Cell lines were established as previously published (8). Briefly, small pieces of the resection specimen, not required for diagnostic analyses, were received fresh from surgery. Single-cell suspensions were prepared by mechanic dissection of the tissue. The suspension was passed through a 100-µm cell strainer and washed with PBS. The cell pellet was resuspended in culture medium as described (8) and seeded in collagen-coated six-well plates. Continually growing cells were passaged and stocked regularly. Cells were routinely checked for absence of mycoplasma contamination.

The names of the cell lines consist of the following information: pseudonymized patient ID containing information on place of material collection (HRO = Hanse City of Rostock) and tumor entity/organ of origin (Lu = lung tumor, BML = brain metastasis of a primary lung tumor). Passage numbers are given for all experiments.

All processes involving patients and patient-derived material were approved by the ethics committee of the University Medical Center Rostock (UMR): A 2019-0187. General guidelines for working with patient material were followed; this included obtaining written consent for each patient in advance.

Cells were cultured in standard cell culture flasks using DMEM/F12 medium supplemented with 10% fetal calf serum (FCS) and 2 mM L-glutamine in a humidified CO₂ incubator at 37°C. All cell culture media and reagents were purchased from PAN (PAN-Biotech GmbH, Aidenbach, Germany), and all culture plates and flasks were from Sarstedt (Sarstedt AG & Co. KG, Nümbrecht, Germany).

Cells (2–5 × 10⁴ cells per well) were seeded in a 24-well plate and incubated for 24 h to allow attachment. One column of the 24-well plate was washed with PBS and stained with crystal violet solution every 24 h for 7 consecutive days resulting in quadruplicates for each time point. On the last day, plates were washed three times with PBS and left to dry at room temperature. After complete drying, 100 µl/well 1% sodium dodecyl sulfate solution was added, and plates were placed on a shaker for 10 min to dissolve the crystal violet. Absorbance measurements at 590 nm were performed using a Tecan Infinite 200 Pro (Tecan Group AG, Männedorf, Switzerland) plate reader. Measurements were normalized to the first time point measurement, and doubling time was calculated with GraphPad Prism 9.4.1 using the exponential growth equation for nonlinear regression.

Cells were harvested, washed, and resuspended in PBS, resulting in a concentration of 2 × 10⁵ cells/100 µl. The antibodies (1 µg per antibody and tube containing 100 µl of cell suspension) were added and incubated at 4°C for 30 min in the dark (for detailed information on antibodies, see Table 1 ). For measurement of stainings with anti-EGFR and -PD-L1 antibodies, cells were incubated with 1 µg cetuximab and durvalumab, respectively. For fluorescent detection, the primary antibodies were stained using an FITC-labeled anti-human IgG secondary antibody. Following antibody incubation, cells were washed three times with PBS and resuspended in 200 µl of PBS for measurement with a BD FACS Calibur (Becton, Dickinson & Co., Franklin Lakes, USA) device. Data analysis was done using FCSalyzer 0.9.22-alpha software.

Cells were seeded on a 96-well plate (1–2 × 10⁴ cells per well) in 150 µl/well standard medium and incubated for 24 h to facilitate attachment. The chemotherapeutics were added to the wells in 50 µl of standard medium yielding the desired final concentrations. For every single agent tested, detailed information can be found in Supplementary Data 1 ( Supplementary Table 1 ). After a 72-h incubation period, a second treatment cycle was initiated by the removal of old medium and the addition of new medium and chemotherapeutics. Plates were incubated for an additional 72 h. After a total of 144 h of treatment, cells were washed with PBS and stained with 50 µl/well crystal violet solution for 20 min. The crystal violet solution was then removed, and plates were washed three times with PBS. After complete drying at room temperature, 100 µl/well 1% sodium dodecyl sulfate solution was added, and plates were placed on a shaker for at least 10 min to dissolve the crystal violet. Absorbance measurements at 590 nm were performed using a Tecan Infinite 200 Pro plate reader and normalized viability was calculated in relation to untreated control samples using the following formula:

