Introduction

Esophageal cancer is a kind of malignant tumor that seriously threatens human health, and its incidence and morbidity in China rank first in the world [1]. Esophageal squamous cell cancer (ESCC) accounts for 90% of all esophageal cancer cases in China.

In Japan, the infiltrative growth pattern (INF) has been routinely assessed as a pathologic characteristic of surgically resected specimens. Although the INF can be easily determined by hematoxylin and eosin (HE)-stained slices and used without specialized training, it has not gained widespread use in the clinic, and there are few reports on the predictive value of the INF regarding the outcome of ESCC [2,3,4,5,6].

Recently, a revolutionary concept shifted the focus from the tumor towards the microenvironment to determine the clinical disease course. Tumor-infiltrating lymphocytes (TILs), as a main component of the immune microenvironment, accumulate in many solid tumors, and their role in tumor progression remains controversial. Several reports have indicated that tumors with abundant TILs are associated with a better prognosis than tumors with scarce TILs in various cancers [7,8,9,10]. Tertiary lymphoid structures (TLSs), as the location where the generation of an efficient adaptive immune response against cancer occurs, reflect lymphoid neogenesis occurring in tumor peripheral tissues. Their presence is associated with a favorable prognosis in most solid malignant tumors [11].

In this study, we analyzed a series of 593 prospectively recruited stage T1 ESCC patients in China. The aim was to characterize the association between the INF type and the tumor immune microenvironment (TILs and TLSs), as well as to determine the predictive value of the INF type for lymph node metastasis and prognosis. In addition, we generated a novel grouping system to process the risk stratification of prognosis in T1 stage ESCC.

Materials and methods

Patient selection

This study enrolled 593 patients with a diagnosis of stage T1 ESCC who underwent esophagectomy and regional lymphadenectomy in our hospital between 2009 and 2018. The clinical and pathological data were obtained through a detailed retrospective review of the medical records. None of the patients had received any therapy before surgery. The study protocol was approved by the ethics committee of Shanxi Cancer Hospital, China (Reference number 201990).

Evaluation of histopathological factors

All available HE-stained tumor slides were reviewed by two pathologists separately who were blinded to clinical data at the time of the histologic evaluation. Whenever there was uncertainty, a third experienced pathologist was consulted. The invasive depth of stage T1 ESCC was subclassified into 6 groups: (m1) intraepithelial carcinoma that did not break the basement membrane, (m2) lesions between m1 and m3, (m3) carcinoma that was extremely close to or infiltrating the lamina muscularis mucosa, and (sm1, sm2, and sm3) submucosal carcinomas equally dividing the depth of the submucosal layer.

Evaluation of the INF, TILs, and TLSs

According to the Japanese classification of esophageal cancer [12], the INF indicates the growth and infiltrative pattern of tumors and can be classified into one of the following three types with regard to the predominant pattern observed at the tumor margin: INFa (expansive type), expansive growth of tumor nests with a well-demarcated border from surrounding tissue; INFb (intermediate type), intermediate growth pattern between INFa and INFc; and INFc (infiltrative type), infiltrative growth of tumor nests with an ill-defined border from surrounding tissue.

TILs were assessed according to the criteria previously published by the International TILs Working Group 2014 for breast cancer [13]. In brief, TILs were defined as the mean percentage of stroma in the invasive carcinoma infiltrated by lymphocytes and plasma cells and are reported as the area occupied by mononuclear inflammatory cells over the total stromal area, not the number of stromal cells. TILs were scored as a continuous variable. In this study, we categorized patients into low-grade and high-grade groups based on the median value.

TLSs are ectopic lymph node-like structures characterized by lymphoid aggregation with high endothelial venules. Similar to secondary lymphoid organs, T and B lymphocytes in TLSs are present in separate areas and may contain a germinal center. We classified TLSs as present or absent according to TLS formation in the area adjacent to the tumor.

All these indexes were evaluated within the tumor invasive margin, which was defined as a region centered on the border separating the host tissue from tumor nests, with an extent of 1 mm.

Statistical analysis

The start of the follow-up period was defined as the date of surgery, and the end of the follow-up period was defined as either the date of death or the date of last censorship, whichever occurred first.

Interrelationships between the INF, TILs, TLSs, and clinicopathological characteristics were analyzed using the chi-square test. Univariate logistic regression and multivariate logistic regression analysis using a backward conditional method were utilized to identify the risk factors for and independent factors of lymph node metastases, respectively. To test the reproducibility of the evaluation method, kappa statistics were calculated for both intra- and interobserver agreement.

Overall survival (OS) was examined using the Kaplan-Meier survival analysis, and univariate Cox proportional hazards regression was used to calculate the hazard ratios and 95% confidence intervals. Variables found to be statistically significant (p < 0.05) on the univariate analysis were entered into the multivariate analysis.

