Introduction
Lymph node metastases (LNM) is a major prognostic factor and determines treatment for operable non-small cell lung cancer (NSCLC) patients [
1,
2]. LNM positive NSCLC patients with PD-L1 > 1% have been recommended to use anti-PD-(L)1 immunotherapy, but its efficacy is only 20–40% [
3]. Moreover, it is estimated that there are about 20% NSCLC patients missing the optimal window for immunotherapy due to ambiguous lymph node states [
4]. The main function of anti-PD-(L)1 immunotherapy is to activate CD8+ T cells, the central role in immune-mediated control of cancer [
5]. Consequently, it is imperative to enhance our understanding of CD8+ T cells in NSCLC patients, especially those with LNM, prompting precise immunotherapy.
The anti-tumor effect of CD8+ T cells largely depends on their spatial architecture and functional status. The spatial structure of tumor-infiltrating CD8+ T cells, such as the topologically distinct distribution and spatial interplay with other neighboring cells, determine the prognosis and treatment response of patients. Previous studies demonstrated that high density of CD8+ T cells in tumor center (TC) as well as invasive margin (IM) was associated with good prognosis of patients with NSCLC, while some studies believed that CD8+T cells in different regions embodied inconsistent prognostic significance, suggesting that there were some unexpected causes [
6‐
9]. Several studies reported that a higher density of CD8+ T cells neighboring CD8- T cells or regulatory T cells was associated with better prognosis of patients with NSCLC, indicating the potential significance of the CD8+ T cells limited in certain spatial scale [
10,
11]. Additionally, Peng et al. provided a novel insight to decode spatial information of immune cell, through variables based on the intercellular distance reflecting the level of interaction [
12]. Therefore, a comprehensive decipherment of the spatial structure is necessary for intratumoral CD8+ T cells in NSCLC.
The functional status of CD8+ T cells is modified by the expression of inhibitory molecules, which can be used to distinguish CD8+ T cells with different states [
13]. PD-1 has been used as an indicator of dysfunction of CD8+ T cells and PD-1+CD8+ T cells was thought as a biomarker to predict better response for immunotherapy and longer survival of patients, while some studies hold contradictory views [
14‐
18]. Using single-cell sequencing, Guo et al. identified a variety of CD8+ T-cell functional subsets via multiple transcriptional markers, such as PD-1, CD103 and Tim3, in NSCLC, and demonstrated that a high ratio of predysfunctional to dysfunctional T cells correlated with better prognosis in 11 patients with adenocarcinoma [
19]. However, these conclusions are limited by the lack of adequate sample size and histopathological studies reflecting the real infiltration.
The infiltration of CD8+ T cells are regulated by other stromal cells in the tumor microenvironment (TME). Some studies have demonstrated that endothelial cells of cancer microvessels (CMVs) can reduce the ability of CD8+ T cells to adhere to the vasculature and induce apoptosis of CD8+ T cells [
20,
21]. However, their effect on the functional status of intratumoral CD8+ T cells in tumor remains unclear. Grout et al. found that cancer-associated fibroblasts (CAFs) obstruct T cells from lung cancer cells [
22], but it is unclear whether these barriers are unselective for immune cells and are able to affect their functional states.
Despite the importance of understanding the functional status and spatial architecture of CD8+ T cells, traditional single-color immunofluorescence or immunohistochemistry methods are unable to distinguish complex cell subsets, whereas prevalent flow cytometry and single-cell transcriptome sequencing methods abandon important spatial information [
23,
24]. Multiplex immunofluorescence (mIF), enabling to stain cells by multi-protein labels in situ, provides an opportunity to analyze the spatial characteristics of CD8+ T cell functional subsets.
To elucidate the characteristics of intratumoral CD8+ T cells in NSCLC patients with LNM, using mIF and machine learning-assisted image analysis, we acquired the amount and single-cell localization of CD8+ T-cell functional subsets and other cells in the tumor microenvironment (TME) in 1116 tissue sites from 279 NSCLC patients. We noted that low density of predysfunctional CD8+ T cells (Tpredys) in TC, high density of dysfunctional CD8+ T cells (Tdys) in IM, and shorter distance from CD8+ T cells to neighboring cells were significantly associated with LNM and worse prognosis. CMVs and CAFs might act as “selective barriers” and were correlated with the dysfunction of CD8+T cells.
