Datasets

We downloaded RNA-Seq gene expression profiling (level 3 and RSEM normalized), protein expression profiling, and clinical data for the TCGA-LUAD cohort from the Genomic Data Commons Data Portal (https://portal.gdc.cancer.gov/). We downloaded microarray gene expression profiling (normalized) and clinical data for other four LUAD cohorts (GSE12667 [38], GSE30219 [39], GSE31210 [40], and GSE50081 [41]) from the Gene Expression Omnibus (GEO) (https://www.ncbi.nlm.nih.gov/geo/). Moreover, we downloaded two scRNA-seq data for LUAD, including GSE131907 [12] and Maynard corhort [13]. The proteomic dataset CPTAC-LUAD was downloaded from CPTAC (https://gdc.cancer.gov/about-gdc/contributed-genomic-data-cancer-research/clinical-proteomic-tumor-analysis-consortium-cptac). In addition, we collected 100 blood samples from LUAD patients and 20 blood samples from healthy persons from Jiangsu Cancer Hospital, China. The studies were “approved by Jiangsu Cancer Hospital.” According to the diagnosis and treatment guidelines for non-small cell lung cancer (CSCO 2020), LUAD patients in this study were divided into two groups: 50 patients in early stage (stage I) and 50 patients in late stage (stage III-IV). We log2-transformed the RNA-Seq gene expression values before further analyses. A description of these datasets is shown in Supplementary Table S1.

Patient and public involvement

The study was done in accordance with both the Declaration of Helsinki and the International Conference on Harmonization Good Clinical Practice guidelines and was approved by the institutional review board.

scRNA-seq data pre-processing

We analyzed two LUAD scRNA-seq datasets GSE131907 [12] (10x) and Maynard cohort [13] (smart-seq2). In GSE131907, the gene expression values were the unique molecular identifier (UMI) data which we normalized using the “NormalizeData()” function in the R package “Seurat” (v4.0.6) with the default parameters. That is, the UMI value of each cell was normalized by size-factor 10,000 and then ln(x + 1) transformed. For the Maynard cohort dataset, we used the normalized count values of gene expression.

Gene-set enrichment analysis

We quantified the enrichment levels of immune signatures, pathways, and tumor phenotypes in tumors by the single-sample gene-set enrichment analysis (ssGSEA) [24] of their marker gene sets. The ssGSEA was performed with the R package “GSVA” [24]. The marker gene sets are presented in Supplementary Table S2. We used GSEA [42] to identify KEGG [43] pathways significantly associated with a gene set with a threshold of adjusted p value < 0.05. We used WGCNA [25], an R package, to identify gene modules and their associated GO terms enriched in the high- (upper third) and low-TMPRSS2-expression-level (bottom third) LUADs.

Survival analysis

We compared OS and DFS between the high- (upper third) and low-TMPRSS2-expression-level (bottom third) LUAD patients. Kaplan–Meier curves were utilized to display survival time differences, whose significances were evaluated by the log-rank test. We performed the survival analyses using the R package “survival”. Moreover, we performed multivariate survival analysis using the Cox proportional hazards model to explore the correlation between TMPRSS2 expression and OS prognosis after correcting confounding variables, including TMPRSS2 expression, age, tumor stage, and enrichment levels of immune cells (CD8 + T cells and CD4 + regulatory T cells). The “age”, “CD8 + T cells enrichment”, and “CD4 + regulatory T cells enrichment” were continuous variables, and both “TMPRSS2 expression” (high versus low) and “tumor stage” (early versus late) were binary variables. We implemented the multivariate survival analysis using the function “coxph” in the R package “survival”.

Statistical analysis

We used the Spearman correlation to evaluate associations between TMPRSS2 expression levels and ssGSEA scores of gene sets; the Spearman correlation coefficients (ρ) and p values were reported. In addition, we used the Pearson correlation to evaluate associations between TMPRSS2 expression levels and gene or protein expression levels and the ratios of immune signatures; the Pearson correlation coefficients (r) were reported. The ratios between immune signatures were the log2-transformed values of the ratios between the geometric mean expression levels of all marker genes in immune signatures. In comparisons of TMPRSS2 expression levels between different groups of samples, we used the two-tailed Student’s t test for two groups and the one-way ANOVA test for more than two groups. We performed the statistical analyses using the R programming software (https://cran.r-project.org/).

