The role of spatial interplay patterns between PD-L1-positive tumor cell and T cell in recurrence of locally advanced non-small cell lung cancer

Abstract

Purpose

To explore the relationship between the spatial interaction of programmed death-ligand 1(PD-L1)-positive tumor cell and T cell with specific functions and the recurrence of non-small cell lung cancer (NSCLC) and optimize prognostic stratification.

Materials and methods

This study retrospectively included 104 patients with locally advanced NSCLC who underwent radical surgery. Tissue microarrays were constructed including tumor center (TC) and invasion margin (IM), and CK/CD4/CD8/PD-L1/programmed death-1 (PD-1) was labeled using multiplex immunofluorescence to decipher the counts and spatial distribution of tumor cells and T cells. The immune microenvironment and recurrence stratification were characterized using the Mann–Whitney U test and Cox proportional hazards model.

Result

Compared with the IM, the proportion of tumor cells (especially PD-L1⁺) was increased in the TC, while T cells (especially PD-1⁺) were decreased. An increase in TC PD-1⁺ CD8 T cells promoted relapse (HR = 2.183), while PD-L1⁺ tumor cells alone or in combination with T cells had no predictive value for relapse. In addition, in both TC and IM, CD8 were on average closer to PD-L1⁺ tumor cells than CD4, especially exhausted CD8. The effective density and percentage of PD-1⁺ CD4 T cells interacting with PD-L1⁺ tumor cells in the IM were both associated with recurrence, and the HRs increased sequentially (HRs were 2.809 and 4.063, respectively). Patients with low PD-1⁺CD4 count combined high PD-1⁺CD4 effective density showed significantly poorer RFS compared to those with high PD-1⁺CD4 count combined low PD-1⁺CD4 effective density, in both the TC and IM regions (HRs were 5.810 and 8.709, respectively).

Conclusion

Assessing the relative spatial proximity of PD-1/PD-L1 contributes to a deeper understanding of tumor immune escape and generates prognostic information in locally advanced NSCLC patients.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00262-023-03380-z.

Introduction

Tumor evasion from surveillance by the tumor microenvironment (TME) plays an important role in cancer progression, leading to poor patient outcomes [1, 2]. Programmed death-ligand 1(PD-L1) on the surface of tumor cells gives it the ability to block T lymphocyte function and promote its exhaustion [3, 4]. This process can be reversed with immune checkpoint inhibitors (ICIs) such as programmed death-1(PD-1)/PD-L1 antibodies, which have achieved great success in multiple cancer types [3, 5–9]. At present, ICIs therapy has become the standard treatment for locally advanced non-small cell lung cancer (NSCLC) [10, 11]. However, how to effectively screen the benefited population is still an urgent clinical problem to be solved.

The positive expression of PD-L1 in tumor cells, as a potential therapeutic biomarker, was controversial in predicting the efficacy and prognosis of ICIs in NSCLC [12–17], the combined evaluation of PD-L1 expression in tumor cells and the number of T lymphocytes to predict prognosis was also unsatisfactory [14, 18, 19]. Single-cell sequencing studies suggested that T lymphocytes with specific functional states affect the survival of NSCLC patients [20]. Theoretically, whether there are a certain number of tumor-specific T lymphocytes that are functionally reversible due to PD-1/PD-L1 pathway blockade/depletion around tumor cells expressing PD-L1 may be more accurate and effective biomarkers for predicting ICIs efficacy and prognosis [4, 21].

Multiplex immunofluorescence (mIF) technology can simultaneously detect multiple target proteins on a single tissue section, accurately identify the phenotype, function, and spatial location of target cells, and is a reliable tool for analyzing cell–cell interactions [22, 23]. Preliminary studies have found that the increased numbers of PD-1 and CD8 interacting with PD-L1 on the surface of tumor cells are negatively correlated with the prognosis of HPV-negative oropharyngeal squamous cell carcinoma [24]. In NSCLC, the functional status of T lymphocytes surrounding PD-L1 in tumor cells and its relationship with patient prognosis have not yet been fully studied.

