This study included a total of 378 patients with NSCLC from two cohorts, who had received treatment at the First People’s Hospital of Yunnan Province from November 2011 to November 2019. Cohort A (328 cases) received non-immunotherapy (Surgery, EGFR-TKIs, chemotherapy, etc.), and cohort B (50 cases) received immunotherapy (anti-PD1, and anti-PD1 plus chemotherapy). Clinical and pathological data, including age at diagnosis, gender, smoking history, tumor pathological type, clinical stage, epidermal growth factor receptor (EGFR) mutation status, and therapy was collected in accordance with study protocol requirements (online supplemental table 1) and online supplemental table 2). Patients with ALK or ROS1 translocation were not included in this study. The workflow of this study was shown in online supplemental figure 1. Patient’s paraffin-embedded tumor tissue was collected to examine immune microenvironment by immunohistochemistry (IHC) staining and to identify tumor genetic variations by next-generation sequencing. The paraffin-embedded tissue blocks were stored at low temperature (−80°C) before use. All of the tissue sections were cut from paraffin-embedded tissue blocks just before starting the experiment.

Two tissue sections of each case were used for immunohistochemical staining of CD8, PD-L1, CD68, CD163, and PD-1. Section 1 was used for double fluorescence costaining of CD8 and PD-L1. Section 2 was used for traditional diaminobezidin (DAB) staining of PD-L1, and the staining result was also used as a reference and quality control for the fluorescent staining result of PD-L1 on section 1. An additional section 3 from samples in cohort A was used for multiplex fluorescence costaining of CD8, CD68, CD163, and PD-1. The sections were stained in multiple batches. The identification and control of batch-to-batch differences were based on quality control samples with known positive degrees of each marker derived from a lung adenocarcinoma tissue, a lung squamous cell carcinoma tissue, and a tonsil tissue from autopsy.

For multiplex fluorescence immunohistochemical staining, paraffin-embedded tissue blocks were serially sectioned into 3 µm sections. Subsequently, a series of processing and staining was performed on the slides according to the kit instruction (Opal 7-Color IHC Kit, NEL797B001KT, PerkinElmer) as described in our previous study,15 including epitope retrieval, endogenous peroxidase and protein blocking, antigen labeling, and tyramide signal amplification (TSA) visualization. CD8 antibody (clone 144B, ab17147, Abcam; dilution 1:25), PD-L1 antibody (clone SP142, Ventana; dilution 1:25), CD68 antibody (clone KP1, ZM-0060, Zsbio; dilution 1:400), CD163 antibody (clone 10D6, ZM0428, Zsbio; dilution 1:200), and PD-1 antibody (clone CAL20, ab237728, Abcam; dilution 1:100) were used as primary antibodies in this study. For section 1, TSA visualization was performed with the fluorophores DAPI (nucleus), Opal 540 (CD8), and Opal 690 (PD-L1). For section 3, the fluorophores were 4’,6-diamidino-2-phenylindole (DAPI) (nucleus), Opal 520 (CD8), Opal 570 (CD68), Opal 620 (CD163), and Opal 690 (PD-1). Slides were scanned using the PerkinElmer Vectra (Vectra V.3.0.5, PerkinElmer). Multispectral images were unmixed using spectral libraries built from images of single-stained tissue samples for each reagent using inForm Advanced Image Analysis software (inForm V.2.3.0, PerkinElmer). The “tumor mask” related function of inForm software was used to define the tumor compartment on the sections. 20 representative multispectral images were selected as training samples to build algorithm (tissue segmentation, cell segmentation, phenotyping tool, and positivity score) using inForm software. Then, the algorithm was applied to batch analysis of all the images. Next, the infiltration levels of CD8+, CD8+PD-L1+, PD-1+, CD8+, CD8+PD-1+, CD68+, CD68+PD-1+, CD68+CD163+, CD163+, and CD163+PD-1+ cells were determined by the proportion of positive cells to total cells in the tumor compartment. Two pathologists confirmed the quality and results of the experiment.

