The study population comprised 633 consecutive stage I to IIIB NSCLC patients who underwent curative-intent radical resection at the University Hospital of North Norway or the Nordland Central Hospital between 1990 and 2010. Of these, 80 patients were excluded due to neoadjuvant radio-chemotherapy, other malignancy within 5 years before NSCLC diagnosis, inadequate tissue in paraffin-embedded formalin-fixed blocks, or poor tissue quality. Consequently, 553 patients were available for analysis. All patients were restaged according to the 8th edition of the Union for International Cancer Control (UICC) TNM classification system (4). The median follow-up of survivors was 86 months (range 34–267 months). Follow-up data were last updated October 1, 2013. Detailed information regarding the study population has been previously published (28).

Formalin-fixed and paraffin-embedded (FFPE) tumor and control specimens were obtained from the hospital archives. An experienced pathologist reviewed the hematoxylin and eosin (H&E)-stained slides of the face of all FFPE blocks and marked representative tumor and stromal areas. A tissue-arraying instrument (Beecher Instruments, Silver Spring, MD, USA) was used to make tissue microarrays (TMAs). Four replicate tissue cores, each 0.6 mm, were transferred from each donor block to a recipient block. The procedure in detail has been previously reported (29). Sections were cut at 4 µm using Microm microtome HM355S for H&E, immunohistochemistry (IHC), or in situ hybridization (ISH) staining.

ISH on FFPE tissue was performed on the Ventana Discovery Ultra platform for IHC and ISH (Basel, Switzerland) based on the “DAKO” protocol by Jorgensen et al. (30). The same method has been used in previous publications by our group (31, 32). Buffers and detection reagents were purchased from Roche^® (Basel, Switzerland).

Protocol in short: Slides were baked at 60°C overnight and then transferred to the Discovery Ultra for ISH staining. Sections were deparaffinized at 68°C for three cycles in Ventana EZ buffer. Heat retrieval was performed at 95°C with Discovery Cell Conditioning Solution (CC1) for 40 minutes to make access for the probes. Optimized concentrations of probe controls and target miR probes were manually applied (miR-20a-5p, 50 nM; and miR-17-5p, 20 nM). The hybridization reaction was carried out for 60 minutes at 54°C for miR-17-5p and 40°C for miR-20a-5p, followed by two stringency washes with 2.0X SSC buffer. Possible unspecific bindings were blocked with AB blocking solution for 16 minutes. Alkaline phosphatase-conjugated anti-DIG (Anti-DIG-AP) was incubated for 20 minutes for immunologic detection. Substrate enzymatic reactions were carried out using NBT/BCIP for 60–120 minutes to give a blue precipitate. The slides were counterstained with Nuclear Fast Red for contrast staining. Slides were then dehydrated through an increasing gradient of ethanol solutions to xylene and then mounted using a Histokitt mounting medium.

The good sensitivity level of the ISH method and minimal RNA degradation in tissue was confirmed using U6, snRNA control probe at a concentration of 1.5 nM. A scramble miR negative control probe at 10 nM indicated no unspecific staining from reagents or tissues. MicroRNA expression in tissues other than Non-Small cell lung cancer (NSCLC) was also confirmed using a multi-tissue TMA control. Optimizations of temperatures, time, and concentrations were conducted for each probe and reagent.

The TMA slides were digitized using a Panoramic 250 Flash III slide scanner (3DHistech, Budapest, Hungary) and processed using QuPath version 0.1.2 (33). Briefly, 1) each slide was preprocessed according to Bankhead et al.; 2) tissue cores were assigned to the corresponding patient ID; 3) the tissue within each TMA core was identified and tiled into 20 × 20-µm tiles; 4) a classifier for each miR, separating tumor, and others (stroma, necrosis, etc.) was constructed; 5) the classifier was applied to all TMA cores; and 6) the median marker intensity in the tumor compartments for each patient was calculated and exported for further analyses.