In this study, mIF technology was used to precisely quantify PD-L1-positive tumor cells and target T lymphocytes with different functional states in the tumor center and invasive marginal region of locally advanced NSCLC patients, and to explore the relationship between their spatial interaction and recurrence, to further optimize individualized treatment.

Materials and methods

Patient cohorts and tissue microarray (TMA)

The study cohort consisted of consecutive patients with primary locally advanced NSCLC who had undergone surgical resection at the Shandong Cancer Hospital between 2014 and 2018 and for whom slides and formalin-fixed, paraffin-embedded (FFPE) tissue blocks were available. None of the patients received neoadjuvant chemotherapy or radiotherapy before radical surgery. All tissue specimens from the patient cohorts were obtained from FFPE tissue and were prepared as a TMA with a previously described standard procedure [25–27]. For each TMAs, two different blocks containing the tumor center (TC) and invasion margin (IM) of the same patient were used. IM was defined as the region centered on the border separating the host tissue from the malignant nests, with an extent of 1 mm [26, 28, 29]. All slides were examined by two experienced pathologists. A total of 104 patients were ultimately included in the study due to missing tissue, missing data, or poor staining quality. Inclusion and exclusion criteria are shown in Supplementary Fig. S1.

mIF staining

Manual mIF staining was performed to visualize the expression of CK, CD4, CD8, PD-L1, and PD-1 in TMA tissues in two panels. The staining procedure was as previously described [23, 25, 30, 31]. The slides (3 μm) were placed in a 65 °C oven for 2 h, deparaffinized and dehydrated with gradient ethanol, antigen retrieval, and blocking to remove endogenous peroxides, followed by dropwise addition of primary antibodies (diluted according to the antibody instructions) and horseradish peroxidase-labeled secondary antibodies. This process is repeated until all antigens have been stained with their respective fluorophores. Finally, nuclei were stained with 4′,6-diamidino-2-phenylindole (DAPI). Supplementary Table S1 lists the reagents and equipment used in mIF.

Multispectral imaging and imaging data analysis

The stained slides were scanned by a Vectra multispectral microscope (Akoya Biosciences). Monochromatic spectral decomposition of multispectral images was performed using the spectral library built into the inForm (version 2.4.8) software, and all spectrally unmixed images were subsequently subjected to the inForm active learning phenotyping algorithm. At least 2 investigators individually identified each DAPI-stained cell based on fluorophore expression patterns and nuclear/cellular morphological characteristics and correlated its phenotype to a specific x, y spatial coordinate. The quantitative and spatial distribution of cellular phenotypes was then automatically quantified using the R software (version 3.6.3) phenoptrReports package. The above analyses were performed without considering sample characteristics and clinical outcomes.

Statistical analysis

Chi-square tests determined pairwise comparisons between clinical variables and recurrence stratification. Comparisons between the two groups were performed using the Mann–Whitney U test. This nonparametric test was selected as the observations did not satisfy the Kolmogorov–Smirnov test of normality (P < 0.005). X-tile software determined cutoff values. Relapse-free survival (RFS) was defined as the time interval between radical surgery and the first recurrence or death, or between surgery and the last observation in recurrence-free patients. The cumulative incidence of RFS in the groups was described by the Kaplan–Meier method and compared with the log-rank test. Univariate and multivariate RFS analyses were conducted using Cox proportional hazards modeling. All statistical analyses were performed using IBM SPSS Statistics (version 25.0) and GraphPad Prism (version 9.0.0) software. P values were bilateral.

Results

Clinicopathological features of the locally advanced NSCLC patients

Totally 104 patients with locally advanced NSCLC were enrolled in this study. Table 1 lists the baseline clinicopathologic characteristics. The median age of the patients was 59 years (range, 53–66 years), and most patients were male (70%). Adjuvant chemotherapy and radiotherapy were received in 92 (88%) and 22 (21%) patients, respectively. The median follow-up time was 37 months (range, 27–48 months). The median RFS time was 27 months (range, 11–40 months). At the time of follow-up, 55 (53%) patients relapsed. Among the 104 NSCLC patients, 28% (29/104) of the patients recorded the epidermal growth factor receptor (EGFR) mutation status in the medical record system, of which 62% (18/29) were mutant and 38% (11/29) were wild type.

