Forty-nine patients with stage III primary LUAD who underwent surgical resection at Tianjin Medical University Cancer Institute and Hospital between January 2015 and May 2016 were enrolled in this retrospective study. Pathological TNM staging was histologically diagnosed based on the 7th edition of the Union for International Cancer Control TNM classification. The inclusion criterion was complete clinical data, standardized postoperative treatment and accurate pathological diagnosis. All patients underwent surgical resection of R0, and adjuvant therapy was mainly platinum-based chemotherapy, supplemented by radiotherapy or targeted therapy, when necessary. The exclusion criteria were those who had received anti-cancer treatment before surgery, had a second primary tumor, or were lacking follow-up. In this study, TLS positive tissues were selected for subsequent experiments which confirmed through hematoxylin and eosin (HE) staining slices. Formalin-fixed paraffin-embedded tumor tissues were collected from 49 patients for subsequent immunohistochemical staining and multiple immunofluorescence staining (Table 1). The study was approved by the Ethics Committee of the Tianjin Medical University Cancer Institute and Hospital. All patients signed relevant informed consent forms.

Multiple immunofluorescence staining was performed using a PerkinElmer Opal 7-color Technology Kit (NEL81001KT). The tumor specimens in paraffin-embedded blocks were cut into 4-μm-thick sections. The sections were deparaffinized in xylene and rehydrated in ethanol. Microwave repair was performed using EDTA buffer (PH=9.0) for 20 min. After cooling, the tissue was sealed with an antibody blocker at room temperature. The sections were then incubated overnight with primary antibody in a refrigerator at 4°C, and on the second day, the sections were co-incubated with poly-HRP-MS/Rb for 10 min at room temperature. Visualization was performed using Opal TSA (1:100). EDTA buffer was then heated by MWT to remove the AB-TSA complex. These steps were repeated for each round of the multiple staining. TSA-stained sections were washed with MWT and counterstained with DAPI (1:100) for 10 min. Using this staining method, all samples were stained with the primary antibody for CD20 (1:600 dilution, clone L26, Abcam) visualized with Opal520 TSA, CD3 (1:400 dilution, clone SP162, Abcam) visualized with Opal540 TSA, CD103 (1:500 dilution, clone EPR4166(2), Abcam) visualized with Opal570 TSA, Bcl6 (1:200 dilution, clone LN22, Novus) visualized with Opal620 TSA, CD4 (1:1000 dilution, clone EPR6855, Abcam) visualized with Opal650 TSA, CD21(1:800 dilution, clone EP3093, Abcam) visualized with Opal690 TSA. Finally, the sections were covered with an anti-fluorescence attenuating tablet and cover glass.

Tumor sections were scanned using a PerkinElmer Mantra Quantitative Pathology Imaging System at 200× magnification. Multispectral images were obtained using PerkinElmer inform Image Analysis software (version 2.4.0). Spectral libraries were built from the images of single-stained tissues with each antibody. The TLSs were then manually distinguished. We collected all TLSs of every tumor section and randomly collected three to five fields from areas outside the TLSs. A total of 958 fields were collected, including 807 TLSs and 151 outside fields of TLS.

The density of TLS was calculated as the number of TLSs per mm² of the tumor region in the sections. Immune subsets were determined by antibody expression, including CD4⁺ T cells, CD8⁺ T lymphocytes (CD3⁺CD4^-), B cells (CD20⁺), FDC (CD21⁺), CD3⁺ T_RM (CD3⁺CD103⁺), CD4⁺ T_RM (CD4⁺CD103⁺), CD8⁺ T_RM (CD3⁺CD4^-CD103⁺), and GC reaction (CD20⁺Bcl-6⁺) (25). The proportion of the immune subsets in each TLS (or field) was calculated as the percentage of this subpopulation to all nucleated cells in the TLS (or field). The proportion of the immune subsets in each patient was calculated by the average proportion in all fields (within the TLS and outside the TLS) across the entire section.

Disease-free survival (DFS) was defined as the time from the date of surgery to tumor recurrence. The surv_cutpoint function in the survminer R package (version 4.1.2) was used to obtain the cutoff value of immune subsets proportion. Then the different immune subsets inside and outside TLS were divided into “high” and “low” groups. Kaplan-Meier curve was drawn with the survminer R package (4.1.2). The log-rank test in survival R package (4.1.2) was used to calculate the P value. Both the survminer and survival R package were downloaded from the public resource website: https://cran.r-project.org/. When comparing the prognostic differences of more than two of sub-groups after combining TLS score and T_RM, P value and HR ratio was calculated with log-rank test in GraphPad Prism software.

