Tissue samples were provided by the Lung Biobank Heidelberg, a member of the accredited Tissue Bank of the National Center for Tumor Diseases (NCT) Heidelberg, the Biomaterial Bank Heidelberg, and the Biobank Platform of the German Center for Lung Research (DZL). The local ethics committees of the Medical Faculty Heidelberg and Hannover Medical School [S-270/2001 (biobanking vote) and 9155_BO_K_2020 (study-specific vote)] approved the use of the biomaterial and data. All patients (for cohort overview see Table I) included in the study signed an informed consent, and the study was performed according to the principles set out in the WMA Declaration of Helsinki.

Tumor and matched distant (>5 cm) tumor-free lung tissue samples from NSCLC patients who underwent therapy-naïve resection for primary lung cancer at Thoraxklinic at University Hospital Heidelberg, Germany were collected between 2006 and 2011. Tissues were snap-frozen within 30 min after resection and stored at −80°C until the time of analysis. More detailed information is provided elsewhere (15).

RNA was isolated from tumor tissue and adjacent lung tissue. For RNA isolation from patient tumor tissue, a tumor content of ≥50% was the minimum prerequisite. A total of 10-15 tumor cryosections (10-15 µm) from each patient were sliced, and the first as well as the last section of the series was stained with hematoxylin and eosin (H&E). A lung pathologist determined the proportion of viable tumor cells, stromal cells, healthy lung cells, and necrotic areas. Total RNA was isolated from patient tissue using an AllPrep DNA/RNA/miRNA Universal kit (Qiagen, Inc.) according to the manufacturer's instructions. Afterwards, the quality of total RNA was assessed by utilizing an Agilent 2100 Bioanalyzer and an Agilent RNA 6000 Nano kit (Agilent Technologies). Using the Transcriptor First Strand cDNA Synthesis kit (Roche), total RNA was transcribed to complementary DNA and used for quantitative polymerase chain reaction (qPCR). A complete description of the procedure is provided elsewhere (15).

For gene expression analyses of patient tissues, volumes of 5 µl cDNA (corresponding to 5 ng of isolated total RNA) were utilized for qPCR with the LightCycler480^® (Roche) in 384-well plates according to the Minimum Information for Publication of qPCR Experiments (MIQE) guidelines (16). Universal ProbeLibrary (UPL) assay (Roche) was used as the amplification and detection system. Gene-specific primers (TIB Molbiol were combined with the primaQuant 2X qPCR Probe-MasterMix (Steinbrenner Laborsysteme). Threshold cycle (Ct) values were evaluated with the LightCycler480^® software release 1.5 and the 2nd derivative maximum method (Roche). For the comparison of gene expression in tumor and non-malignant samples, the relative expression of the genes was calculated (ΔCt values). A panel of housekeeper genes was evaluated in tumor and lung samples in regard to stable expression, and TMPRSS2 expression was normalized to the housekeepers esterase D (ESD) and ribosomal protein S18 (RPS18). For the waterfall plots analyzing the fold-change expression changes between tumor and normal tissues, the 2^−ΔΔCt values were calculated. The following primers and UPL were used for the detection of TMPRSS2: TMPRSS2 forward (UPL #71, 5′-ACC AGT GTG TCT GCC CAA C-3′) and TMPRSS2 reverse (UPL #71, 5′-GCG TTC AGC ACT TCT GAG G-3′); ESD forward (UPL#50, 5′-TCA GTC TGC TTC AGA ACA TGG-3′) and ESD reverse (UPL#50 5′-CCT TTA ATA TTG CAG CCA CGA-3′); RPS18 forward (UPL#46, 5′-CTT CCA CAG GAG GCC TAC AC-3′) and RPS18 reverse (UPL#46, 5′-CGC AAA ATA TGC TGG AAC TTT-3′). The complete procedure including housekeeper selection is described elsewhere (15).