For traditional DAB immunohistochemical staining, paraffin-embedded tumor tissue samples were sectioned at a thickness of 4 µm and transferred to coated glass slides. For PD-L1 staining on section 2, the slides were stained with a Ventana GX automated system (Ventana, AZ, USA). The primary antibody specific for PD-L1 (clone SP142) was diluted 1:25 and incubated for 32 min at room temperature. The antibody was detected with the Ventana Amplification Kit and Ventana ultraView Universal DAB Detection Kit. Digital images were captured using an Aperio Scanscope AT Turbo slide scanner with 20× magnification. Scoring of PD-L1 expression was performed using a digital image analysis software, namely, Aperio Membrane V.9 and Aperio Genie Classifier. PD-L1 expression was reported as a continuous variable of the percentage of TC staining with any intensity. PD-L1 expression in each sample was classified as PD-L1 negative, low, or high. PD-L1 negative was defined as <1% of TC staining. PD-L1 high was defined as ≥50% of TC staining. PD-L1 low was defined as 1%–49% of TC staining. These scoring systems were based on previous studies using an SP142 assay.3 16 17

DNA extraction from the formalin-fixed and parrffin-embedded (FFPE) tumor specimens and targeted gene capture sequencing for the tumor mutational burden (TMB) test were performed using the standard protocols mentioned previously.18 A total of 2 mL of whole blood was collected from each patient, and DNA from the peripheral blood lymphocytes was extracted as a normal control. DNA libraries were captured with a designed Genescope panel of 543 genes (Genecast, Beijing, China) which are mainly tumor-related genes and cover 1.7 Mb of the entire genome. The captured samples were subjected to Illumina NovaSeq 6000 platform using the pair-end sequencing method.

The TMB was defined as the number of somatic, coding, base substitutions, and indel mutations per megabases of the genome examined. Only the regions with sequencing depth larger than 100× after deduplication were taken into TMB calculation. All base substitutions and indels in the coding region of the targeted genes, including synonymous alterations, were initially counted before filtering as described above. Alterations that were predicted to be germline by the somatic-germline-zygosity algorithm were not counted. Known germline alterations in dbSNP were not counted. Germline alterations occurring with two or more counts in the ExAC database were not counted.21 To calculate the TMB per megabases, the total number of mutations counted was divided by the size of the coding region of the targeted territory. TMB was reported as a continuous variable. According to the TMB levels, the patients were divided into high, moderate, and low groups. The grouping criteria were based on the 75th percentile and 25th percentile of this batch of data. Then, TMB levels greater than or equal to the 75th percentile was defined as high. TMB levels less than the 25th percentile were defined as low, and the moderate levels occurred between the 25th and 75th percentiles.

The CD8+PD-L1+ Tils are common in NSCLC tissues. Figure 1A shows an example of adenocarcinoma tissues infiltrated by CD8+PD-L1+ Tils. The infiltration levels of those cells in cohort A ranged from a minimum of 0% to a maximum of 15%. The median frequency was 0.13% (IQR: 0.01%–0.77%). Only 23% of the NSCLC samples showed no such cells, which were reported as 0%. Correlation analysis showed that there was a weak positive correlation between the CD8+PD-L1+ Tils and the overall CD8+ Tils (Spearman r=0.5243, p<0.0001), as well as between the CD8+PD-L1+ Tils and the CD8+PD-L1⁻ Tils (Spearman r=0.3114, p<0.0001) (online supplemental figure 2A,B). In contrast, the CD8+PD-L1⁻ Tils had a strong positive correlation with the overall CD8+ Tils (Spearman r=0.9426, p<0.0001) (online supplemental figure 2C).

According to the quartile values of the overall infiltration level, we classified the samples from cohort A into CD8+PD-L1+ Tils high (infiltration levels within the top 25%) and low (the rest samples) groups. Figure 1B shows representative images of the immune microenvironments of CD8+PD-L1+ Tils high group (case 1) and low group (case 2). The levels of total CD8+ cells and CD68+ (macrophage) cells in the CD8+PD-L1+ Tils high group were significantly higher than those in the CD8+PD-L1+ Tils low group (p<0.0001, p=0.0003) (figure 1C). However, the expression levels of immune checkpoint molecules PD-L1 and PD-1, as well as the level of CD163+ M2 macrophages, were also significantly higher in the environment of the CD8+PD-L1+ Tils high group than the low group (p<0.0001, p=0.0002, p<0.0001) (figure 1D). Based on the classification result of tumor PD-L1 expression by traditional IHC test, further analysis showed that the levels of total CD8+ Tils, CD8+PD-L1+ Tils, and CD8+PD-L1⁻ Tils in the PD-L1 high group were higher than those in the PD-L1 negative group. However, only the level of CD8+PD-L1+ Tils in the PD-L1 low group was significantly higher than that in the PD-L1 negative group, the levels of total CD8+ Tils and CD8+PD-L1⁻ Tils were not significantly different between the two groups (online supplemental figure 3).

