3 expression. (E) Patients with higher CD8⁺ T cells infiltration had superior OS than those with lower CD8⁺ T cells infiltration.

Acquisition of data

We obtained the RNA-sequencing data and corresponding clinical data of PC patients from The Cancer Genome Atlas (TCGA, https://cancergenome.nih.gov/) in February 2020. Mutation information for KRAS, TP53, SMAD4, and CDKN2A in the TCGG PC dataset were downloaded through the cBioportal database (http://www.cbioportal.org/). Of the 177 patients in TCGA PC dataset, 171 had an OS > 1 month. In addition, the relevant PC datasets were retrieved from the Gene Expression Omnibus (GEO) (http://www.ncbi.nlm.nih.gov/geo/) using the following search words: “pancreatic ductal adenocarcinoma” and “pancreatic cancer”. The exclusion criteria included: (1) studies involving only blood samples; (2) studies with fewer than 30 PC samples; (3) studies involving only PC cell lines or xenografts. Finally, four GEO datasets, including GSE62452 [GPL6244 (HuGene-1_0-st) Affymetrix Human Gene 1.0 ST Array (transcript (gene) version); 61 non-tumor samples and 69 pancreatic tumor samples], GSE60979 [GPL14550 SurePrint G3 Human GE 8x60K Microarray; 12 non-tumor samples and 49 pancreatic cancer samples], GSE28735 [GPL6244 (HuGene-1_0-st) Affymetrix Human Gene 1.0 ST Array (transcript (gene) version); 45 pairs of pancreatic tumor and adjacent non-tumor tissues], and GSE79668 (GpL11154 Illumina HiSeq 2000 (Homo sapiens); 51 pancreatic tumor samples], were selected for further analysis [47]. All datasets involved in this study are publicly available. Thus, local ethics approval was not required. A flowchart of our study is provided in Supplementary Figure 3.

Differential expression analysis for B3GNT3

Multiple datasets were utilized to evaluate B3GNT3 expression in PC. First, differential expression analysis for B3GNT3 was conducted using GSE62452 dataset and GSE60979 dataset. Then, the Oncomine database and GEPIA database were further used to analyze B3GNT3 expression in PC. Paired differential expression analysis for B3GNT3 was also conducted using GSE28735 dataset. Moreover, the HPA database was used to further validate B3GNT3 protein expression in PC samples and normal pancreatic tissues, and B3GNT3 expression in cancer cell lines.

Kaplan-Meier survival analysis

GSE62452, GSE79668 and TCGA datasets were used in survival analysis to assess the prognostic value of B3GNT3 expression in PC. We divided patients in these three PC datasets, respectively, into high- and low-B3GNT3 expression groups, according to the medium cutoff of B3GNT3 expression. KM survival curves were conducted using the “survminer” package in R.

Functional enrichment analysis

First, we performed GSEA to evaluate the differences in possible biological pathways between the high- and low-B3GNT3 expression groups in the TCGA PC dataset. An annotated gene set c2.cp.kegg.6.2.symbols.gmt obtained from the Molecular Signatures Database (MSigDB) was used as the reference gene set. Genes significantly related to B3GNT3 in the TCGA PC dataset were sniffed out with |Pearson correlation coefficient (Cor)| >0.75 and P < 0.05. These co-expressed genes of B3GNT3 were subsequently uploaded into ConsensuspathDB (http://cpdb.molgen.mpg.de/) for functional enrichment analysis, with P < 0.01 considered significant [23].