Cryo-conserved tumor and corresponding lung tissues from patients were evaluated by a pathologist as described in the 'Total RNA isolation and cDNA synthesis' section and cut in 100-µm pieces with a HM 500 OM microm (Thermo Fisher Scientific, Inc.). Tissues were disrupted using 500 µl PBS (#14190-094, Thermo Fisher Scientific, Inc.) per 100 mg of tissue, Halt™ Protease Inhibitor Cocktail (#78430, Thermo Fisher Scientific, Inc.) and a TissueLyser Mixer-Mill Disruptor (Qiagen) for 2 min at 25 Hz. Afterwards, the cell lysates were centrifuged for 10 min at 13.000 × g and 4°C. The supernatants were transferred to a new tube, and the protein concentration was determined using a BCA protein assay (#23225, Thermo Fisher Scientific, Inc.). A total of 24 randomly selected patients were used as representative examples. Total protein (100 µg) was used for the immunoblot analysis of TMPRSS2 and AAT and heated for 5 min at 95°C in an SDS (sodium dodecyl sulfate)-sample buffer containing ß-mercaptoethanol, glycine and pyronin. Samples were separated using 10% SDS-PAGE and blotted on a nitrocellulose membrane (#GE10600002 Merck KGaA) that was blocked with nonfat dried milk powder (5%, #A0830, AppliChem) in PBS/Tween 20 (0.1%, #A4974, AppliChem) for 1 h at room temperature. For the detection of TMPRSS2, a rabbit monoclonal antibody (#MA5-35756, Thermo Fisher Scientific, Inc.) was applied at a dilution of 1:750 overnight at 4°C. For the detection of AAT, a rabbit polyclonal antibody (#A0012, Dako) was used at a dilution of 1:10,000 overnight at 4°C. A β-actin antibody (#A5441 Merck KGaA) was applied at a dilution of 1:10,000 for 1 h at room temperature as a loading control. A peroxidase system was used for signal development. A secondary anti-mouse IgG antibody (#A4416, Merck KGaA) was applied at a dilution of 1:10,000 for 1 h at room temperature, and a secondary anti-rabbit IgG antibody (#A6154, Merck KGaA) was applied at a dilution of 1:5,000. Signals were visualized with Chemiluminate-HRP Picodetect (#A3417, Applichem). Quantification of the signals was evaluated with Image Studio Lite V. 5.2 (LI-COR Biosciences, GmbH) and adapted to actin levels (Figs. S3-S6).

For detection of TMPRSS2, a monoclonal TMPRSS2 antibody (MABF2158, Merck KGaA) was used. Before tissue microarray (TMA) construction, a H&E-stained slide of each block was analyzed to select tumor-containing regions. A TMA machine (AlphaMetrix Biotech) was used to extract tandem 1.0-mm cylindrical core sample from each tissue donor block (for cohort overview see Table II). Paraffin-embedded tissue sections were deparaffinized with the following steps: 2×10 min in xylol, 2×5 min in 100% ethanol, 1×3 min in 98% ethanol and 1×3 min in 70% ethanol. Antigen retrieval was performed in a steamer with sodium-citrate-buffer (10 mM sodium citrate, 0.05% Tween 20, pH 6.0) for 15 min. Peroxidases were blocked for 10 min at room temperature (RT) using 3% H₂O₂ (Applichem). Slides were incubated with normal goat serum for 1 h at RT to avoid unspecific background staining (Cell Signaling Technology, Inc.). The primary antibody was incubated at a dilution of 1:200 overnight at 4°C in a humid chamber. The staining procedure was performed with SignalStain^® DAB Substrate Kit (#8059, Cell Signaling Technology, Inc.) according to manufacturer's instructions. The last developing step was performed for 2 min. Cell nuclei were stained using Mayer's Hematoxylin Solution (Sigma-Aldrich/Merck KGaA). Slides were mounted using ImmunoHistoMount™ (Sigma-Aldrich/Merck KGaA). Staining was observed with an Olympus IX-71 inverted microscope. Images were captured with an Olympus Color View II digital camera and Olympus Cell-F software (cellSense dimension, V1.11, Olympus). Tiffs were assembled into figures using Photoshop CS6 (Adobe Systems, Inc.). Only changes in brightness and contrast were applied. TMA slides were scanned and analyzed using Aperio ImageScope (v12.4.3.5008, Leica Biosystems). Scoring was performed by multiplication of staining intensity (0-3) with the proportion of positive cells (0-4) (Fig. 4A).