Chi-square (and Fisher’s exact) test was used to evaluate the relationship between grade score, CD3⁺ CD103⁺ TRM and clinicopathological features. Wilcoxon rank test (paired nonparametric t test) was used to compare the difference of CD103⁺subsets inside and outside TLS. Kruskal-Wallis H test was used to compare the differences of immune subsets proportion among different sub-group. All statistical analyses, except survival analyses, were performed with GraphPad Prism (version 9.1.0, US). P values of < 0.05 were considered statistically significant.

According to the increasing prevalence of FDCs and the maturation of B cells, TLSs were classified into three maturity stages: 1) early TLS (E-TLS), characterized by dense lymphocytic aggregates without CD21 and Bcl-6 expression (Figure 1A); 2) primary follicle-like TLS (PFL-TLS), characterized by lymphocytic clusters with central network CD21 expression, but no GC reaction (Bcl-6^-) (Figure 1B); and 3) secondary follicle-like TLS (SFL-TLS), characterized by lymphocytic clusters with GC reaction (CD20⁺Bcl-6⁺) (Figure 1C).

For the first time, we divided patients into three levels based on the maturity of TLSs: 1) grade 1: patients with TLSs characterized by only E-TLSs, and without PFL-TLSs and SFL-TLSs; 2) grade 2: patients with TLSs characterized by E-TLSs and at least one PFL-TLS, but no SFL-TLS; and 3) grade 3: patients with TLSs characterized by at least one SFL-TLS in the tumor tissue (Table 2).

We first evaluated the prognostic impact of the number and density of TLSs in patients. Kaplan–Meier analysis showed that patients with higher numbers of TLSs had a much better prognosis (median DFS 18.7 months vs. 7.4 months, P = 0.011, Figure 2A, left). A higher density of TLS was also positively associated with a good DFS (median 17.3 months vs. 12.4 months, P = 0.009, Figure 2A, right).

Furthermore, we analyzed the prognosis of patients with different grades. The results showed that the prognosis of grade 3 patients was significantly higher than that of those in grade 1 (median DFS 19.5 months vs. 4.3 months, P <0.001, Figure 2B). The grade 2 patients also had a better DFS than those in grade 1 (median 12.6 months vs. 4.3 months, P =0.039). The prognosis of grade 3 patients tended to be better than those in grade 2 (median DFS 19.5 months vs. 12.6 months, P =0.059, Figure 2B). These results indicate that the maturity of TLS is crucial for the prognosis of patients.

By comparing the difference in the proportion of T_RM inside and outside the TLS, we determined that all T_RM subsets were mainly located in TLS, especially CD4⁺ T_RM (Figures 3A, B). The proportion of CD3⁺ T_RM in TLS (mean ± SD: 1.34% ± 1.13%) was significantly higher than that outside (mean ± SD: 0.71% ± 0.84%), P <0.001. The proportion of CD4⁺ T_RM in TLS (mean ± SD: 0.83% ± 0.75%) was significantly higher than that outside (mean ± SD: 0.26% ± 0.33%), P <0.001. The proportion of CD8⁺ T_RM in TLS (mean ± SD: 0.78% ± 0.72%) was significantly higher than that outside (mean ± SD: 0.55% ± 0.67%), P <0.05.

Survival analysis showed that both CD3⁺ T_RM and CD8⁺ T_RM in TLS could predict longer survival (median DFS 17.3 months vs. 6.7 months, 17.5 months vs. 12.7 months, P<0.05; Figure 3C). Likewise, CD4⁺ T_RM within TLS tended to prolong DFS of patients although there was no significant difference between groups (median 15.2 months vs. 6.9 months, P =0.078, Figure 3C). However, TRM outside the TLS had no effect on prognosis.

The above findings suggested that T_RM is mainly located within the TLS, and the T_RM within the TLS can affect prognosis. Therefore, we next focused our research on T_RM inside TLS. The relationships between patient score, CD3⁺ T_RM within TLS, and clinical features of patients including sex, age, TNM stage, micropapillary, EGFR mutation, and smoking are shown in Table 3. The results showed that patients with stage IIIA LUAD had more mature TLS than patients with IIIB LUAD (P =0.041). There were no significant associations between other clinical features and patient scores, as well as between CD3⁺ T_RM and TLS.