Same as for the single-agent testing, cells were seeded in a 96-well plate in standard medium. One row was used as blank control (cells incubated with PBS instead of standard medium, thus leading to cell death by starvation) and one row was used as living control (≙ untreated cells), leaving two 6 × 6 dose response matrices for drug sensitivity testing. We tested drug combinations that are typically used in clinical practice, by combining a platinum-based drug (cisplatin and carboplatin) with etoposide, vinorelbine, and paclitaxel, resulting in six paired combinations (for detailed information, please refer to Supplementary Data 2 ( Supplementary Table 2 ). Each dose–response pair experiment was repeated at least three times. Chemotherapy was performed as described above. Data obtained from the spectrometer were annotated, and normalized viability was calculated using in-house software. Synergy was calculated using the bayesynergy R package (16) and RStudio. This uses a probabilistic approach based on the Bliss independence model. The bayesynergy R package essentially describes drug interaction by comparing the zero-interaction model to the observed data using differences in normalized volume under the surface [VUS(Δ)], resulting in measures that can be directly used as percentage points of efficacy gained or lost. Values are calculated separately for antagonism VUS(Δ+) and synergism VUS(Δ−). The final synergy score is calculated by dividing VUS(Δ−) by its standard deviation (see formula below).

Single-cell suspension was seeded at 100 cells/well in quadruplicate in a 24-well plate in 2 ml of standard medium. Outgrowing colonies were photo documented using a Primo Vert microscope and Axiocam USB camera (Zeiss, Jena, Germany).

After preparing a single-cell suspension, cells were seeded at 1,000 cells/well in duplicate in a cell-repellant six-well plate (Greiner AG Kremsmünster, Austria) in 5 ml of spheroid medium (CLS, Eppelheim, Germany) coated with 0.4% base agar and 0.35% top agar. Formation of spheres was checked daily and the final outcome (outgrowth or no outgrowth) was scored as positive or negative at day 14.

Cells were incubated in FCS-free standard medium for at least 24 h and then cells were harvested and a cell suspension of 1 × 10³ cells/µl was prepared in FCS-free standard medium. A 24-well plate with TC inserts (Greiner) was prepared and 500 µl of the cell suspension (≙ 5 × 10⁵ cells) was added to each insert. For the invasion assay, inserts were coated with Matrigel^® Basement Membrane Matrix (Corning Inc., Corning, New York, USA) prior to the addition of the cell suspension. Then, 750 µl of 10% FCS containing medium was added to the lower wells. After incubation for 72 h at 37°C and 5% CO₂, the medium was removed, and inserts were washed twice with PBS. Finally, 1 ml of 0.2% crystal violet solution was added to each insert for cell staining. After 15 min of incubation, removal of the staining solution, and three washing steps with PBS, the non-invasive cells on the upper side of the insert were scraped off with a cotton swab. The addition of 0.75 ml of 1% SDS solution and 15 min incubation on a shaker led to solvation of crystal violet. The inserts were removed, the plate was placed in the Tecan reader, and absorbance was measured (measurement: 570 nm; reference: 620 nm). Measurements were normalized to the HROC24 cell line (17). For data analysis, again, the GraphPad Prism software was used.

Cells were seeded on six-well plates and allowed to grow until full confluency was reached. After starving cells for 24 h with FCS-free standard medium, a scratch was performed using a standard 200-µl pipette tip. Photos were taken with the Primo Vert microscope and Axiocam USB camera at different time periods to account for the different wound closure rates. For each cell line, the assay was done at least three times and the mean wound closure rate and standard deviation was calculated. Photos were analyzed using ImageJ 1.53 and wound healing size tool (18). Wound closure rate in µm/h was calculated in Microsoft Excel by finding the beginning ( t 0 )   and end point ( t 1 ) of the linear migration phase, calculating the difference in wound width during linear migration and dividing the difference by time in hours.

For each cell line, the assay was done at least three times and the mean wound closure rate and standard deviation were calculated.

Per cell line, 5 × 10⁶ cells were harvested by scraping, resuspended in PBS, and embedded in paraffin, and 4-µm sections were stained by applying the same SOP for diagnostic immunohistochemistry assessments. The following reagents/antibodies were used for staining: hematoxylin and eosin for H&E, clone 22C3 for PD-L1, polyclonal AE1/3 for pan-cytokeratin, and clone Ber-EP4 for Ep-CAM.

Three patients operated on at the UMR in the years 2009–2020 and enrolled in the BioBank Rostock (BBR) presented with either primary lung tumors or brain metastases (see Table 2 ). Patient HROBML01 was a male patient who presented with a metastatic lung tumor in the brain at age 67 and had no smoking habit. Patient HROLu22 was a female patient who presented with a primary tumor of the lung at age 64 and also did not smoke. Patient HROLu55 was a male patient who presented with a primary lung tumor at age 48 and was, at least at times, a heavy smoker.

