Clinical relevance of tertiary lymphoid structures in esophageal squamous cell carcinoma

Tumor samples were obtained from 293 patients (mean age, 66 years) with primary ESCC who underwent surgical resection at the Department of Gastroenterological Surgery, Osaka City University Hospital, between 2011 and 2014. In this study, the following cases were excluded: 7 patients undergone endoscopic submucosal dissection, 9 cases of non-radical resection, 31 patients of preoperative radiotherapy, 4 patients who died during the perioperative period, 4 patients with intramural metastasis in the stomach, and 2 patients with pathological complete response by NAC. A total of 236 patients (mean age, 67 years) who had undergone esophagectomy with reconstruction for ESCC were enrolled in this study. The median follow-up time was 51 months (range 0–100 months). Indications for preoperative chemotherapy and details of the regimens were based on the esophageal cancer treatment guidelines. The preoperative regimen was 5-FU + cisplatin (FP), 5-FU + nedaplatin (FGP), or 5-FU + cisplatin + docetaxel (DCF). The chemotherapy dose was reduced as needed according to patient condition and adverse events. None of the patients received immunotherapy prior to surgery. Tumors were diagnosed histologically based on the 11^th Edition of the Japanese Classification of Esophageal Cancer [10, 11]. Upper gastrointestinal endoscopy, CT scan and fluorodeoxyglucose-position emission tomography imaging were performed in all patients for clinical diagnosis. The diagnosis was based on the consensus of all esophageal surgeons in our hospital. Informed consents were obtained from all patients. The differentiation of squamous cell carcinoma was classified as follows: well-differentiated type (n = 20), moderately differentiated type (n = 118), poorly differentiated type (n = 87), and others (n = 11). Patients received adjuvant chemotherapy in case with pathologically diagnosed lymph node metastasis.

Formalin-fixed paraffin-embedded (FFPE) tumor blocks with representative tumor areas were collected from 236 consecutive patients with ESCC. Immunohistochemical staining was performed according to previous reports from our department [4]. In briefly, immunohistochemical staining was performed on 4-μm-thick sections of paraffin-embedded tumor blocks from ESCC patients. After incubation at 60 ˚C for 10 min, the sections were deparaffinized in xylene and rehydrated with a graded ethanol series (70%, 80%, 90% and 100%). Endogenous peroxidase activity was blocked using absolute methanol containing 3% hydrogen peroxidase for 15 min. After washing the sections in PBS, antigens were retrieved by microwave treatment. Nonspecific binding was subsequently blocked using nonspecific staining blocking reagent (DAKO, Kyoto, Japan). The sections were then reacted with the primary antibody listed below, standing at 4 ˚C overnight. All reactions were performed using appropriate positive and negative controls. Antibodies used for immunohistochemical analyses were as follows: a mouse monoclonal anti-CD3 antibody (clone: F7.2.38, 1:50, high pH retrieval; Abcam, Cambridge, UK), a mouse monoclonal anti-CD8 antibody (clone: F7.2.38, 1:100, high pH retrieval; Abcam, Cambridge, UK), a mouse monoclonal anti-CD20 antibody (clone: IR604, prediluted, low pH, Merck, Darmstadt, Germany), a mouse monoclonal anti-CD21 antibody (clone: 2G9, prediluted, low pH, Nichirei, Tokyo, Japan), a rabbit monoclonal anti-CD23 antibody (clone: SP23, prediluted, low pH, Nichirei, Tokyo, Japan), and a rat monoclonal anti-PNAd (clone: 120,801, 1:25, high pH, Nichirei, Tokyo, Japan). Sections were incubated with secondary antibody histofine reagent (Nichirei, Tokyo, Japan), and the signal was visualized using 3–3’-diaminobenzidine (DAB). Sections were counterstained with hematoxylin before mounting.