A study of pancreatic cancer in mouse models found that the engagement of PD-L1 on CD8+ cells with PD-1 on the neighboring CD8+ cells can lead to a decrease in the antitumor ability of both cell types.14 Our results from cohort A showed that the level of CD8+PD-1+ Tils in the CD8+PD-L1+ Tils high group was significantly higher than that in the low group (p<0.0001) (figure 1E). Meanwhile, the level of CD68+PD-1+ cells in the CD8+PD-L1+ Tils high group was higher than that in the low group (p=0.0150) (figure 1E). CD68+PD-1+ macrophages would convert to the M2 type with impaired antigen producing ability when engaging with the PD-L1 molecule on CD8+PD-L1+ Tils.14 Indeed, we found in cohort A that the level of CD68+CD163+ M2 type macrophages was significantly higher in the CD8+PD-L1+ Tils high group than that in the low group (p<0.0001) (figure 1E).

Previous study reported that the expression of PD-L1 on Tils is modulated by tumor antigens.14 To investigate whether there is an association between the infiltration of CD8+PD-L1+ Tils and tumor antigens in NSCLC tumors, we performed deep sequencing with a panel of 534 tumor-related genes on 123 samples from cohort A and analyzed the tumor mutation spectrum and the total mutation burden. The landscape of tumor mutation spectrum, CD8+PD-L1+ Tils infiltration, and TMB levels was shown in figure 2A.

The CD8+PD-L1+ Tils high group has a significantly higher mutation burden than the low group (p=0.0240) (figure 2B). As shown in figure 2C and D, the levels of TMB and CD8+PD-L1+ Tils in the EGFR mutated group were significantly lower than those in the EGFR wild-type group (p=0.0250, p=0.0143) (figure 2C–D). On the other hand, mutations in DNA polymerase coding genes, DNA polymerase epsilon (POLE) and polymerase δ catalytic subunit gene 1 (POLD1), will cause a large number of genetic variations in tumors, which will in turn lead to the production of a large number of new antigens.22 23 In order to verify whether the neoantigen-related tumor gene mutations are associated with CD8+PD-L1+ Tils infiltration, we analyzed the levels of TMB and CD8+PD-L1+ Tils between POLE or POLD1 mutated group and wild-type group. TMBs of the POLE and POLD1 mutant groups were indeed higher than those of the wild-type group (p=0.0252 and p=0.1627, respectively) (figure 2E), and the level of CD8+PD-L1+ Tils of the POLE mutant group was also higher than that of the wild-type group (p=0.0472) (figure 2F). The difference in the levels of CD8+PD-L1+ Tils between POLD1 mutant and wild-type groups exhibited a similar trend, although not statistically significant (p=0.0776) (figure 2F).

To determine whether the infiltration of CD8+PD-L1+ Tils will affect the prognosis of patients receiving non-immunotherapy, we analyzed the prognostic difference between the CD8+PD-L1+ Tils high and low groups in patients diagnosed at different stages, including stage I–II patients treated with surgical resection, stage IV patients treated with EGFR-TKIs, and stage IV patients treated with chemotherapy in cohort A. Analysis on the relationship between CD8+PD-L1+ Tils infiltration and clinical characteristics revealed that the levels of CD8+PD-L1+ Tils infiltration were higher in patients younger than 61 years old and in stage IV patients (online supplemental figure 4A,C), but it was not associated with gender, smoking history, or tumor pathological type (online supplemental figure 4E). In order to determine whether the EGFR mutation status affected the age-related result, we divided the ≥61 group and the <61 group into four groups. There was no significant difference in the levels of CD8+PD-L1+ Tils between the “≥61, EGFR mut” group and the “≥61, EGFR wt” group (p>0.9999), as well as between the “<61, EGFR mut” group and “<61, EGFR wt” group (p>0.9999) (online supplemental figure 4B). Similar results were found in the groups related to clinical stage (online supplemental figure 4D). In stage I–II patients treated with surgical resection, the PFS time of patients in the CD8+PD-L1+ Tils high group was significantly shorter than that of patients in the low group (HR=2.09, p=0.0198) (figure 3A). Multivariate analysis incorporating clinical characteristics showed that a high level of CD8+PD-L1+ Tils is an independent predictor of a short PFS time (HR=3.24, p=0.005) (figure 3B). However, there was no significant difference in the PFS time between CD8+PD-L1+ Tils high and low groups of stage IV patients receiving either targeted therapy or chemotherapy (p=0.4709 and p=0.4727, respectively) (online supplemental figure 5A,B).