Table 1.

The clinical and pathological characteristics of locally advanced NSCLC patients

Parameter	Total (n, %) 104(100)	Relapse (n, %) 55(53)	Non-relapse (n, %) 49(47)	P
Age
≤ 65 y	75(72)	40(53)	35(47)	0.883
> 65 y	29(28)	15(52)	14(48)
Gender
Male	73(70)	42(58)	31(42)	0.145
Female	31(30)	13(42)	18(58)
Smoking index
≤ 400	55(53)	24(44)	31(56)	0.045*
> 400	49(47)	31(62)	18(38)
Tumor location
Central	52(50)	27(52)	25(48)	0.844
Peripheral	52(50)	28(54)	24(46)
Visceral pleural invasion
Yes	52(50)	27(52)	25(48)	0.884
No	52(50)	28(54)	24(46)
Histology subtype
Squamous cell carcinoma	50(48)	28(56)	22(44)	0.292
Adenocarcinoma	52(50)	25(48)	27(52)
Adenosquamous carcinoma	2(2)	2(100)	0(0)
Pathologic stage (AJCC8)
IIB	50(48)	22(44)	28(56)	0.184
IIIA	50(48)	30(60)	20(40)
IIIB	4(4)	3(75)	1(25)
Adjuvant chemotherapy
Yes	92(88)	50(54)	42(46)	0.408
No	12(12)	5(42)	7(58)
Adjuvant radiotherapy
Yes	22(21)	10(45)	12(55)	0.432
No	82(79)	45(55)	37(45)
EGFR mutation status				0.962
Mutant type	18(17)	9(50)	9(50)
Wild type	11(11)	5(45)	6(55)
Not clear	75(82)	35(47)	40(53)

Statistical significance was determined by two-sided Chi-square test. EGFR, epidermal growth factor receptor. *P < 0.05

Immune microenvironment was different between tumor center and invasion margin

To characterize the tumor immune microenvironment in locally advanced NSCLC, we used mIF staining to quantify the proportion and spatial distribution of cell subsets in tissue microarrays containing TC and IM regions. Serially sectioned tissues were stained with CD4 and CD8 mIF panels (Fig. 1a). Cell representations were identified using a supervised image analysis system (inForm 2.4.8). Nine cell subtypes (tumor cell/PD-L1⁺ tumor cell/PD-L1⁻ tumor cell/CD4 T cell/PD-1⁺ CD4 T cell/PD-1⁻ CD4 T cell/CD8 T cell/PD-1⁺ CD8 T cell/PD-1⁻ CD8 T cell) were classified based on positivity and relative intensities of all markers in a single panel (Fig. 1b).

Fig. 1.

Multiplex analysis of PD-1/PD-L1 checkpoints in the locally advanced NSCLC microenvironment. Examples of multiplex immunofluorescence (mIF) images (a) and summary of each defined cell phenotype (b)

The composition of the locally advanced NSCLC microenvironment showed a higher proportion of tumor cells to total cells and a lower proportion of CD4 and CD8 to total cells in the TC compared to the IM (TC vs. IM, 57 vs 18%, P = 0.000; 20 vs 34%, P = 0.000; 5 vs 16%, P = 0.000) (Fig. 2a, c). In addition, tumor cells in the TC were more likely to express PD-L1 (35% vs 3%, P = 0.000) (Fig. 2b, d), and CD4 in the TC was less PD-1 expressing compared with the IM (5 vs 8%, P = 0.802) (Fig. 2b, e). The component of PD-1/PD-L1 showed that most immune cells express PD-L1 except tumor cells and dominantly IM (51 vs 97%, P < 0.001) (Fig. 2g). In addition to T cells, a minority number of non-T cells express PD-1, mainly in the TC (39 vs 31%, P = 0.026) (Fig. 2h). Collectively, these results suggested that T cell infiltration was more abundant at the IM compared to the TC, where PD-1, especially CD4, was upregulated on the surface of tumor-infiltrating T lymphocytes.