Data of qPCR and IHC analyses were statistically analyzed under REMARK criteria (17) with SPSS 25.0 for Windows (IBM Corp.). The endpoint of the study was overall survival. Overall survival time was calculated from the date of diagnosis until the last date of contact or death. The cut-offs used for survival analyses were selected using the software tool 'Cutoff-Finder' (http://molpath.charite.de/cutoff/index.jsp). Multivariate survival analyses were performed using the Cox proportional hazards model (Table III). Univariate analysis of survival data was performed according to Kaplan and Meier (Figs. 1C-F, 2 and 4). The log-rank test was used to test the significance between the groups. The non-parametric, Wilcoxon matched-pairs signed rank test and the Mann-Whitney test were used to investigate significant differences between the patient groups (Figs. 1A, 4D and S1). The Bonferroni's correction for multiple testing was considered for Fig. 1A. Correlation analyses were performed using the nonparametric Spearman's rank correlation analysis (Figs. 3, 4B, 5C and D and S2). A P-value <0.05 was considered significant. Data were visualized with GraphPad Prism 9 (GraphPad Software, Inc.) and SPSS 25.0 (IBM Corp.).

First, we analyzed TMPRSS2 gene expression in a large NSCLC patient cohort (n=347, Table I). We observed a significantly lower expression of TMPRSS2 in the tumor tissues of patients with SQCC and ADC compared to that noted in the adjacent non-tumor lung tissues (Fig. 1A; Please note that higher values indicate a lower expression). TMPRSS2 was significantly upregulated in pathological stage I ADC compared to stage II, but not in SQCC (Fig. S1; Please note that higher values indicate a lower expression). Comparing tumor and normal lung tissues, we found an approximately 10-fold downregulation of TMPRSS2 in the tumor tissues of patients with SQCC and an approximately 2-fold downregulation in the tumor tissues of patients with ADC (Fig. 1B). In the SQCC group, TMPRSS2 was downregulated in 137/143 patients (~96%, Fig. 1B).

In a further approach, we focused on the influence of TMPRSS2 on patient overall survival. We used the software Cut-off Finder to separate patients into two groups: one with a higher and one with a lower TMPRSS2 expression. A multivariate survival analysis (Table III) including the most important clinical parameters of the entire NSCLC cohort revealed that age (HR=1.036, P=0.001), sex (male vs. female, HR=1.502, P=0.041) pathological tumor stage (HR=1.032, P<0.001) and a low expression of TMPRSS2 (HR=1.536, P=0.033) are independent prognostic indicators of a dismal prognosis (Table III). A more detailed analyses of the two subgroups of NSCLC revealed that age and pathological tumor stage are prognostic indicators for both, SQCC and ADC, patients. However, the lower expression of TMPRSS2 was found as a significant independent prognostic factor only for ADC (HR=2.065, P=0.002) (Table III). A low expression of TMPRSS2 resulted in a more than 2-fold hazard ratio. Kaplan-Meier analyses including all patients (Fig. 1C) confirmed that low TMPRSS2 expression in tumor tissue is prognostic for poor overall survival (P=0.017). In contrast, the expression of TMPRSS2 in adjacent normal lung tissue was not correlated with survival prognosis (Fig. 1D). Remarkably, we found that low TMPRSS2 expression had a prognostic value only for patients harboring an ADC (P=0.006) (Fig. 1F) but not for patients with SQCC (P=0.736 (Fig. 1E).