Moreover, we analyzed the clinical and immune characteristics that affected the DFS of patients in this study. The results identified some univariate factors that could affect the DFS of patients, including the TLS number, TLS density, patient score, CD20⁺ B cells in TLS, FDC in TLS, CD103⁺ T_RM, and CD8⁺CD103⁺ T_RM in TLS, average CD3⁺ CD103⁺ T_RM, average CD4⁺CD103⁺ T_RM, and average CD8⁺CD103⁺ T_RM (Table 4).

We analyzed the distribution of immune subsets in the TLS in different grades of patients. The results showed that the proportion of CD20⁺ B cells in grade 3 patients was higher than that in grade 2 and grade 1 patients (mean ± SD: 21.08% ± 6.72% vs. 15.59% ± 4.47%, 21.08% ± 6.72% vs. 11.41% ± 7.84%, P =0.037, and P =0.008, respectively; Figure 4A). The proportion of CD3⁺ T_RM in grade 3 patients was higher than that in grade 2 and grade 1 patients (mean ± SD: 1.98% ± 1.23% vs. 1.10% ± 0.96%, 1.98% ± 1.23% vs. 0.57% ± 0.38%, P =0.027, and P =0.019, respectively; Figure 4A). Patients in grade 3 had higher CD4⁺ T_RM than patients in grade 1 (mean ± SD: 1.33% ± 0.87% vs. 0.59% ± 0.53%, P =0.005; Figure 4A). The proportion of CD8⁺ T_RM in grade 3 patients was higher than that in grade 1 (mean ± SD: 1.06% ± 0.84% vs. 0.69% ± 0.63%, P =0.039, Figure 4A).

Next, we analyzed the proportion of T_RM in all 807 TLSs, including E-TLSs, PFL-TLSs, and SFL-TLSs. The proportion of CD3⁺ T_RM in SFL-TLS was significantly higher than that of E-TLS and PFL-TLS (mean ± SD: 3.60% ± 6.80% vs. 1.57% ± 2.71%, 3.60% ± 6.80% vs. 1.23% ± 1.62%, P<0.001; Figure 4B), respectively. The proportion of CD4⁺ T_RM in SFL-TLS was significantly higher than that in E-TLS and PFL-TLS (mean ± SD: 2.17% ± 2.19% vs. 0.98% ± 2.21%, 2.17% ± 2.19% vs. 0.67% ± 1.19%, P<0.001; Figure 4B), respectively. The proportion of CD8⁺ T_RM in SFL-TLS was significantly higher than that in E-TLS and PFL-TLS (mean ± SD: 1.30% ± 1.78% vs. 0.98% ± 2.46%, 1.30% ± 1.78% vs. 0.81% ± 1.41%, P<0.001; Figure 4B), respectively.

Patients were stratified into four groups according to the proportion of CD103⁺ T_RM and patient scores. The prognosis of patients in the group of CD3CD103^High and grade 3 was significantly higher than that of CD3CD103^High and grade 1 + 2 (median DFS 19.7 months vs. 12.7 months, P =0.046) and that of CD3CD103^Low and grade 1 + 2 (median DFS 19.7 months vs. 7.2 months, P =0.003), respectively (Figure 5A). Patients in the CD4CD103^High and grade 3 groups had a significantly better prognosis than those in CD4CD103^High and grade 1 + 2 (median DFS 19.5 months vs. 12.6 months, P =0.037) and in CD4CD103^Low and grade 1 + 2 (median DFS 19.5 months vs. 6.9 months, P =0.011), respectively (Figure 5B). Similarly, the prognosis of patients in the group of CD8CD103^High and grade 3 tended to be better than that of CD8CD103^High and grade 1 + 2 (median DFS 19.7 months vs. 12.8 months, P =0.052), and significantly higher than that of CD8CD103^Low and grade 1 + 2 (median DFS 19.7 months vs. 7.3 months, P <0.001; Figure 5C). However, there was no significant difference in prognosis between patients in the CD103^High and grade 1 + 2 groups and the CD103^Low and grade 1 + 2 groups, regardless of the CD3⁺ T_RM, CD4⁺ T_RM, or CD8⁺ T_RM subsets.