Discussion
In this study, utilizing mIF combined with machine learning-assisted image analysis, we deciphered the specific TME affecting the density and distribution of CD8+T cell functional subpopulations associated with LNM. Our results demonstrated that tumor-infiltrating CD8+ T cells were featured by decreased Tpredys cells and increased Tdys cells, driven by a special immunosuppressive microenvironment, which prompted us to consider whether it is appropriate to give anti-PD-(L)1 perioperative immunotherapy in all PD-L1>1% patients with LNM.
In current clinical practice for patients with operable NSCLC, anti-PD-(L)1 immunotherapy has been floundering due to limited efficacy to activate the anti-tumor effects [
30,
31]. In our observation, there are two plausible explanations for the limited efficacy of immunotherapy: First, we found a high proportion of CD8+ T
dys in NSCLC patients, which was even worse in patients with LNM; Second, some immunoregulatory barriers, such as CMVs and CAFs, were associated with poor infiltration and dysfunction of CD8+ T cells.
Dysfunctional CD8+ T cells used to be characterized using PD-1 alone in previous studies [
17]. Actually, the dysfunction of T cells is a successive process. Some immune inhibitory molecules, such as PD-1 and CD103, are mainly expressed on transitory dysfunctional CD8+ T cells, whereas Tim3 and LAG3 are mainly expressed on terminally dysfunctional CD8+ T cells [
13,
16,
19]. Therefore, we present a more exact landscape of T cells in NSCLC. Firstly, we found more CD8+ T cells instead of CD4+ T cells in IM than in TC. More importantly, there were more CD8+ T
dys in the IM, which may account for the inconsistent prognostic results of CD8+ T cells in previous investigations [
32‐
35]. Moreover, in the multivariate models, our results demonstrated that a low density of CD8+ Tpredys in TC and a high density of CD8+ Tdys in IM were significantly associated with LNM and poor RFS. A previous study found that NSCLC patients with LNM harboring more PD-1+ CD8+ cells, and more PD-1+ CD8+ cells existed in the IM [
17]. Using single-cell sequencing, Guo et al. demonstrated that a high ratio of pre-exhausted to exhausted T cells correlated with better prognosis in treatment-naive lung adenocarcinoma [
19]. More importantly, the function of dysfunctional CD8+ T cells were considered not able to be recovered by PD-(L)1 inhibitors [
36]. Notably, adaptive resistance to PD-1 inhibitors is associated with the upregulation of Tim3 [
37]. Collectively, the dysfunction of T cells represents a distinct status of T-cell differentiation with considerable clinical relevance. However, we still understand little about where, when and how to initiate or execute a dysfunction program.
Intercellular interactions play an important role in the regulation of immune cell function. We revealed a stronger CD8+ T-cell immunomodulatory network in NSCLC patients with LNM than in those without LNM. Specifically, shorter mNNDs between intratumoral CD8+ T cells and CD4+ T
con, CD4+ T
reg, CMVs and CAFs were observed in lymph node positive patients and were associated with worse RPS. A previous study showed that stronger CD8-T
reg proximity was associated with poor OS in NSCLC [
11], while another study found that a higher density of CD3+CD8+ cells neighboring CD3+CD8- cells was associated with better prognosis, despite the small sample size (n=20) [
10]. Furthermore, using single-cell transcriptome sequencing, Li et al. found that, in melanoma, both T
con cells and T
reg cells displayed levels of proliferation comparable to those observed in dysfunctional CD8+ T cells [
38]. Overall, although we provided novel insights, the interaction between CD4+ and CD8+ T-cell subsets remains ambiguous and requires more efforts.