Fig. 2.

Relative distribution of cell phenotypes in the tumor center and the invasive margin. a–b Representative images of multiplex immunofluorescence (mIF) at the tumor center and corresponding invasive margins in a squamous cell carcinoma patient. c Relative distribution of cell phenotypes in TC and IM. Relative distribution analysis of tumor cell (d), CD4 T cell (e), CD8 T cell (f), PD-1⁺ cell (g), and PD-L1⁺ cell subpopulations (h). Significance determined by Mann–Whitney U test. *P < 0.05, **P < 0.01, ***P < 0.001

Survival analyses of tumor cell expressing PD-L1 and its target cell

Next, we sought to understand whether the number of defined cell subsets is correlated with patient RFS (Fig. 3). The cell fraction was defined as the number of cell subsets divided by the total number of cells. We found that lower levels of IM PD-1⁺CD4⁺ (low vs high, median RFS, 19.5 vs not yet reached months; P = 0.019) and TC PD-1⁺CD8⁺ (20.0 vs 45.0 months; P = 0.033) were associated with inferior RFS in 104 patients (Fig. 3h, i). The clinicopathological characteristics of the patients were not associated with recurrence (Supplementary Table S2). TC PD-1⁺CD8⁺ T cells [low vs. high, hazard ratio (HR) = 2.183, 95% confidence interval (CI) = 1.140–4.184, P = 0.019] was significantly associated with RFS, as revealed by multivariate Cox analysis (Table 2). Collectively, these data highlighted the clinical relevance of PD-1⁺ T cells in the survival of locally advanced NSCLC patients.

Fig. 3.

Kaplan–Meier analyses of relapse-free survival (RFS) for PD-L1⁺ tumor cell and T cell in subgroups of patients. Cumulative RFS of PD-L1⁺CK⁺ and T cell was calculated by the Kaplan–Meier method, and statistical analyses were generated using the log-rank test

Table 2.

Multivariate Cox regression analysis for relapse-free survival

	Multivariate HR (95% CI)	P
Model 1
Gender (female vs male)	0.906 (0.402, 2.041)	0.812
Smoking index (> 400 vs ≤ 400)	1.584 (0.782, 3.210)	0.202
Pathologic stage (IIIA vs IIB)	1.310 (0.741, 2.315)	0.352
Pathologic stage (IIIB vs IIB)	2.083 (0.600, 7.232)	0.248
TC PD-1+CD8+ (low vs high)	2.183 (1.140, 4.184)	0.019*
IM PD-1+CD4+ (low vs high)	1.675 (0.965, 2.907)	0.066
Model 2
Gender (female vs male)	0.717 (0.314, 1.639)	0.431
Smoking index (> 400 vs ≤ 400)	1.367 (0.680, 2.748)	0.380
Pathologic stage (IIIA vs IIB)	1.079 (0.595, 1.958)	0.801
Pathologic stage (IIIB vs IIB)	0.782 (0.203, 3.016)	0.721
IM PD-1+CD4+ effective density (high vs low)	2.809 (1.477, 5.343)	0.002*
Model 3
Gender (female vs male)	0.760 (0.333, 1.734)	0.514
Smoking index (> 400 vs ≤ 400)	1.637 (0.804, 3.334)	0.174
Pathologic stage (IIIA vs IIB)	1.192 (0.671, 2.116)	0.550
Pathologic stage (IIIB vs IIB)	1.773 (0.514, 6.116)	0.365
IM PD-1+CD4+ effective percentage (high vs low)	4.063 (1.979, 8.341)	0.000*
Model 4
Gender (female vs male)	0.570 (0.227, 1.429)	0.230
Smoking index (> 400 vs ≤ 400)	0.791 (0.351, 1.782)	0.571
Pathologic stage (IIIA vs IIB)	1.496 (0.756, 2.962)	0.248
Pathologic stage (IIIB vs IIB)	2.137 (0.469, 9.734)	0.326
TC PD-1+CD4+ density & effective density (lohi vs hilo)	5.810 (1.248, 27.045)	0.025*
Model 5
Gender (female vs male)	0.290 (0.075, 1.119)	0.072
Smoking index (> 400 vs ≤ 400)	1.120 (0.395, 3.173)	0.831
Pathologic stage (IIIA vs IIB)	0.792 (0.301, 2.082)	0.637
IM PD-1+CD4+ density & effective density (lohi vs hilo)	8.709 (2.950, 25.714)	0.000*