The functional properties of miR-17-5p or miR-20a-5p were evaluated in two different human epithelial lung cancer cell lines: the squamous cell carcinoma H520 (ATCC^® HTB-182) (34) and the adenocarcinoma A549 (ATCC^® CCL-185) (35). Cell lines were authenticated at the Forensic Center at the UiT The Arctic University of Norway using Short tandem repeat (STR)-based DNA typing (PowerPlex 16HS kit from Promega, Promega, Madison, USA). For all experiments, H520 and A549 were below passages 60 and 20, respectively.

For all experiments, cells were transiently transfected with either 10 µM hsa-miR-17-5p Pre-miR™ miRNA Precursor (catalog# PM12412; Thermo Fisher Scientific, IL, USA) or hsa-miR-20a-5p Pre-miR™ miRNA Precursor (catalog# AM17100, Thermo Fisher Scientific, USA).

To assess the transfection efficacy, miRs were transfected alongside the Cy3™ Dye-Labeled Pre-miR Negative Control #1 (catalog# AM17120, Thermo Fisher Scientific, USA) using the transfection reagent Lipofectamine™ RNAiMAX (catalog# 13778075, Thermo Fisher Scientific, USA). Transfected Cy3™ Dye-Labeled Pre-miR Negative Control emits fluorescent light when exposed to UV light, and using a fluorescence microscope, the transfection efficiency of our transfections was evaluated to be between 80% and 95% (31).

For the colorimetric proliferation assay, 5 × 10³ cells/well were cultured in 96-well plates. At different time points, cells were incubated with 12 mM of [3-(4,5-dimethylthiazol-2-yl)-2,5-diphenyltetrazolium bromide] (MTT; 5 mg/mL) (catalog# M6494, Invitrogen, OR, USA) for 4 h at 37°C. The formazan crystals produced were solubilized by the addition of 0.01 M HCl/sodium dodecyl sulfate (SDS) (catalog# 28312; Thermo Fisher Scientific, IL, USA) and mixed thoroughly using the pipette. The cells were incubated at 37°C overnight to dissolve the formazan, and the absorbance was measured in the CLARIOstar^® plate reader (BMG Labtech, Ortenberg, Germany) at 570 nm.

The detailed methodology of the invasion assay has been previously reported (36). In brief, cells were seeded in ThinceRt chambers (Greiner Bio-One, Kremsmünster, Austria) and transfected for 48 h at 37°C. The membranes containing invading cells (under the surface of the membrane) were photographed using an inverted optical microscope. Three random microscope fields were selected for analysis. The area of invading cells was determined using the ImageJ software, and results were plotted (mean ± SEM of three independent experiments) in relation to the control (C = 1).

Detailed methodology has been previously reported (36). Cells were incubated in a culture medium containing mitomycin C (10 µg/L) to avoid cell proliferation, “wounded” using a 200-µL sterile pipette tip, and then washed to remove detached cells and debris. After 4 h, the cells were transfected for 24 h at 37°C. Photographs of the same area of the wound were taken at 0 and 24 h. Images were captured using an inverted optical microscope and analyzed using the Micrometrics SE Premium 4 software. The areas that were occupied by migrating cells after 24 h of incubation (control and transfected cells) were calculated by subtracting the background levels at 0 h. Results were plotted (mean ± SEM) in relation to control (C = 1). Images are representative of three different experiments.

SPSS 28.0 (Chicago, IL, USA) was used for all statistical analyses. The mean miR expression across TMA cores was normalized to values between 0 and 100 and divided into high and low expression using the mean score as a cut-off. Correlations were analyzed using Spearman’s rank coefficient. For survival analyses, the primary endpoint used was disease-specific survival (DSS), defined as the time from the date of surgery to lung cancer death. Univariate survival curves were estimated using the Kaplan–Meier method, and the statistical significance between survival curves was assessed using the log-rank test. Curves were terminated when less than 10% of the patients were at risk. Clinicopathological variables associated with significant prognostic values from the univariate analyses and the miRs were entered in the multivariate analyses. For multivariate analyses, the backward conditional Cox regression analysis was used. Probability for stepwise entry and step removal was set to 0.05 and 0.10, respectively. p-Values < 0.05 were considered statistically significant for all analyses.

Clinicopathological and demographic variables and their impact on DSS are presented in Table 1 . Age at diagnosis ranged from 28 to 85 years, with a median of 67. Of the patients, 68% were male, and 96% of the patients were current or previous smokers.