Tumor slides stained with anti-CD20 and anti-CD23 antibodies were scanned at low magnification to select three fields (9.1 mm²; original magnification 20 × ; 1600 × 1200 resolution) with the greatest number of intratumoral and peritumoral CD20⁺ or CD23⁺ cells. The microscopic images were imported from the digital photo filing system DP-73 (Olympus, Tokyo, Japan). The area (mm²) of CD20⁺ cells was measured, and the percentage area (%) of each field that was covered by the CD20⁺ or CD23⁺ area was calculated with Image J software (NIH, Bethesda, MD). For the CD20⁺ and CD23⁺ areas, clusters occupying ≥ 0.04% and ≥ 0.01%, respectively, were counted as positive. CD20 and CD23 densities were determined as the mean values of the percentage areas in the three fields. Tumor slides stained with anti-CD8 antibody were scanned at high magnification, and five fields (original magnification 400 ×) were selected with staining of intratumoral and peritumoral CD8⁺ cells. The mean numbers of positively immunostained cells per field were counted using Image J software. To determine the cutoff value, each histogram was generated, and the quartile was calculated. A histogram of the obtained numbers was created to set the cutoff values for CD20, CD23, and CD8, respectively. The cutoff value was set to 0.863 for the first quartile for CD20, 0.128 for the second quartile for CD23, and 64 for the first quartile for CD8.

The association with relapse-free survival (RFS) was analyzed initially by a Kaplan–Meier curve and the log-rank test. RFS curves were drawn using the Kaplan–Meier method, and the log-rank test was used to assess the significance of differences in survival. RFS was defined as the time between the day of surgery and recurrence. P-values < 0.05 were considered significant. Each statistical analysis was performed using the JMP software program (SAS Institute, Cary, NC, USA).

The “T.T” ESCC cell lines were used in this study. T.T human esophageal squamous cell carcinoma cells were obtained from the Health Science Research Resources Bank (Osaka, Japan). The cells were cultured at 37˚C and in 5% CO2. The media used were DMEM (FUJIFILM Wako Pure Chemical Corporation, Tokyo, Japan) and RPMI-1640 (FUJIFILM Wako Pure Chemical Corporation, Tokyo, Japan), supplemented with 10% fetal bovine serum (FBS; Gibco, NY, USA), 100 IU/ml penicillin (ICN Biomedicals, CA, USA), 100 mg/ml streptomycin (ICN Biomedicals, CA, USA), and 0.5 mM sodium pyruvate (Bioproducts).

T.T cells (1 × 10⁶/ml) were plated in 10 ml of full medium treated with or without treatment with 5-FU and CDDP (30 µM) for 24 h. PBMCs were isolated from total 10 healthy donors using Ficoll density gradient centrifugation with Ficoll-Paque™ PLUS (Cytiva, Tokyo, Japan) at 1,025 × g for 30 min at 20 ˚C, with the brakes off. The PBMCs (1 × 10⁷/ml) were then co-cultured with naïve or chemical-treated T.T cells (1 × 10⁶/ml) for 24 h (n = 7 and 3, respectively). The ratio of CD20⁺ B cells, CD23⁺ B cells, and CD8⁺ T cells in the PBMCs after co-culture with T.T cells was examined by flow cytometry. In brief, after saturation with BD Fetal Bovine Serum Stain Buffer (BD Biosciences, NJ, USA), mononuclear cells were incubated with the primary antibodies and Fc-receptor blocking buffer (cat. no. 564220, 2% HS in PBS) for 30 min at 4 ˚C in the dark. The following monoclonal directly labeled anti-human antibodies were used: PE-labeled anti-CD20 (cat. no. 590961), BV421-labeled anti-CD23 (cat. no. 562707) and FITC- labeled anti-CD8 (cat. no. 555634). The evaluation of dead cells was performed with 7-Amino-Actinomycin D (cat. no. 559925). All antibodies used for flow cytometry were purchased from BD Biosciences. Flow cytometric analyses were performed with the BD LSRFortessa™ X-20 (BD Biosciences).