HR hazard ratio, CI confidence interval. *P < 0.05

In addition, we analyzed the prognostic value of the number of PD-L1⁺ tumor cells combined with T cells (Fig. 4). Patients with high IM PD-L1⁺CK⁺ and low IM PD-1⁺CD4⁺ showed significantly poorer RFS compared to those with a low IM PD-L1⁺CK⁺ and IM PD-1⁺CD4⁺ (16.5 vs not yet reached months; P = 0.019), but was not an independent risk factor for recurrence (Fig. 4f and Supplementary Table S3).

Fig. 4.

Survival analysis of locally advanced NSCLC patients based on the expression levels of the combination of PD-L1 (on tumor cells) and T cells. P values were calculated by the Kaplan–Meier method, and statistical analyses were generated using the log-rank test

Spatial analysis showed nearest neighbor distances of PD-1 and PD-L1

The nearest neighbor distances (NND) analysis was performed, which measures the distance from each T cell to the nearest PD-L1⁺ tumor cell (Fig. 5a). This analysis revealed that CD8 T cell was, on average, closer to PD-L1⁺ tumor cells than CD4 T cell, both in the TC and the IM (Fig. 5b, c). We also extended this assessment to the spatial location of CD4 and CD8 functional subsets and found that in the TC region, PD-1⁺CD8⁺ T cells were closer to PD-L1⁺ tumor cells than CD8 T cells (Fig. 5b, c). These findings led us to hypothesize that the different spatial distributions of CD4 and CD8 T cells may represent differences in their biological functions with prognostic significance.

Fig. 5.

Nearest neighbor distance (NND) analysis for T cell and PD-L1⁺ tumor cells. a Schematic representation of the quantification of NND. NND calculated from each T cell to their nearest PD-L1⁺ tumor cells (b) and individual value plot of the average NND (c). P values were calculated with the Mann–Whitney U test, and all data are presented as the mean ± SEM

PD-L1⁺ tumor cell-adjacent PD-1⁺ CD4 T cell was correlated with patient recurrence

Given the ability of multiplexed imaging and machine learning to precisely define the location of individual cell phenotypes, we further detailed the prognostic value of effective density (the absolute number of PD-L1⁺ tumor cells near T cells within a 30 μm radius) (Fig. 6a). This radius was preselected to identify target cell populations most likely to engage in potent, direct, cell-to-cell interactions with PD-L1 on the surface of tumor cells, consistent with previous studies in lung cancer [24, 29, 31]. Interestingly, we found that patients with higher effective densities of IM PD-1⁺CD4⁺ had significantly shorter RFS than those with lower effective densities (11.5 vs 47.0 months; P = 0.000) (Fig. 6b). The prognostic value remained significant after adjustment using a multivariate Cox model (high vs. low, HR = 2.809, 95% CI = 1.477–5.343, P = 0.002) (Table 2). Other T cell phenotypes were not linked with RFS (Fig. 6b).

Fig. 6.