The vasculature is pivotal for the transportation of immune cells to the target tissues. Our results showed that CMVs were correlated with increased proximity of all T cell subsets in the TC but were correlated with decreased proximity of CD8+ T
total in the IM, indicating that CMVs may be selective for T cells in the IM, reducing the infiltration of CD8+ T cells rather than CD4+ T cells. More importantly, we found that CMVs were associated with high proximity of CD8+ T
dys in the IM. Previous studies have demonstrated that cancer vasculature could impede T-cell trafficking through endothelial cell anergy via the downregulation of adhesion molecules [
20] and through the establishment of a death barrier via the upregulation of apoptotic ligands [
21]. Notably, CD4+ T cells in human primary lung tumors are Th2 skewed [
39], and Th2 cells express a higher level of anti-apoptotic signals [
40,
41]. Our findings strengthen the hypotheses of tumor vasculature as a “selective barrier” and indicate that there might be additional mechanisms for tumor vasculature to restrict T cells. However, further explorations are required to confirm our results. CAFs are another important stromal component of TME. A recent study identified two types of CAF subsets, MYH11+ αSMA+ CAF and FAP+ αSMA+ CAF, which contribute to T-cell exclusion and could restrict T cells contact with cancer cells [
22]. Likewise, our results demonstrated that αSMA+ CAFs were negatively correlated with cancer-proximal CD8+ T
total cells in both IM and TC, strengthening the evidence of the “CAF barrier” in lung cancer. Of note, we also observed that CAFs were positively correlated with CD4+ T
total cells and CD4+ T
con in IM, and CD8+ T
dys in TC, which indicated that CAFs might be more than a unselective mechanical barrier. Lakins et al. found that the upregulation of PD-1/PD-L2 and FAS/FASL in T cells/CAFs, respectively, drove the deletion and dysfunction of tumor-specific T cells [
42], which provided rationality to our findings.
Collectively, our study revealed that NSCLC patients with LNM were characterized by more dysfunctional intratumoral-infiltrating CD8+ T cells and a more immunosuppressive TME impeding CD8+ T cells. These findings indicated that anti-PD-(L)1 immunotherapy may not work as well in patients with immunosuppressive microenvironment until these adverse factors are eliminated. A combination strategy is a prospective orientation in facilitating immunotherapy. Radiotherapy, which can suppress CAFs and increase the infiltration of T cells, is therefore considered a wonderful partner for immunotherapy [
43]. Antiangiogenic agents at mild doses can induce the normalization of tumor vessels and some combination schemes are under clinical trials [
44]. More importantly, next-generation immunotherapies, such as CAR-T, in combination with other therapies could improve the infiltration of the modified immune cells or reverse their dysfunction, which may be the key to improving their efficacy [
45]. Given that the primary tumor may be an antigen source for activating and expanding tumor-specific T cells and systemic surveillance of micrometastases, neoadjuvant immunotherapy is promising [
46]. Unfortunately, some patients may miss the opportunity for neoadjuvant immunotherapy due to occult LNM (LNM is negative in preoperative evaluation but confirmed as positive by postoperative pathology) [
4]. Given a high risk of LNM and recurrence, future studies and clinical practice should give more concern over the patients with the above characteristics, requiring adequate clinical assessments and treatments to improve prognosis, such as combination therapy or close follow-up.
Some limitations of this study have to be acknowledged. First, excluding other pathological types, such as large cell lung cancer, from our cohort may restrain the utilization of our findings in a prevalent NSCLC cohort. Second, although rarely, it needs to be known that CD4 and CD8 are also slightly expressed on non-T cells, such as a few macrophages and dendritic cells [
47,
48]. It is uncertain whether the setting of a positive threshold can eliminate this effect during the image analysis. Lastly, the lack of functional tests on T cells and comprehensive analysis of other relevant immune cells may reduce the embodiment of the complete immune microenvironment. Therefore, further independent studies are warranted to confirm our findings.
Despite these limitations, our study has several explicit advantages. Firstly, our customized mIF methods allowed for the synchronous detection of various protein markers in single tissue slide, providing a thorough perspective of the TME. Secondly, our tissue-based analysis revealed an unreported spatial interaction network of CD8+ T cells that could not be uncovered using dissociative techniques such as flow cytometry or single cell RNA sequencing. Meanwhile, quantitative digital analysis emphasizes the superiority of computer-assisted quantitation over the visual counting of positive cells by pathologists alone. Lastly, our exploration of the role of stromal cells in T cells, reported as the correlation between the CCPS of stromal cells and T cells, in contrast to most prior studies that provide only cell quantity, can better corroborate the hypothesis of stromal barriers based on a rational parameter combining quantity and spatial structure.
In summary, we performed a comprehensive decipherment of the spatial structure of intratumoral CD8+ T cells in NSCLC patients with LNM based on mIF images, including special spatial distribution of CD8+ T-cell functional subsets and spatial interplay between CD8+ T cells and their neighboring cells, which revealed the complexity of TME with significant implications for facilitating precise immunotherapy.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.