Routine diagnostic procedures included histological assessment of the tumor tissue, determination of the subentity, TNM classification, and grading. The primary lung tumors were further analyzed for TTF1 and PD-L1 protein expression and an NGS Illumina focus panel assessment was performed (see Table 3 , and for detailed information on the Illumina focus panel, see Supplementary Data 3 ). The compiled data for the tumor of patient HROLu22 reveal a primary adenocarcinoma of the lung grade 3 with no nodal or distant metastases. Tumor cells stained positive for both TTF1 and PD-L1. The detailed molecular profiling revealed a complex mutation in the EGF receptor (EGFR). Assessments of patient HROLu55’s tumor revealed a rare pleomorphic cell tumor of the lung. The tumor was graded as G3/4 and distant lymph nodes had already been infiltrated with tumor cells. While the majority of tumor cells stained positive for PD-L1, no TTF1 protein could be detected. Tumor HROBML01 was classified as a brain metastasis of a primary squamous cell carcinoma of the lung. The origin of derived PDC models from the NSCLC patients was ensured by confirming matching STR profiles of the patients with respective PDCs ( Table 4 ).

The cells of HROBML01 and HROLu22 grow as tumor islands, and the cells possess a morphology frequently associated with epithelial tumor (i.e., cobblestone-like) cells. The cells of HROLu55 start of as large(r) cells, which reduce their size according to the space available and finally form a 100% confluent monolayer consisting of small cells in the end. Micro-photographic images of all three cell lines can be found in Figure 1 .

Cell doubling times were between 3 and 5 days. The highest doubling time with 69.31 h (ranging from 57.60 h to 87.00 h; 95% CI) was calculated for HROBML01, closely followed by HROLu55 with 72.06 h (61.85 h to 86.32 h; 95% CI). The lowest proliferation was observed for HROLu22 with a doubling time surpassing 100 h: 101.6 h (91.77 h to 113.7 h; 95% CI).

Paraffin-embedded immunohistochemical assessments performed by an expert pathologist (FP) of the three cell lines (see Figure 2 ) confirmed in H&E overview staining that the cell line HROBML01 is most likely derived from a squamous cell type, HROLu22 from an adenocarcinoma, and HROLu55 from a pleomorphic carcinoma. Additionally, the embedded cells of HROLu55 stained positive for pan-cytokeratin and Ep-CAM as well as PD-L1, which further supports the pleomorphic subentity of the cell line HROLu55 as well.

After confirmation that the cell lines represent the subentities of the tumors they were derived from, the expression of several surface markers was assessed by flow cytometry ( Figure 3 ). All three PDCs expressed HLA I and lacked HLA II expression (data not shown). Because high levels of CD326 (Ep-CAM) were considered as a marker for the epithelial origin of cells, the absence of CD90 as a marker for fibroblast cells, and expression of proteins as a marker for lung tumors, LYPD3, DSG3, and CCD59 (39) were analyzed. All cells expressed CD326, were negative for CD90, and had varying degrees of the lung tumor markers. Additionally, the presence of the immune checkpoint protein PD-L1 and the growth receptor EGFR was analyzed. The cell lines derived from the primary lung tumors were strongly positive for PD-L1 and EGFR. The brain metastasis cell line HROBML01 only showed weak staining for both markers.

In addition to the focus panel sequencing, which was performed during routine pathological assessment for the tumors of patients HROLu22 and HROLu55, we performed whole exome sequencing (WES) analyses, comparing the cell lines with the original patient tumors for all three models. For patients HROLu22 and HROLu55, tumor adjacent normal tissue was assessed in parallel. Due to the fact that HROBML01 was derived from a brain metastasis, no normal brain tissue was available.

The mutational signature analysis of HROLu55 shows high similarity to cosmic signature 4, which is consistent with the mutational pattern caused by exposure to tobacco smoke ( Figure 4 ). No such clear association with a mutational pattern could be observed for either HROLu22 or HROBML01 (see Figure 4 ). However, HROLu22 shows similarities to signatures 1, 5, and 16. Signature 1 is believed to be related to an endogenous mutational process, signature 5 is not yet associated with an underlying mechanism but is often observed in lung adenocarcinomas (40), and signature 16 does not have any described underlying mechanism so far. HROBML01 shows again the highest similarity to signature 5.

In a second step, we identified the top 30 mutated genes by the WES approach across all three tumor tissues and cell lines (see Figure 5 ). Among these, common general cancer mutations in the genes TP53 and MUC16 (41) and lung cancer-associated mutations in the genes MXRA5 (42, 43) and MUC19 (44) could be observed. The genes MXRA5 and MUC16 were mutated in all three tumors. However, the mutation in MXRA5 could not be detected in the cell line HROBML01 and the MUC16 mutation was not detected in the cell line HROLu22. Mutations in the gene TP53 could be observed for both tumors of patients HROLu55 and HROBML01 as well as their cell line counterparts.