A recent publication demonstrated an increased expression of TMPRSS2 in lung airway of smokers compared to non-smokers(18). Therefore, we further analyzed TMPRSS2 in relationship to the smoking status (Fig. 2). Indeed, we found that the prognostic effect of TMPRSS2 expression depends on the smoking history. While TMPRSS2 expression had no influence on overall survival in non-smokers (Fig. 2A, P=0.395), a trend was observed for former smokers (Fig. 2B, P=0.172) and a significantly worse prognosis was noted in current smokers (Fig. 2C, P=0.014) with lower TMPRSS2 expression in tumor tissue. Since TMPRSS2 is an androgen-regulated gene (19), we analyzed its relationship to sex. We observed a tendency for male patients (Fig. 2D) and a nearly significant survival benefit in female patients expressing higher levels of the TMPRSS2 gene in the lungs (P=0.059, Fig. 2E).

AAT has been shown to inhibit TMPRSS2 activity (4). Moreover, TMPRSS2 and SERPINA1 are co-expressed in human lungs (GTEx Multi Gene Query. GTEx Portal 2020. https://gtexportal.org/home/multiGeneQueryPage/TMPRSS2,SERPINA1). These facts prompted us to perform correlation analyses between TMPRSS2 and SERPINA1 expression in our NSCLC patient cohort (Fig. 3). Overall, we found a weak correlation (r=0.44) between TMPRSS1 and SERPINA1 expression in NSCLC tumor tissues (Fig. 3A). However, in normal lung tissues, the correlation between these two genes was higher (r=0.56, Fig. 3B).

When we further stratified NSCLC cases into SQCC and ADC, we found that TMPRSS2 and SERPINA1 genes were correlated in SQCC (r=0.51, Fig. 3C) but not in the ADC subgroup (r=0.24, Fig. 3D).

We next performed IHC staining for TMPRSS2 using tissue microarrays (TMA) to include protein expression data in our analyses. In total, 104 patients with ADC and 84 patients with SQCC were evaluated (Fig. 4A). Each patient was evaluated in duplicates for the proportion of positive tumor cells (upper panel) and intensity of staining (lower panel). The total IHC score for each patient was calculated by multiplication of both values. Out of all the evaluated IHC patients, 95 were from the qPCR cohort presented in Fig. 1. As shown in Fig. 4B, protein and RNA levels of TMPRSS2 showed a slight correlation. The same TMAs were evaluated for AAT expression in our previous study (3). While AAT was detected heterogeneously and found in both tumor and stroma (Fig. 4C, left panels), TMPRSS2 protein mainly occurred in tumor cells (Fig. 4C, right panels). While AAT protein levels were similar between ADC and SQCC (Fig. 4D), TMPRSS2 levels were significantly higher in ADC than that noted in SQCC. Furthermore, we found that high protein levels of TMPRSS2 are associated with better patient overall survival (Fig. 4E) and have prognostic value for ADC (P=0.006, Fig. 4G), but not for SQCC (Fig. 4F). This latter finding is in line with our gene expression data (Fig. 1E and F). However, we did not find any correlation between TMPRSS2 and AAT protein levels in the tissue samples analyzed (Fig. S2).

Since AAT has been described as an inhibitor of TMPRSS2 activity (4), we further investigated TMPRSS2 and AAT protein profiles in NSCLC patients. Therefore, we homogenized normal and tumor tissues (tumor content >50%) from 24 randomly selected patients (12 SQCC and 12 ADC). Equal amounts of proteins were separated electrophoretically followed by western blot analyses (Fig. 5A). By using specific anti-TMPRSS2 and anti-AAT antibodies we were able to detect various molecular forms of TMPRSS2 (upper panels) and AAT (lower panel) proteins. Using semi-quantitative densitometric analyses, we compared the protein profiles of TMPRSS2 in tumor and normal samples (Figs. 5B and S3-S6). We observed an increased amount of TMPRSS2 cleaved form in tumor tissues of both SQCC and ADC (with a median increase of 238 and 359%, respectively). The levels of full-length TMPRSS2 and AAT protein were lower in SQCC (73% for TMPRSS2 and AAT) and ADC (87% for TMPRSS2 and 69% for AAT) as compared to normal lung tissues. We further investigated the relationship between TMPRSS2 and AAT protein levels and did not find any correlation in patients with SQCC (Fig. 5C). Interestingly, a slight negative correlation between the two proteins was found in normal lung tissues of patients with SQCC and ADC (Fig. 5D).