Chi-square test and Fisher’s exact test were used to assess the associations of the expressions of CD20⁺ B cells, CD23⁺ B cells, and CD8⁺ T cells with clinicopathological features. The degree of infiltration between the density of CD20⁺/CD23⁺ B cells and the number of CD8⁺ T cells was compared by Spearman’s correlation coefficient. The regression line was determined by plotting the number of each cell, and the correlation was examined. A Cox proportional hazard model was used for univariate and multivariate analyses of prognostic factors. P-values < 0.05 were considered significant. Correlation strength was defined as weak (0.2 < r ≤ 0.4) and strong (r > 0.7). The Wilcoxon rank sum test was used to compare the expressions of CD20⁺ B cells, CD23⁺ B cells, and CD8⁺ T cells between the two groups.

A representative stained slide is shown in Fig. 1. Peritumoral regions of clinical samples that were stained with HE were examined microscopically, and lymphocytes aggregating in the peritumoral regions were confirmed (Fig. 1A). Anti-CD20 staining showed positive cells at the same site (Fig. 1B). Regardless of whether CD20 expression was high or low, CD20⁺ B cells appeared mostly as aggregates that were located mainly at the invasive margin of the tumor and were present to line the tumor. The clusters of CD20⁺ B cells were surrounded by a CD3⁺ T cell rich area, and CD21⁺ FDCs formed a tight network in the B cell zone of the follicle (Fig. 1C, D). On anti-CD8 staining, CD8⁺ cells were observed outside aggregating CD20⁺ cells and inside tumoral areas as tumor infiltrating lymphocytes (Fig. 1E). The structure of HEV within the CD20-TLS was identified (Fig. 1F). Therefore, TLS was defined as CD20⁺ B cell aggregation, and its association with clinical outcomes was evaluated in this study. In the loupe image, TLSs were present to line the vicinity of the tumor (Fig. 1H). To examine the impact of tumor-associated TLSs on clinicopathological features, the patients were divided into two groups based on the density of CD20⁺ cells: a TLS-rich group (n = 177) and a TLS-poor group (n = 69). The relationship between the patient’ clinicopathological factors and each lymphocyte type is shown in Table 1. Young patients tended to have a lower TLS density than elderly patients (p = 0.055). Regarding TLS density, there was no difference between the two groups due to other factors. Patients were also classified into two groups based on CD8: a CD8-high group (n = 178) and a CD8-low group (n = 58). Regarding CD8 infiltration, there were significant differences in the sex ratios, clinical tumor depth, clinical nodal metastasis, clinical stage, with or without NAC, and pathological tumor depth. Kaplan–Meier survival analysis showed that survival tended to be better in the TLS-rich group than in the TLS-poor group in patients who had not received adjuvant chemotherapy (p = 0.1344, log-rank test) (Fig. 2A). In contrast, there was no significant difference in survival in the patients who received NAC (Fig. 2B).

The presence of TLSs detected by CD20 alone showed no relationship to the prognosis of patients treated with NAC. B cells usually differentiate from naïve B cells in lymphoid tissues by germinal centers (GC) and are thought to transform into memory B cells. The CD23 expression of GCs in TLSs was examined to evaluate the function of TLS. CD23⁺ cells were aggregated within TLSs and were consistent with the light zone on HE staining (Fig. 1G). All of the aggregates of CD23⁺ cells were in the TLS formation and were not lacking in TLSs. It was previously reported that TLS maturation stage was evaluated, and secondary follicle-like TLS containing lymphocytic aggregates of CD23⁺ cells was indicated as mature TLS [12]. Following these previous studies, a colony of CD23⁺ cells within TLS was defined as a TLS containing germinal center (GC-TLS). The patients were also classified by CD23 into two groups, a GC-TLS-rich group (n = 59) and a GC-TLS-poor group (n = 177). It was found that GC-TLSs, as well as normal TLS formations, and the infiltration number of CD8⁺ T cells had a weak positive correlation in patients with or without NAC (r = 0.412, p < 0.0001, r = 0.320, p < 0.0001, respectively) (Fig. 3A,B). Subsequently, the correlation between GC formation in TLSs and CD8⁺ T cells with or without NAC was examined. Of the 236 patients, the number of patients who received NAC was 152 (64%). The number of infiltrating CD8⁺ T cells was significantly higher in patients with NAC than in those without NAC (p = 0.010) (Fig. 3C). Regarding GC formation, patients who underwent NAC tended to have a higher CD23 density (p = 0.055) (Table 2).