Higher IM PD-1⁺CD4⁺ effective percentage was associated with significantly shorter RFS. a Schematic representation of spatial analysis involving PD-L1⁺ tumor cells and T cells. Red and green dots represent PD-L1⁺ tumor cells and T cells, respectively. The white line connecting the red and green dots indicates that the distance between the two cells is less than 30 microns. The absolute number of green dots is defined as the effective density. Scale bar, 50 μm. b RFS based on the effective densities (30 μm) of T cells. Statistical relevance was defined using the log-rank test. c Schematic diagram for calculating the effective percentage of PD-1⁺ CD4 T cells. Effective percentage = the effective density of PD-1⁺ CD4 cells/the effective density of CD4⁺ cells. EP: effective percentage. d RFS based on the effective percentage of PD-1⁺ CD4 T cells. Statistical relevance was defined using the log-rank test

Based on the above results, we had reason to hypothesize that the effective density ratio of PD-1⁺ CD4 to CD4 T cell may be a more effective marker for predicting the recurrence of locally advanced NSCLC. Therefore, we introduced the concept of effective percentage (Fig. 6c). A significantly higher effective percentage of PD-1⁺ CD4 T cells within a 30-μm radius was associated with poorer RFS (6.5 vs 45.0 months; P = 0.000) (Fig. 6d). Multivariate Cox analysis showed that patients with a higher effective percentage of PD-1⁺ CD4 T cell had the highest risk of recurrence, with an HR of 4.063 (Table 2). These results suggested that the impact of PD-1/PD-L1 on patient survival depends not only on their numbers but also on their proximity.

T cell count combined with effective density can assess more accurate recurrence prediction

Next, we sought to assess the effect of T cell number combined with effective density on relapse. Patients with low PD-1⁺CD4⁺ count and high PD-1⁺CD4⁺ effective density showed significantly poorer RFS compared to those with high PD-1⁺CD4⁺ count and low PD-1⁺CD4⁺ effective density, in both the TC and IM regions (TC, 6.5 vs 45.0 months; P = 0.002; IM, 7.5 vs not yet reached months; P = 0.000), and were independent risk factors for recurrence (TC, HR = 5.810, 95% CI = 1.248–27.045, P = 0.025; IM, HR = 8.709, 95% CI = 2.950–25.714, P = 0.000) (Fig. 7; Table 2; Supplementary Fig. S2). Overall, our data emphasize that both T cell count and proximity to PD-L1⁺ tumor cells should be considered when predicting prognosis.

Fig. 7.

Survival analysis based on the T cell count combined effective density. P values were calculated by the Kaplan–Meier method, and statistical analyses were generated using the log-rank test

Discussion

In the present study, we established a single-cell resolution-based point-to-point pattern analysis platform in tissue in situ using mIF to guide prognostic stratification in patients with locally advanced NSCLC. Notably, this approach allowed us to map cell subtypes with specific functions, characterize in situ the number and proximity of PD-L1⁺ tumor cells and their target T cells, and assess their value in relapse. We demonstrated that an increased effective density of PD-1⁺ CD4 T cells that efficiently interact with PD-L1⁺ tumor cells in the invasion margin area was a key risk factor for recurrence in locally advanced NSCLC.

Previous studies have shown that quantification of PD-L1⁺ tumor cells and T cells alone or in combination was controversial in predicting prognosis [12–14, 18, 19], which may be related to the failure to distinguish between tumor center and invasive margin [16, 22]. Tumor borders or core regions have distinct microenvironmental characteristics [32, 33]. The invasive margin is the front line of the body's anti-tumor response, and its infiltrating T cells are more likely to be transformed into an exhausted state under the continuous stimulation of tumor antigens. The expression of PD-1 by T cells indicates that T cells are exhausted [20, 34], and our study confirmed an increased proportion of exhausted T cells in the invasive margin region compared to the tumor center. Furthermore, although we did not find a relationship between PD-L1⁺ tumor cells and recurrence, our findings suggested that increased PD-1⁺ CD8 T cell infiltration in the tumor center promotes NSCLC recurrence, which was consistent with the results of single-cell sequencing [20].