Cell culture and transfection

We purchased Human PC cell lines, PANC-1 and Aspc-1 from the Cell Bank of the Chinese Academy of Science (Shanghai, China). Cells were subsequently cultured in RPMI-1640 medium (Invitrogen, Carlsbad, CA, USA) containing 10% fetal bovine serum (FBS) (HyClone, Logan, UT, USA) and 1% penicillin/streptomycin in a humidified 5% CO₂ incubator at 37° C [48]. Cells were plated in 6-well plates at a density of 1 × 10⁶ cells per well before transfection [48]. Lentiviral small hairpin RNA (shRNA) targeting B3GNT3 was synthesized and cloned into the GV248 vector (Genechem Co., Ltd, Shanghai, China). As previous studies introduced [48, 49], at 80% confluence, the B3GNT3 lentiviruses (sh-B3GNT3) and the negative control shRNA (NC) were transfected into the cells using Lipofectamine 2000 (Invitrogen). At 48 h post-transfection, the cells were harvested for further analyses. The efficacy of sh-B3GNT3 for each cell line was assessed using qRT-PCR. The PCR primers were as follows: B3GNT3 forward, 5′-CAGCACGTTCAGAACTTCCTC-3′; B3GNT3 reverse, 5′-GCGCACATAGTTGCTAGGGG-3′; GAPDH forward, 5′-TGACTTCAACAGCGACACCCA-3′; GAPDH reverse, 5′-CACCCTGTTGCTGTAGCCAAA-3′.

MTT assay

The 3-(4,5-dimethylthiazal-2-yl)-2, 5-diphenyl-tetrazolium bromide (MTT) assay was used to measure PC cell proliferation. Cells were plated in 96-well plates at 2,000 cells/well and cultured for 1-5 days. At the indicated periods, 20 μL of MTT (5 mg/mL) (Sigma-Aldrich, St. Louis, MO, USA) solution was added to each well and further incubated for 4 h at 37° C [50]. Then, 100 μL of DMSO (Corning Inc., Corning, NY, USA) was added to each well to solubilize the formation product. Finally, a microplate reader (Bio-Tek Company, Winooski, VT, USA) was utilized to measure the absorbance at 490 nm, as previously described [48, 51].

Transwell invasion assays

The cell invasion assay was conducted using Transwell Permeable Supports (Corning Inc., Corning, NY, USA), as previously described [48, 52]. Cells (1.0 × 10⁵) suspended in serum-free medium were plated in triplicate onto 8-μm Transwell filter inserts of 24-well plates pre-coated with Matrigel (10 mg/L) (BD Biosciences, San Jose, CA, USA). And 10% fetal bovine serum (FBS) was added into the lower chamber. After 18 h, the cells in the upper chamber were wiped with a cotton-tipped swab. Then, the chambers were fixed with 100% methanol and stained with 0.5% crystal violet for 20 min. Images of invaded cells were obtained through an inverted light microscope (Olympus Corp., Tokyo, Japan). Nine fields were observed at random in each group.

Western blot analysis

Radio immunoprecipitation assay lysis buffer containing proteinase inhibitors (Beyotime, Jiangsu, China) was utilized to extract protein from cultured cells. Protein concentrations were determined using the BCA Protein Assay Kit (Pierce, Rockford, IL, USA). Protein samples (30 μg) from each group were resolved on 10% SDS-PAGE and transferred onto nitrocellulose membranes [51]. Then, the membranes were then incubated with anti-B3GNT3 (NBP1-32539; Novus Biologicals), E-Cadherin (ab40772; Abcam), N-cadherin (ab76011; Abcam), vimentin (ab92547, Abcam), and anti-GAPHD (ab181602; Abcam). After washing, the membranes were subsequently incubated with secondary horseradish peroxidase-conjugated antibody (1:5000; Santa Cruz Biotechnology) for 1 h at room temperature. Protein-expressing signals were visualized using the Western Lightning Plus ECL kit (PerkinElmer, Waltham, MA, USA) and quantified by densitometry, as previously described [50].