For the logistic and Cox regression analyses, a two-tier system for the INF was used along with the presence or absence of INFc to facilitate risk stratification. For all analyses, a two-sided p value of < 0.05 was considered statistically significant. Analyses were conducted using the STATA software program (version 14.0; Stata Corp., College Station, TX).

Results

INF: association with TILs, TLSs, and clinicopathologic features

Of 593 patients, the INFa type was present in 179 patients, the INFb type was present in 218 patients, and the INFc type was present in 196 patients. The relationship between the INF type and TILs, TLSs, and clinicopathologic features of patients with T1 ESCC is shown in Table 1. INFc was associated with low-grade TILs (p < 0.001) and the absence of TLSs (p < 0.001), deep tumor invasion (p < 0.001), poorly differentiated phenotype (p < 0.001), and a strong possibility of lymphovascular invasion (p < 0.001) and lymph node metastasis (p < 0.001). All these statistically significant parameters gradually increased or decreased along the order of INFa-INFb-INFc.

Table 1 The relationship between INF and clinicopathologic features of patients with T1 ESCC
Full size table

INF: an independent risk factor for lymph node metastases

The univariate logistic regression analysis showed that the following variables were significantly associated with lymph node metastases: male sex, tumor size ≥ 20 mm, deep invasion, poorly differentiated phenotype, lymphovascular invasion, INFc, low-grade TILs, and the absence of TLSs. Multivariate backward stepwise logistic regression identified tumor size ≥ 20 mm (p = 0.022), deep invasion (p = 0.009 and p = 0.000), lymphovascular invasion (p = 0.001), and INFc (p = 0.045) as independent risk factors for lymph node metastasis (Table 2).

Table 2 Univariate and multivariate logistics analysis of risk factors for lymph node metastases
Full size table

INF and TILs: independent predictive factors for OS

The median follow-up period was 42 months (range, 1–112 months). During the follow-up, 61 patients died. The overall 1-, 3-, and 5-year survival rates were 97.0%, 92.1%, and 90.1%, respectively.

The univariate Cox analysis revealed that INFc, low-grade TILs, the absence of TLSs, tumor size ≥20 mm, deep invasion, poorly differentiated phenotype, lymphovascular invasion, and lymph node metastases were significantly associated with poor OS. Furthermore, INFc (p = 0.015) and low-grade TILs (p = 0.002) were also independent inferior predictive factors on the multivariate analysis (Table 3) (Fig. 1a–c).

Table 3 Univariate and multivariate Cox analysis of overall survival in stage T1 ESCC
Full size table
Fig. 1
figure 1

Kaplan-Meier survival curve of overall survival of patients with stage T1 ESCC stratified by the following histologic parameters. a INF type. b TILs. c TLSs. d INF and TILs

Full size image

INF and TILs: a novel histologic risk stratification model

Based on the independent prognostic factors INF and TILs, a novel histologic risk stratification model for OS was generated as follows: INFa/b and low-grade TILs (n = 107, 18%), INFa/b and high-grade TILs (n = 287, 48%), INFc and low-grade TILs (n = 89, 15%), and INFc and high-grade TILs (n = 110, 19%) (Fig. 2). The Kaplan-Meier curves showed that patients with INFa/b and high-grade TILs had the best prognosis, and patients with INFc and low-grade TILs had the worst prognosis. The other two groups demonstrated similar prognoses. The log-rank test showed that the difference between groups was significant (p < 0.001) (Fig. 1d).

Fig.2
figure 2

A novel histologic risk stratification model based on hematoxylin and eosin–stained slices. a Tumors with INFa/b and low-grade TILs. b Tumors with INFa/b and high-grade TILs. c Tumors with INFc and low-grade TILs. d Tumors with INFc and high-grade TILs

Full size image

Intra- and interobserver variation

To test the reproducibility of the evaluation method, two observers independently evaluated the INF, TILs, and TLSs on 40 randomly selected patients from this study. After 1 month, one observer performed another evaluation to assess the intraobserver variation. The kappa values were 0.73, 0.64, and 0.70 for the intraobserver assessment and 0.41, 0.50, and 0.55 for the interobserver assessment of the INF, TILs, and TLSs, respectively, which showed moderate to good agreement and validated the reproducibility of the evaluation method.

Discussion

The interdependent interaction between the tumor and its microenvironment is a crucial topic in cancer research, and this interaction orchestrates the fate of tumor progression. In this study, we focused on the tumor INF type and local immune microenvironment (TILs and TLSs) in stage T1 ESCC. In some studies, according to their location, TILs and TLSs are classified as intratumoral and stromal TILs and intratumoral and peritumoral TLSs. In this study, we chose the invasive margin to evaluate these parameters. Because the invasive margin is reported to constitute a critical interface between pro- and antitumor factors, tumor-immune cell interactions at the invasive margin are a biologic driver of tumor aggressiveness [14].