The relative proximity of PD-L1 and PD-1, as evidence of physical interaction between the two, is necessary for tumors to develop adaptive immune resistance, and assessing their spatial proximity is expected to provide more accurate prognostic and predictive information than assessing the amount of PD-L1/PD-1 alone or in combination. For example, increased numbers of PD-1 and CD8 adjacent to tumor cells expressing PD-L1 were both negatively correlated with OS in HPV-negative oropharyngeal squamous cell carcinoma [24], and frequent interaction between PD-1 and PD-L1 predicted metastatic melanoma PD-1 inhibitor treatment response [35]. Our study confirmed that at the invasion margin, an increase in the effective amount of exhausted CD4 (PD-1⁺CD4⁺) interacting with PD-L1 on the tumor cell surface promoted recurrence after radical resection of locally advanced NSCLC. To our knowledge, this is the first relevant study to evaluate the effect of the effective interaction of PD-L1-expressing tumor cells with PD-1 on the surface of which target cells on the prognosis of NSCLC. It is worth noting that a previous preclinical study found that PD-1 blockade directly targets PD-1⁺ CD4 T cells [36]; other studies have shown that specific CD4 T cells in tissue in situ were blocked by PD-1 inhibitors can promote the proliferation of CD8 T cells, and the tumor regresses significantly [4]. Furthermore, recent studies have demonstrated that the spatial interaction of PD-1 and PD-L1 can predict NSCLC immunotherapy outcomes. Therefore, although our results have not been validated in immune checkpoint inhibitor efficacy prediction models, it is reasonable to speculate that the frequent proximity of PD-L1 on the surface of tumor cells to PD-1 on the surface of CD4⁺ T lymphocytes is expected to more accurately predict PD -1/PD-L1 inhibitor treatment efficacy.

This study also has limitations. This study is retrospective and has a small sample size, and a larger study cohort is needed to confirm the results. In addition, the study included patients with locally advanced NSCLC, which may limit the applicability of the results to early-stage NSCLC. Furthermore, we did not identify CD4 subsets that might interact with PD-L1-positive tumor cell. The expression of PD-1 in non-T lymphocytes and PD-L1 in immune cells is abundant, but the specific cell type and its value in the recurrence of locally advanced NSCLC are not clear, and it is necessary to explore this subset of cells in future studies prognosis and efficacy prediction.

Altogether, our results highlighted the value of assessment of two-dimensional spatial markers of the TME in predicting recurrence in locally advanced NSCLC. Exploring the invasive margin region will provide important insights into the complex and heterogeneous tumor immune microenvironment associated with the efficacy of ICIs therapy.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

DAPI

4′,6-Diamidino-2-phenylindole

EGFR

Epidermal growth factor receptor

FFPE

Formalin-fixed, paraffin-embedded

IM

Invasion margin

ICIs

Immune checkpoint inhibitors

mIF

Multiplex immunofluorescence

NSCLC

Non-small cell lung cancer

NND

Nearest neighbor distances

PD-L1

Programmed death-ligand 1

PD-1

Programmed death-1

RFS

Relapse-free survival

TME

Tumor microenvironment

TMA

Tissue microarray

TC

Tumor center

Author Contributions

LY, LX, and XS contributed to conception and design of the study. LY, WZ, JS, GY, SC, and FS organized the database. LY, WZ, JS, GY, and SC performed the statistical analysis. LY wrote the first draft of the manuscript. LX and XS wrote sections of the manuscript. All authors contributed to manuscript revision, read, and approved the submitted version.

Funding

This work was supported by grants from the National Natural Science Foundation of China (Grant No. 82172866), the Shandong Provincial Natural Science Foundation (Grant No. ZR2021LZL005), the Shandong Provincial Natural Science Foundation (Grant No. ZR2019LZL019), and the Department of Science & Technology of Shandong Province (Grant No. 2021CXGC011102).

Data Availability

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

Declarations

Conflict of interest

The authors declare that they have no conflict of interest.

Ethics approval

This study was approved by the Ethics Review Committee of Shandong Cancer Hospital and complied with the provisions of the Declaration of Helsinki. This study was a retrospective analysis, and informed consent was not required.