In the current study, we analyzed a TMA set containing tumor specimens obtained from a large cohort of patients with stage I-III NSCLC using 23 markers, including T-cell, B-cell, immune checkpoint, and myeloid cell markers, placed in 5 mIF panels. Cord plots and UMAP clusters based on marker co-expression were used to visualize different TAIC phenotypes across different panels. In addition, we identified rare cell types, showing the diversity of cell populations in NSCLC. Immune checkpoints expressed by malignant cells were observed in several combinations within specimens, as well as in TAICs across specimens. Overall, we observed two patterns of cellular distribution—mixed and unmixed—related to T-cells, B-cells, macrophages, and PMNs as the main cell phenotypes. Cellular patterns of distribution, as well as the distance between malignant cells and various cell phenotypes, showed different associations with clinicopathologic characteristics, including smoking status, tumor size, final tumor stage, adjuvant therapy, and mutational status. Kaplan-Meier survival curves showed that patterns and distances from malignant cells to different cell phenotypes were also associated with OS and RFS. Finally, we identified four groups of cellular immunologic patterns, which were also associated with OS and RFS in univariate analysis and Cox proportional hazards regression models.
We found that malignant cells expressing B7-H3 were most commonly observed in NSCLC, followed by malignant cells expressing PD-L1, OX40, B7-H4, and IDO-1. These checkpoints have adverse regulatory functions over T lymphocytes (21–24). Other immune checkpoints observed in our cohort, such as OX40 and ICOS, have co-stimulatory signals for T-cell activation in normal and pathologic conditions (25–27). Cords plots and UMAP plots showed that various TAICs and malignant cells could express these checkpoints. Although it was thought for a long time that OX40 expression was restricted to activated conventional T-cells, other TAICs, including malignant cells, have since been shown to express this marker (28). Overall, we observed higher densities of immune checkpoint markers in SCC than in ADC, particularly PD-L1, B7-H3, and B7-H4, showing that these immune checkpoints are predominantly expressed in solid tumors (29), creating a more immunosuppressive microenvironment. It is clear that in NSCLC, these immune checkpoint pathways are expressed simultaneously and are an essential mechanism of immune resistance against T-cell response (30, 31). Our findings confirm that malignant cells could express more than one checkpoint marker simultaneously, indicating that lung tumors can use more than one pathway to avoid the immune system (32, 33). Immune checkpoints are essential regulators of the immune system, initiating a productive immune response, preventing the onset of autoimmunity, or using tumors to avoid the immune system (34). In agreement with other studies (31), our results show that TAICs and malignant cells essentially drive these suppressive pathways. This knowledge of simultaneous co-expression can guide the study of rational combinations of agents for potential new therapeutic approaches.
Although we observed overall increased amounts of CD3 + CD8 + cytotoxic T-cells, CD3 + CD45RO + memory T-cells, CD3 + CD8 + CD45RO + cytotoxic memory T-cells, and CD20 + B-cells in the NSCLC specimens, we also detected a number of TAICs expressing co-inhibitory and co-stimulatory signatures, including PD-1, LAG3, TIM3, FOXP3, ICOS, and OX40, in higher amounts in SCC than in ADC. Similarly, we observed that CD68 + macrophages, CD68 + CD11b + myeloid dendritic cells, and CD66b + PMNs were more predominant in SCC than in ADC. These observations suggest that various immunosuppressive cells are present in these tumors, possibly reflecting their relation to other factors such as smoking status or chronic obstructive pulmonary disease, increasing these cell populations as described previously (35). Although myeloid cell phenotypes such as CD68 + Arg-1 + type II macrophages, CD68 + Arg-1 + CD11b + type II immature tumor-associated macrophages, CD66b + CD11b + immature PMNs, Arg-1 + CD33 + CD14 + CD11b + monocytic MDSCs, and CD33 + CD66b + CD11b + MDSC-PMNs were observed in low densities, these were also present, playing their immunosuppressive roles (36).
By mapping the spatial organization of the TAICs as previously described (17), we observed mixed and unmixed cellular distribution patterns. Likely related to the dysfunctional signature observed in melanoma tumor tissues (37), we observed two groups of cells in relation to malignant cells: an immunosuppressive group, which has a mixed pattern indicating close interaction with malignant cells, and an immunoprotective group, which has an unmixed pattern with apparently less interaction with malignant cells. This immunoprotective group included CD3 + CD8 + cytotoxic T-cells, CD3 + CD8 + CD45RO + cytotoxic memory T-cells, CD3 + CD45RO + memory T-cells, and CD68 + macrophages.