We demonstrated that INFc was more closely associated with the immunosuppressive microenvironment (low-grade TILs and the absence of TLSs) and tumor aggressiveness (deep invasion, poor differentiation) than INFa and INFb. Some reports have shown that INFc is significantly associated with a poorly differentiated phenotype and deep invasion in gastric and pharyngeal cancers [3, 6], compatible with our results. However, there is very little published literature describing the association between the INF type and the tumor immune microenvironment. A previous report indicated that some small tumor nests at the invasive front were inversely associated with CD8+ TILs [15]. These small tumor nests were described as cancer-initiating cells or tumor budding (a cluster of fewer than five cancer cells) in several tumor types [16, 17]. Ito et al. found that tumor budding was strongly correlated with INFc, and the concordance rate was approximately 80% [2]. In our opinion, the evaluation of the INF is based on the invasive form of tumor nests in whole tissue at low magnification, while tumor budding is based on the cellular level at high magnification. In addition, “small tumor nests” and “tumor budding” were included in the concept of an infiltrative-type growth pattern (INFc), which refers to tumor nests at the tumor margin with an ill-defined border from surrounding tissue. van Wyk et al. described an inverse association between tumor budding and the intensity of peritumoral inflammation in colorectal cancer, which is in accordance with our results regarding the relationship between the INF and immune reaction [18]. Viktor et al. revealed that these small tumor nests frequently lost major histocompatibility complex-I expression when its expression was maintained in differentiated tumor cells in colorectal cancer; consequently, these small tumor nests may be able to reduce immunogenicity and evade the antitumoral host immune response and then undergo epithelial mesenchymal transition, invasion, and metastasis [19]. This explains why patients with the INFc type in our study always exhibited deep invasion and strong lymph node metastasis ability. We also found that the shallower the infiltration was, the richer the TILs and TLSs, and this result was also confirmed in a previous study on colorectal cancer [20]. At an early disease stage, an active immune reaction could be a major determinant for controlling tumor evolution, as with tumor progression, immune cells gradually decreased. Therefore, we focused on stage T1 ESCC to explore the immune microenvironment in our study. Further work is required to elucidate the molecular basis of the interaction between the INF and the immune microenvironment, which can help us design more efficient immunotherapeutic strategies.

According to the univariate logistic analysis, low-grade TILs and the absence of TLSs were associated with lymph node metastasis. We believe that a weak immune reaction reflects a defect of the host response to tumor challenge, leading to an increased likelihood of lymph node metastases. In the multivariate analysis, INFc was identified as an independent risk factor for lymph node metastasis, which was also validated in early pharyngeal cancers. The evaluation of the INF type needs to be performed in the tumor margin, which is present only in surgical specimens; it cannot be evaluated using biopsy specimens, and therefore cannot be used to select the preoperative strategy. However, in T1 ESCC, for which the treatment is complicated, the INF type can be applied in the histopathological evaluation of endoscopically resected specimens to assess the likelihood of metastasis and whether any additional treatment is necessary. This is another reason why we treated T1 ESCC as the subject in our study. Kanda et al. reported that the INFc type was associated with a high frequency of peritoneal recurrence, and the INFa or INFb type was associated with a high frequency of hepatic recurrence in gastric cancer [3]. INFc was also associated with recurrence in bladder cancer and colorectal cancer [4, 5]. Thus, we assumed that it may be helpful for the selection of more aggressive postoperative treatment.

The prognostic value of the INF and TILs has been demonstrated in several studies [2,3,4,5, 7,8,9]. In this study, we evaluated for the first time the combination of the INF and TILs as a histopathological prognostic factor in stage T1 ESCC. We demonstrated that INFc patients with high-grade TILs had a better prognosis than patients with low-grade TILs; the same was also true for INFa or INFb patients. Using this novel risk stratification model, we can assign patients based on the findings of a microscopic evaluation with HE staining alone, making it a useful and practical system that can be applied in any pathology laboratory and that requires no special ancillary testing. This prognostic impact was independent of the tumor–node–metastasis (TNM) stage, so it may be able to supplement the current TNM staging system.

This study has several limitations. First, it is a retrospective observational study that enrolled stage T1 ESCC patients who underwent esophagectomy and regional lymphadenectomy. As endoscopic resections were increasingly performed in tumors confined to the mucosa, esophagectomy was only performed in patients with poorly differentiated, larger lesions, or suspected LN metastasis by examination. Thus, a selection bias may exist due to surgical indications and patients’ decisions. Second, the number of events expected in OS was low, and the analysis may be under-powered. Therefore, more studies and a larger sample size will be necessary in future research. Third, other immune cells, such as neutrophilic and eosinophilic granulocytes and macrophages, etc., were not addressed in this study; however, they may be involved in immune microenvironment.

In conclusion, INFc is an independent risk factor for lymph node metastasis and an independent yet inferior prognostic factor of OS in T1 stage ESCC. Furthermore, it is associated with immunosuppressive low-grade TILs and the absence of TLSs, and the combination of the INF and TILs is useful for the risk stratification of prognosis in stage T1 ESCC.