In vitro experimentsAntibodies, reagents and cell lines

All antibodies were used at a dilution of 1:1000 unless otherwise specified. Anti-PD- L1 (ab213480), anti-CD8 (ab22378), anti-CD49b (ab181548), anti-MSH6 (ab92471), anti-TMPRSS2 (ab109131) and anti-GAPDH (ab181603) were purchased from Abcam (Burlingame, CA). Anti-PD-L1 (66248-1-Ig) and anti-MSH6 (66172-1-Ig) in supplementary materials were purchased from Proteintech Group, Inc.PE anti-mouse TNF-α antibody (12-7321-81), APC anti-mouse IFN-γ antibody (17-7311-81), APC anti-mouse CD279 (PD-1) antibody (12-9985-81), and APC anti-mouse CD223 (LAG-3) antibody (12-2231-81) were purchased from eBioscience (San Diego, CA). The human lung cancer cell lines A549, H1975, and H1299 were from the American Type Culture Collection. They were cultured in 90% F12K (GIBCO, USA) supplemented with 10% fetal bovine serum in a humidified incubator at 37 °C and 5% CO2. NK92 cells (KeyGEN BioTECH, Nanjing, China) were cultured in Alpha MEM (GIBCO, USA) with 2 mM L-glutamine, 1.5 g/L sodium bicarbonate, 0.2 mM inositol, 0.1 mM 2-mercaptoethanol, 0.02 mM folic acid, 100–200 U/mL recombinant human IL-2 (PeproTech, Rocky Hill, New Jersey, USA), and a final concentration of 12.5% horse serum and 12.5% fetal bovine serum.

TMPRSS2 knockdown with small interfering RNA (siRNA)

A549 cells were transfected with TMPRSS2 siRNA or control siRNA by using Effectene Transfection Reagent (Qiagen, Hilden, Germany, B00118) according to the manufacturer’s instructions. The medium was replaced after 24 h incubation with fresh medium, and the cells were maintained for a further 24 h. Quantitative PCR or Western blotting were used to detect the transfection efficiency. TMPRSS2 siRNA and control siRNA were synthesized by KeyGEN Biotech (Nanjing, China). Their sequences were as follows: TMPRSS2 siRNA: 1, 5'- GGAC AUGG GCUA UAAG AAU -3' (sense) and 5'- AUUC UUAU AGCC CAUG UCC-3' (antisense); 2, 5'- ACUC CAAG ACCA AGAA CAA -3' (sense) and 5'- UUGU UCUU GGUC UUGG AGU-3' (antisense); 3,5'-GGAC UGGA UUUA UCGA CAA-3'(sense) and 5'-UUGU CGAU AAAU CCAG UCC-3' (antisense); control siRNA: 5'-UUCU CCGA ACGU GUCA CGU dTdT-3' (sense) and 5'-ACGUGACACGUUCGGAGAAdTdT-3' (antisense).

Lentivirus generation and infection

Lentivirus was prepared according to the manufacturer’s instructions. The heteroduplexes, supplied as 58-nucleotide oligomers, were annealed; the downstream of the U6 promoter was inserted into the pLKO.1 plasmid to generate pLKO.1/ShTMPRSS2. Recombinant and control lentiviruses were produced by transiently transfecting pLKO.1/vector and pLKO.1/ShTMPRSS2, respectively. The lentiviruses were transfected into 293 T cells. After 48 h, lentiviral particles were collected and concentrated from the supernatant by ultracentrifugation. Effective lentiviral shRNA was screened by infecting these viruses with Lewis cells, and their inhibitory effect on TMPRSS2 expression was analyzed by quantitative PCR and Western blotting. The lentivirus containing the ShTMPRSS2 RNA target sequences and a control virus were used for the animal study. The coding strand sequence of the shRNA-encoding oligonucleotides was 5’-ACGGGAACGTGACGGTATTTA-3’ for TMPRSS2.