Analysis of nearest neighbor median distance from malignant cells to CD3 + T-cells and macrophages showed that CD68 + cells are closer to malignant cells than are CD20 + B-cells and CD66b + PMNs. Although CD3 + CD8 + cytotoxic T-cells were among the most abundant cells in both ADC and SCC specimens, not many of these were close to malignant cells. In contrast, T-cells expressing PD-L1, B7-H3, B7-H4, IDO-1, and OX40 were located close to the malignant cells, suggesting that the distances from malignant cells and pattern of distribution, rather than the density of these cells, plays a critical role in cancer. Interestingly, CD3 + CD45RO + memory T-cells, CD3 + CD8negFOXP3 + regulatory T-cells, B-cells, and most myeloid cells were located far from malignant cells. However, CD3 + CD8negFOXP3 + regulatory T-cells play a solid immunosuppressive role in the tumor environment by releasing inhibitory cytokines (38), facilitating the action of other cell inhibitors. Cellular distribution patterns and cellular distances are not frequently studied and are not very well understood, but these patterns can give us essential information about tumor tissue biological processes related to different tumor characteristics (39).
The association of cellular distribution patterns with clinicopathologic features has not been previously described, and we believe that this information can help us better understand the biological behavior of tumors. For example, we observed that ADC specimens from smokers overall had a mixed pattern of distribution of CD3 + T cells and an unmixed pattern of CD3 + CD8 + CD45RO + effector memory T-cells. The largest ADC tumors showed an unmixed pattern of CD3 + PD-L1 + cells. In ADC KRAS-mutant tumors, we observed a mixed pattern of CD68 + PD-L1 + cells. In SCC, we observed changes in cellular distribution patterns predominantly in MDSC populations, which showed a mixed pattern of CD11b + CD66b + CD33 + cells in smaller tumors. Patients who received adjuvant therapy also had tumors with an unmixed pattern of malignant cells, with several T-cell inhibitory proteins and MDSC populations. ADC specimens from smokers also showed a close median distance from malignant cells to PD-L1 + T-cells, CD3 + CD8 + PD-L1 + cytotoxic T-cells, M2 macrophages, and MDSC-PMNs. This can be interpreted as tobacco’s immunosuppressive effect on the tumor microenvironment (35). In contrast, ADC specimens from nonsmokers showed close median distances from malignant cells to CD3 + CD8negFOXP3 + regulatory T-cells and, most importantly, from CD3 + CD8 + cytotoxic T-cells and CD3 + CD45RO + memory T-cells to CD3 + CD8negFOXP3 + regulatory T-cells, suggesting a different mechanism of inhibition than that observed in tumors from smokers.
Small tumors in ADC showed closer proximity from malignant cells to CD3 + CD8 + cytotoxic T-cells, CD3 + CD8 + CD45RO + cytotoxic memory T-cells, CD3 + CD8negFOXP3 + regulatory T-cells, CD3 + CD45RO + FOXP3 + memory regulatory T-cells, IDO-1 + T-cells, IDO-1 + B-cells, and LAG3 + B-cells compared with the larger tumors, suggesting that small tumors are more enriched in inhibitory signals. In SCC, large tumors showed closer distances from malignant cells to CD68 + macrophages and IDO-1 + T-cells than did small tumors. Furthermore, in ADC, stage II tumors showed closer proximity from malignant cells to TIM3 + B-cells than did stage III tumors, and stage I tumors showed closer proximity from cytotoxic T-cells to PD-L1 + macrophages and regulatory T-cells than did stage II tumors, suggesting changes according to the stage of the tumor. In tumors from patients who received adjuvant therapy after surgery, we observed closer distances from malignant cells to PD-1 + PD-L1 + cytotoxic T-cells and PD-L1 + macrophages than in tumors from patients who did not receive adjuvant therapy. This finding indicates that these markers could aid in the decision to administer adjuvant therapy, together with the current recommendations (40).
We found that distribution patterns and distances from malignant cells to different cell phenotypes could be associated with outcomes. Limited penetration among malignant cells, indicated by an unmixed pattern of distribution of MDSC-PMNs, was associated with poor RFS, and an unmixed pattern of CD66b + PMNs and MDSC-PMNs was associated with poor OS in ADC. These findings suggest that these cell phenotypes are acting as a barrier, limiting the actions of other activated T-cells. MDSCs are known to suppress T-cell activation and toxicity using various mechanisms (41). Although TIM3 has been reported in several T-cell populations, as well as dendritic cells and monocytes (42), we observed that a subset of CD20 + B-cells co-expressed coinhibitory molecules such as TIM3, and a mixed pattern of distribution was associated with poor OS. The action of TIM3 and other inhibitor molecules on this type of cell is unknown, and further investigation will be needed to understand the implications of these co-expression patterns. A mixed pattern of CD3 + CD8 + GZB + activated cytotoxic T-cells was associated with better RFS, suggesting that the interaction of these cells with malignant cells could prevent tumor recurrence in SCC. In contrast, a mixed pattern of CD3 + VISTA + T-cells was associated with poor OS, showing that the distribution of these cells next to malignant cells increases their immunosuppressive action. VISTA has a potential immunosuppressive influence on the tumor microenvironment for several types of T-cells in various cancers, including NSCLC, and immunotherapy targeting this protein is currently under investigation (43).