In order to verify our speculation that patients with a high degree of CD8+PD-L1+ Tils infiltration may be more suitable for immunotherapy, we investigated the relationship between the level of CD8+PD-L1+ Tils and the efficacy of anti-PD-1 treatment in patients in cohort B. All patients in this cohort were EGFR wild-type, 25 cases of which were with lung adenocarcinoma and the other 25 cases with lung squamous cell carcinoma. 14 patients received PD-1 inhibitor monotherapy and 36 received a combination treatment with PD-1 inhibitor plus chemotherapy. The efficacy of treatment was evaluated as objective tumor response: 1 case achieved CR (complete response), 21 cases achieved PR (partial response), 14 cases had SD (stable disease), and 14 patients had PD (progressive disease).

Figure 4A shows a landscape of the treatment efficacy and the corresponding gene mutation information and other clinical characteristics of all patients ranked by the infiltration level of CD8+PD-L1+ Tils. The overall level of CD8+PD-L1+ Tils and their proportion in the total CD8+ Tils were higher in the CR/PR group than the SD/PD group (p=0.0337, p=0.0118) (figure 4B). In 20 patients without detectable PD-L1+ TC, there was a similar trend, although the difference was not significant (p=0.1307, p=0.1072) (online supplemental figure 6). In terms of the level of total CD8+ Tils, there was no significant difference between the CR/PR group and the SD/PD group (p=0.1697) (figure 4C). The PFS time of each patient was shown in figure 4A. In general, the PFS time of patients in the CD8+PD-L1+ Tils high group was longer than that of patients in the low group (HR=0.55, p=0.0429) (figure 4D).

From the perspective of the tumor immune microenvironment (TIME) classification based on the levels of PD-L1 expression and CD8+ Tils, the result showed that the difference in ORR rate ((CR+PR) / (CR+PR+SD+PD)) was significant between groups with different levels of CD8+PD-L1+ Tils or groups with different TIME (PD-L1/CD8) type (p=0.0439, p=0.0192) (online supplemental table 3). However, this difference was not significant between groups with different levels of PD-L1 expression or CD8+ Tils (p=0.1030, p=0.5090). Interestingly, the ORR rate of the PD-L1-high and CD8-high group was lower than that of the PD-L1-high and CD8-moderate/low group and the PD-L1-low/negative and CD8-high group (40% vs 67% and 71%, respectively) (online supplemental table 3). In addition, among the five CR/PR patients (cases 1, 23, 33, 34, and 46) in the PD-L1-low/negative and CD8-moderate/low group, two patients (cases 1 and 33) had high level of CD8+PD-L1+ Tils (figure 4A).

The responses were expected to be different in anti-PD-1 plus chemotherapy compared with anti-PD-1 monotherapy. The result showed that the response rates of these two treatment groups were different, but relatively close. The ORR rate of the monotherapy group was 50.0%, and that of the anti-PD-1 plus chemotherapy group was 42%. There was no significant difference in PFS between these two treatment groups (online supplemental figure 7).

Of the 50 patients in cohort B, 25 received genetic mutation analysis. Among them, patient 26 carries POLE and POLD1 mutations, patients 40 and 47 carry POLD1 mutations, and patient 50 carries POLE mutations. The amount of CD8+PD-L1+ Tils in all these four patients’ tumors was at a high level (figure 4A), and the anti-PD-1 treatments achieved good therapeutic effects. The tumor responses to immunotherapy in these four patients were PR (patient 26), PR (patient 40), PR (patient 47), and SD (patient 50), respectively (figure 4A).