Western blotting

A549, H1975 and H1299 cell extracts were lysed by using lysis buffer supplemented with protease inhibitor cocktail immediately before use. Total proteins present in the cell lysates were quantified by using the BCA assay. Proteins were denatured by addition of 6 volumes of SDS sample buffer and boiled at 95 °C for 5 min and were then separated by SDS-PAGE. The resolved proteins were transferred onto a nitrocellulose membrane after electrophoresis. The membranes were incubated with 5% skimmed milk in TBS containing 0.1% Tween 20 (TBS-T) for 1 h to block the non-specific binding and then incubated overnight at 4 °C with specific antibodies. After 2 h incubation with the HRP-labeled secondary antibody, proteins were visualized by enhanced chemiluminescence using a G: BOX chemiXR5 digital imaging system (SYNGENE, UK). The band densities were normalized to the background, and the relative optical density ratios were calculated relative to the housekeeping gene GAPDH.

Quantitative PCR

The total RNA was isolated by Trizol (Invitrogen, USA) and was reversely transcribed into cDNA using the RevertAid First Strand cDNA Synthesis Kit (Thermo Fisher, USA). Quantitative PCR was performed with the ABI Step one plus Real-Time PCR (RT-PCR) system (ABI, USA) using One Step TB Green™ PrimeScript™ RT-PCR Kit II (SYBR Green) (RR086B, TaKaRa, JAPAN). Relative copy number was determined by calculating the fold-change difference in the gene of interest relative to GAPTH. The program for amplification was one cycle of 95 °C for 5 min, followed by 40 cycles of 95 °C for 15 s, 60 °C for 20 s, and 72 °C for 40 s. The relative amount of each gene was normalized to the amount of GAPDH. The primer sequences were as follows: hTMPRSS2: 5'-AACT TCAT CCTT CAGG TGTA-3' (forward) and 5'-TCTC GTTC CAGT CGTCTT-3' (reverse); hGAPDH: 5'- AGAT CATC AGCA ATGC CTCCT-3' (forward) and 5'-ACAC CATG TATT CCGG GTCAAT-3' (reverse).

Cell proliferation assay

A549, H1975 and H1299 cells were plated in 96-well plates at 3 × 104 cells per well and maintained in a medium containing 10% FBS. After 24 h, cell proliferation was determined using the Cell Counting Kit-8 (CCK-8; KeyGEN Biotech, China) following the manufacturer’s instructions. To perform the CCK-8 assay, 10 µl CCK-8 reagent was added to each well and the 96 plates were incubated at 37 °C for 2 h. The optical density was read at 450 nm using a microplate reader. All these experiments were performed in triplicates.

Transwell migration and invasion assays

Cell migratory and invasive abilities were assessed using 24 well transwell chambers (Corning, USA) with membrane pore size of 8.0 µm. A549, H1975 and H1299 cells were seeded into the upper chamber without matrigel at 1 × 105 cells in serum-free medium, while 500 µl medium containing 20% FBS was added to the lower chamber. The chambers were incubated at 37 °C and 5% CO2 for 24 h. The cells on the upper chamber were scraped off with cotton-tipped swabs, and cells that had migrated through the membrane were stained with 0.1% crystal violet at 37 °C for 30 min. The migrated cells were counted at 200x magnification under the microscope using three randomly selected visual fields. All these experiments were performed in triplicates.

Co-culture of tumor cells with NK92 cells

A transwell chamber (Corning, USA) was inserted into a six well plate to construct a co-culture system. A549 cells were seeded on the six well plate at a density of 5 × 104 cells/well, and NK92 cells were seeded on the membrane (polyethylene terephthalate, pore size of 0.4 µm) of the transwell chamber at a density of 5 × 104 cells/chamber. Tumor cells and NK92 cells were co-cultured in a humidified incubator at 37 °C and 5% CO2 atmosphere for 48 h.

EdU proliferation assay

After co-culture of A549 cells with NK92 cells for 48 h, we measured the proliferation capacity of NK92 cells by an EdU (5- ethynyl-2'-deoxyuridine; Invi-trogen, California, USA) proliferation assay. NK92 cells were plated in 96-well plates with a density of 2 × 103 cells/well with 10 µM EdU at 37 °C for 24 h. The cell nuclei were stained with 4',6- diamidino-2-phenylindole (DAPI) at a concentration of 1 µg/mL for 20 min. The proportion of NK92 cells incorporating EdU was detected with fluorescence microscopy. All the experiments were performed in triplicates.