MiR-17-5p and miR-20a-5p were both predominantly expressed by tumor epithelial cells and in immune cell aggregates. When present, staining was found in both the nucleus and cytoplasm ( Figure 1 ). Out of the 553 patients available for analysis, some were excluded based on damaged or missing TMA cores, poor staining or tissue quality, or no/few viable tumor cells identified (reviewed by pathologist LTB). After curation, a total of 460 and 470 patients were available for miR-17-5p and miR-20a-5p scoring, respectively. Representative QuPath exports of single cores at higher magnification are provided in the Supplementary Material .

The classifier for miR-17-5p was trained on a total of 13,332 tiles (4,428 and 8,904 defined as tumor and “other”, respectively), while the classifier for miR-20a-5p was trained on 46,181 tiles (19,382 and 26,799 defined as tumor and “other”, respectively). Both classifiers demonstrated an estimated precision of 98% for separating tumors from other tissues, using 20% of the available tiles as a test set. Manual evaluation revealed that, in some cases, the classifier was inaccurate, especially when differentiating between tumor cells and saturated miR-stained immune cells and when tumor cores were faintly stained. Examples of tumor tissue detected by the QuPath classifiers are shown in Figure 1 . The mean and median of the normalized values of miR-17-5p and miR-20a-5p expression in the tumor compartment are found in Supplementary Figure S1 .

The correlations between miR expression and clinicopathological variables were weak (r < 0.250) or non-significant. Correlations between the miRs and other tested biomarkers in our cohort were weak, with the exception of positive correlations with Her3 (miR-17-5p r = 0.348 and miR-20a-5p r = 0.286), MCT1 and MCT2 (miR-20a-5p r = 0.255 and 0.258), MMP9 (miR-17-5p r = 0.292), VEGFR1 (miR-17-5p r = 0.250), and CTLA4 (miR-17-5p r = 0.264).

Due to the strong prognostic impact of lymphocytes in our cohort, we specifically investigated the association between the miRs and immune cells with emphasis on lung squamous cell carcinoma (LUSC) patients. However, we did not observe any significant correlations between the investigated miRs and immune cell markers previously investigated in the cohort.

Univariate analyses of the dichotomized miR-17-5p and miR-20a-5p expression in the overall cohort and in selected subgroups are presented in Table 1 and Figures 2 and 3 . No significant prognostic impact according to either miR expression or DSS was observed in this population. In subgroup analyses, high expression of miR-20a-5p was significantly associated with increased DSS in patients with lymph node metastases (N+, p = 0.023) and lung squamous cell carcinoma histology (p = 0.022). For miR-17-5p, a similar trend was observed in N+ patients (p = 0.078), while high expression was significantly associated with reduced DSS in patients without lymph node metastases (N0, p = 0.050).

Multivariate analyses for miR-17-5p and miR-20a-5p are presented in Tables 2 and 3 , respectively. In the multivariate analysis of the overall cohort, high expression of miR-20a-5p was an independent predictor of increased DSS (HR 0.73, 95% CI 0.54–0.99, p = 0.040). In subgroup analyses, high expression of miR-20a-5p was an independent predictor of increased DSS in N+ (HR 0.58, 95% CI 0.38–0.89, p = 0.014) and LUSC (HR 0.49, 95% CI 0.32–0.76, p = 0.002) patients.

MiR-17-5p was not associated with DSS in the overall cohort ( Table 3 , Model 1). In the subgroup analysis of N0 patients, high expression of miR-17-5p was an independent predictor of reduced DSS (HR 1.55, 95% CI 1.03–2.31, p = 0.035).

The H520 and A549 cell lines were assessed for proliferation, invasion, and migration using MTT, transwell, and wound healing assays, respectively ( Figures 4 – 6 ). Overexpression of miR-20a-5p led to reduced proliferation in the A549, but not in the H520 cell line ( Figure 4 ), increased invasion in both cell lines ( Figure 5 ), and increased migration in the A549 cell line ( Figure 6 ). Overexpression of miR-17-5p did not lead to changes in proliferation in either cell line but induced increased invasion ( Figure 5 ) and migration ( Figure 6 ) in both cell lines.