It is known that most, if not all, malignancies trigger an innate inflammatory response that builds up a pro-tumorigenic microenvironment that can resist treatment (44). This suggests that the close proximity of immunosuppressive cells to malignant cells and increased interactions between these cells in NSCLC, as we observed in the current study, enable tumors to avoid the immune system. In ADC, we also observed that EGFR-mutant tumors had a significantly closer median distance from malignant cells to PD-1 + PD-L1 + T-cells and from CD3 + CD8 + GZB + activated cytotoxic T-cells to CD3 + CD8negFOXP3 + regulatory T-cells than did wild-type tumors. Compared with wild-type tumors, KRAS-mutant tumors showed closer distances from malignant cells to MDSC-PMNs, suggesting that mutational status affects cellular distribution.
Kaplan-Meier curves showed that in ADC, close distances from malignant cells to PMNs and MDSC-PNMs are associated with worse RFS than long distances. In SCC, close distances from malignant cells to PD-L1 + T-cells and long distances to ICOS + T-cells were associated with worse RFS. Our data also showed that close proximity of malignant cells to CD3 + CD8 + cytotoxic T-cells, CD3 + CD8 + GZB + activated cytotoxic T-cells, and macrophages was associated with better OS than long distances in ADC, suggesting that the cell-to-cell proximity of these cells mitigates the suppressive effect of inhibitory cells and supporting the findings of Barua et al (45). We also found that close distance from malignant cells to B7-H3 + T-cells was associated with worse OS. This suggests that the cellular spatial distribution of specific cell phenotypes is an independent factor associated with poor or better prognosis and can be used to select combinations of therapeutic strategies and determine patient prognosis (39), not only in NSCLC but also in other cancers (12).
Finally, by combining patterns of cellular distribution with cellular distances, we identified four groups of cellular immunologic patterns. In univariate analysis, we found that an unmixed pattern of CD3 + T-cells with long distances from malignant cells was associated with poor OS and RFS in ADC, and an unmixed pattern of CD68 + with long distances from malignant cells was associated with poor OS. An unmixed pattern of CD3 + PD-L1 + cells with close distances from malignant cells and a mixed pattern of CD3 + ICOS + cells with close distances from malignant cells were associated with better RFS. ICOS is known to have a dual role; co-stimulation confers an anticancer response, but ICOS signaling also engages regulatory T-cell activity induction (46), suggesting that the location and distribution of ICOS + cells can influence the activation of one of its roles. Furthermore, in multivariable analysis, a mixed pattern of CD3 + IDO-1 + cells with close distances to malignant cells was identified as an independent marker associated with poor OS, suggesting that location and distance of IDO1 + cells from malignant cells can drive an immunosuppressive microenvironment, which may be responsible for the resulting poor prognosis. IDO expression has been detected in various TAIC populations and has been shown to be related to lung cancer progression (47). In contrast, wild-type EGFR, a mixed pattern of CD3 + CD45RO + memory T-cells with close distances to malignant cells, and a mixed pattern of CD68 + macrophages with long distances to malignant cells were associated with better OS in multivariable analysis, suggesting that not only cellular patterns of distribution but also distance from malignant cells can influence outcomes. These findings highlight the importance of better understanding the complex relationships between malignant cells and immune cells in terms of their spatial distribution to direct the study of new therapeutic approaches.
The current study has some limitations. First, although our NSCLC specimens were collected retrospectively, which allowed us a large enough sample to examine varying cell phenotypes, those phenotypes were displaced in different independent mIF panels, which limited the integration of the different cell phenotypes. Second, most of the patients from our cohort were smokers, which can influence the analysis between nonsmokers and smokers. Lastly, our specimens were placed in TMA format, which may induce under- or overrepresentation of the marker levels and spatial distribution owing to tumor heterogeneity.
In summary, our data showed that tumor cells and TAICs could produce multiple inhibitory factors in NSCLC. In studying the spatial distribution of various cellular populations, we could identify other associations between these cells and clinicopathologic variables in surgically resected ADC and SCC specimens. In addition, we identified several associations between specific cellular patterns of distribution and their distances that can negatively or positively influence patient outcomes; however, validation of our findings using a similar cohort of patients is needed.