In vivo experimentsIn vivo mouse models

Lewis tumor cells were transduced with ShCon (scramble) or ShTMPRSS2 lentivirus and selected by puromycin for 7 days. The stably transfected Lewis tumor cells (1 × 107/ml) were subcutaneously injected into the right armpit of recipient mice after shaving the injection site. After 5 days, when the tumor volume was approximately 4–5 mm3, the mice were randomly divided into six groups, with half of the ShCon and ShTMPRSS2 mice treated with 150 U/L PD1/PDL1 inhibitor BMS-1 (concentration 500 mg/mL; i.p.) (MCE Cat. No. HY-19991) every 3 days. The tumors were isolated from mice after 15 days. Tumor volumes did not exceed the maximum allowable size according to the LJI IACUC animal experimental protocol. The tumor volume was measured every 3 days after the tumor appeared on the fifth day and was calculated as follows: V = 1/2 × width2 × length. The studies were “approved by Nanjing Medical University.”

Isolation of TILs

After the tumor tissues were separated aseptically and rinsed with cold PBS for 3 times, they were excised and chopped with tweezers and scissors and were then digested with 2 mg/mL collagenase (type IV, sigma V900893) for 45 min, until no tissue mass was visible. Following digestion, lymphocytes were separated with lymphocyte separation medium, washed with PBS, and counted. The specific protocol was as follows: tumors were filtered through 70 µM cell strainers, and the cell suspension was washed twice in culture medium by centrifugation at 1500 rpm and 4 °C for 10 min. After the washing, the cells were resuspended with PBS and were layered over 3 mL of 30%-100% gradient percoll (Beijing Solarbio Science & Technology, Beijing, China); this was followed by centrifugation at 2600 rpm for 25 min at 25 °C. The enriched TILs were obtained at the interface as a thin buffy layer, were washed with PBS three times, and finally were resuspended in FACS staining buffer for further staining procedures.

Flow cytometry

TILs were stained with CD8 (eBioscience, 11-0081-81), CD49b (eBioscience, 11-5971-81), PD-1 (eBioscience, 12-9985-81), and LAG3 (eBioscience, 12-2231-81) and were analyzed by flow cytometry. TILs were restimulated with cell stimulation cocktail (eBioscience, San Diego, California, USA), and the expression of IFN-γ and TNF-α (Biolegend) was analyzed by flow cytometry. Staining for cell surface markers was performed by incubating cells with antibody (1:100 dilution) in FACS buffer (0.1% BSA in PBS) for 30 min at 4 °C. Surface markers of intracellular cytokines (IFN-γ (eBioscience, 17-7311-81) and TNF-α (eBioscience, 12-7321-81)) were stained before fixation/permeabi-lization (Intracellular Fixation & Permeabilization Buffer Set, ThermoFisher).

Immunofluorescence of CD8, CD49b and PD-L1

Paraffin-embedded mice tumor tissue section (3 µm thick) were subjected to immunofluorescence with CD8 (Abcam, ab22378), CD49b (Abcam, ab181548), or PD-L1 (Abcam, ab2134808) primary antibodies. Before immunostaining, tumor tissue sections were deparaffinized with xylene, rehydrated and unmasked in sodium citrate buffer (10 mM, pH 6.0), and treated with a glycine solution (2 mg/mL) to quench autofluorescence. After antigen retrieval, 3% H2O2-methanol solution blocking inactivated enzymes, and goat serum blocking, tissue slides were incubated in wet box for 2 h at 37 °C with anti-CD8, CD49b, or anti-PD-L1 rabbit primary antibodies (1:100 dilution) in blocking solution, and were then dropped with FITC (1:100 dilution) secondary antibody 50-100ul and incubated at 37° for 1 h in the dark. The immunolabeled slides were examined with a fluorescence microscope after nuclear counterstaining with DAPI. Green, red and blue channel fluorescence images were acquired with a Leica DFC310 FX 1.4-megapixel digital color camera equipped with LAS V.3.8 software (Leica Microsystems, Wetzlar, Germany). Overlay images were reconstructed by using the free-share ImageJ software.