Immune infiltration analysis in PC through multiple datasets

As previously introduced [53], we were using the “estimate” package in R to perform the ESTIMATE algorithm in the TCGA PC dataset, which was based on ssGSEA to generate a tumor purity score and an immune score [18]. PC samples with high immune scores showed high level of immune cells infiltration within the TME. A higher tumor purity score indicates a lower immune infiltration in tumor tissues. The CIBERSORT algorithm is an analytic tool that estimates specific cell type frequencies using gene expression data. Thus, we used the CIBERSORT algorithm to estimate the fraction of 22 immune cell types in each tumor tissue in the TCGA PC dataset [17]. Next, using the “gsva” package in R, ssGSEA was conducted to calculate the enrichment scores of each immune-related term in the TCGA PC dataset [19]. The gene sets of the following 29 immune-related terms (CD8⁺ T cells, cytolytic activity, regulatory T cells (Treg), tumor-associated macrophages (TAMs), natural killer cells (NK cells), type-2 T helper cells (Th2 cells), type-1 T helper cells (Th1 cells), antigen-presenting cell (APC) co-stimulation, antigen-presenting cell (APC) co-inhibition, type-1 IFN response, major histocompatibility complex (MHC) class-1, parainflammation, plasmacytoid dendritic cells (pDCs), T cell co-stimulation, activated dendritic cells (aDCs), check-point, T cell co-inhibition, B cells, follicular helper T cells (Tfh), neutrophils, tumor-infiltrating lymphocyte (TIL), inflammation-promoting, type-1 IFN response, human leukocyte antigen (HLA), T helper cells, chemokine receptor (CCR), mast cells, dendritic cells, and immature dendritic cells (iDCs)) were obtained from a previous study [54]. Correlation analysis between B3GNT3 expression and immune infiltration was conducted using Pearson correlation coefficients (|Cor| > 0.3 and P < 0.05) [55]. The GSE62452 and GSE79668 datasets were used to validate the immune infiltration landscape of PC using ESTIMATE and ssGSEA. Furthermore, we conducted correlation analysis between B3GNT3 and specific immune cells markers to further confirm the association between B3GNT3 and immune infiltration in PC.

Tissue microarray (TMA) construction

A total of 50 samples of PC patients were collected after R0 resection in Guangdong Provincial People’s Hospital during 2013-2020. All PC patients were regularly followed up in our institution. The OS of these 50 patients was >1 month. Thereafter, a TMA was constructed for the 50 samples using 1.5 mm tissue cores. However, the final number of samples analyzed was lower due to the absence or limited tumor cells in some samples or unavoidable loss of tissue, as commonly occurs in TMA studies [56]. All tissues were approved by the Clinical Research Ethics Committee of Guangdong Provincial People’s Hospital. Informed consent was obtained from all participants.

Multi-color immunofluorescence

We measured the expression level of B3GNT3 using multi-color immunofluorescence in the TMA slides. B3GNT3 (NBP1-32539; Novus Biologicals) and CD8A (ab17147; Abcam) were simultaneously stained in serial sections from the TMA blocks. In short, antigen retrieval was performed with citrate buffer pH 8.0 for 20 min at 97° C in a pressure-boiling container. Blocking was subsequently conducted using 0.3% bovine serum albumin in 0.05% Tween solution for 30 min [56], followed by incubation with primary antibodies at 4° C overnight. At last, fluorescein-labeled secondary antibodies were added for signaling detection at room temperature. Nuclei were detected with DAPI. The samples were photographed using a laser confocal microscope. The area quantification the FL method of Halo software v3.0.311.314 (Indica Labs, USA) was used to quantify the intensities of the B3GNT3 fluorescent signals. Then, we divided the samples into high- and low-B3GNT3 expression groups according to the medium cutoff of the B3GNT3 fluorescent signal intensity. On each slide, three stromal areas of 4 × 10⁴ μm² were randomly selected based on DAPI staining. Within each area, CD8⁺ T cells were counted manually, and the average CD8⁺ T cell counts within the three stromal areas (4 × 10⁴ μm²) were calculated for each slide. Lastly, we evaluated the differences in CD8⁺ T cell count (per 4 × 10⁴ μm²) between the high- and low-B3GNT3 expression groups.

Statistical analysis

All statistical analyses were performed using SPSS 25.0 software, R 3.5.2 software (http://r-project.org/), and GraphPad Prism 8.0 software (GraphPad Software, Inc.). Group differences were analyzed by Wilcoxon test or Kruskal-wallis test, and expressed as the mean ± standard deviation (SD). Correlation analysis was performed with the Pearson correlation coefficient. P < 0.05 was considered statistically significant.