The prognosis tended to be good in the GC-TLS-rich group in the patients who received NAC, but there was no significant difference in both groups with and without NAC (p = 0.340, p = 0.094, log-rank test, respectively) (Fig. 4A, B). Regarding the prognosis of the NAC group, more patients had long-term survival in the GC-TLS-rich group than in the GC-TLS-poor group (Fig. 4C). To understand the prognostic value of GC-TLS, subgroup analysis was performed by clinical stage in patients who received NAC. While there was no significant difference in the 3-year survival rate in patients with clinical stage 0, I, or II, the 3-year survival rate of patients with clinical stage III or IVa was significantly better in the GC-TLS-rich group than in the GC-TLS-poor group (p = 0.022, log-rank test) (Fig. 4C,D). Hazard ratios were calculated by grouping patients according to clinical progression (III, IVa and 0, I, II). We found that GC-TLS had a significantly greater prognostic impact, in terms of hazard ratio, in comparison to TLS in patients with progression III, IVa (Fig. 4E). Furthermore, GC-TLS was an independent prognostic factor in the study of prognostic factors with the addition of CD8⁺TIL (Table 3).

Grade 3 or greater myelosuppression events were observed in 21 (13.8%) of 152 patients undergoing NAC. Postoperative adjuvant chemotherapy was given to 101 patients (43%). The relationship between myelosuppression that developed during NAC and immune cells was investigated. Myelosuppression involved three strains of cytopenia in this study. The patients with Grade 4 myelosuppression had significantly greater density of TLSs and GCs and infiltration of CD8⁺ T cells than other patients (p=0.016, p=0.004, p=0.049, respectively) (Fig. 5). In addition, to assess the association between chemotherapy and TLS maturation at the tumor site, the GCs of patients with and without chemotherapy in locally advanced cases were compared. GC formation was compared by classifying 96 patients with locally advanced cT3 or deeper. A responder was defined as a patient diagnosed as locally advanced cancer of T3 or deeper from preoperative examination and as T1 on postoperative pathology due to the effect of NAC. The responder group had better GC formation than the non-responder group (p=0.013) (Fig. 5D). Co-culturing of PBMCs with an anticancer drug-treated esophageal squamous cell carcinoma cell line increased the CD20 and CD23 expression in PBMCs in vitro (Fig. 6). GC-TLS was observed in abundance in the primary lesions of patients who underwent ICI treatment for postoperative recurrence of ESCC and who showed a clinical response (PR, CR) (Fig. 7).

This investigation was carried out according to the Declaration of Helsinki. All experimental procedures after 2013 were approved by the Osaka City University ethics committee (Approval No. 3138 and 4092), and all patients had provided informed consent for collection and analysis of the specimens. On the other hand, because the other experimental procedures up to 2013 were performed based on only all-inclusive consent, an opt-out form as for an observational study was used. Written informed consent was obtained from all patients.

Not applicable. This manuscript does not contain any individual person’s data. The clinical pathological features are also completely anonymized and there is no personally identifiable information.

The authors declare that they have no competing interests.

The datasets used and analyzed during this study are available from the corresponding author on reasonable request, and most of the original data are included within the article.