Finally, we calculated the TMB for all cancer samples (tissues and cell lines). The tumor tissue always had a higher TMB than the corresponding cell line (see Table 5 ). HROBML01 presented with the highest TMB for both tumor tissue and cell line. In total, all three tumors and cell lines have a comparably high TMB (see Figure 6 ).

In addition to the WES analyses, matched RNA expression levels were assessed by RNA sequencing of the same tissue piece or cell pellet, respectively. Principal component analysis (PCA) for quality control revealed clustering of the normal tissue, the tumor tissue of HROLu22 and HROLu55 with their cell line counterparts, and a third cluster of tumor tissue and cell line for HROBML01 (see Supplementary Data 4 consisting of Supplementary Figures 2 , 3 ). Further “sample-to-sample distance” calculations revealed that the normal tissues of HROLu22 and HROLu55 are (very) close (see Supplementary Data 4 ). This confirms the QC assessment of the PCA.

Thus, the 20 most up- and downregulated genes on RNA expression level were identified ( Figure 7 ). The difference in expression level of normal to tumor tissue and cell lines is at least log2. Among the most overexpressed genes in tumor cells compared to normal tissue are genes coding for transcription factors (HOXB9, SIM2, ZIC5, SP8, TFAP2A, FOXE1, HOXB13, and SALL4), cancer testis antigens (CT83), and cytokines activating regulatory Th17 cells (IL23A). The most downregulated genes on the RNA level code for long non-coding RNA (lncRNA; LANCL1-AS1, LINC00670, BANCR, and LOC100652999) and proteins involved in the regulation of angiogenesis (ANGPT4), signaling cascades (PLA2G1B and RS1), and immune response (SFTPD).

A common feature of cancer cells that serves as a unit to measure the degree of aggressiveness is invasion and migration. Thus, the invasion and migration potential of the three cell lines was assessed in classical transwell assays. In comparison to the reference HROC24, the cells of cell line HROLu55 possessed a higher capacity for invasion and migration. The cells of HROLu22 and HROBML01 were comparable to the reference with regard to both invasion and migration ( Table 6 and Figure 8 ).

These properties of higher invasion and migration capacity resulted in faster wound closure rates ( Table 7 and Figure 9 ) for HROLu55. These cells are capable of a rapid wound closure (25.21 µm/h). Cells of HROLu22 were 8.99 µm/h slower than cells of HROLu55 but substantially faster than HROBML01. The cell line HROBML01 shows a very slow closure rate (1.23 µm/h). Unfortunately, the results for HROBML01 are inconclusive since cells tend to detach from the culture flask during or after wound infliction.

Further properties associated with tumorigenicity are the capacity to form colonies starting with very few tumor cells (≈100 cells) and three-dimensional spheroids ( Figure 10 ). Classical epithelial tumor cell colony formation was observed for HROBML01 and HROLu22. Cells of the cell line HROLu55 also grew in colony-like formations; however, these resemble more a loose accumulation of cells without direct cell-to-cell interaction.

Spheroid formation was observed after 7–10 days for all three cell lines. The earliest onset and most pronounced spheroid formation was observed for HROLu55 (data not shown).

Adherent 2D cell lines, especially in the NCI-60 panel, have a long-standing tradition as tools for in vitro response testing (45). In particular, patient-derived cell lines possess great potential for highly accurate predictability (7). Thus, response to a broad variety of therapeutics commonly administered for treatment of lung tumors was determined in dose-kinetic analyses ( Figure 11 ). All IC₅₀ value calculations are in the range typically found in clinical settings ( Table 8 ). Thus, pre-existing drug resistance for all three cell lines and tested reagents is unlikely. HROLu22 tended to be least sensitive towards treatment with cisplatin, carboplatin, and etoposide. Response to vinorelbine in HROLu22 cells plateaued over a (wide) range of concentrations until viability finally decreased to 0.

Treatment of patients with a single substance is rarely performed. Thus, clinically more relevant are drug combinations, which were tested in a so-called checkerboard assay ( Figures 12 , 13 and Supplementary Data 2 consisting of Supplementary Table 2 and Supplementary Figure 1 ). Observed interaction scores range from −9.63 (for HROBML01 and the combination of cisplatin with vinorelbine) to 10.95 (for HROLu22 and the combination carboplatin with vinorelbine). Most studies suggest a cutoff at about −10 and 10 (46–48) for calling synergy or antagonism, respectively. None of the tested combinations largely surpass these values; thus, observed effects were most likely additive. More importantly, no antagonistic effects occurred.