Ex vivo expansion of lung cancer‐derived disseminated cancer cells from lymph nodes identifies cells associated with metastatic progression

The cellular basis of the apparent aggressiveness in lung cancer is poorly understood but likely associated with functional or molecular features of disseminated cancer cells (DCCs). DCCs from epithelial cancers are mostly detected by antibodies directed against histogenetic markers such as cytokeratin or EpCAM. It has been argued that marker‐negative metastatic founder cells might escape detection. We therefore used ex vivo sphere formation for functional detection of candidate metastasis founders. We generated cell suspensions from 199 LN samples of 131 lung cancer patients and placed them into non‐adherent cell culture. Sphere formation was associated with detection of DCCs using EpCAM immunocytology and with significantly poorer prognosis. The prognostic impact of sphere formation was strongly associated with high numbers of EpCAM‐positive DCCs and aberrant genotypes of expanded spheres. We also noted sphere formation in patients with no evidence of lymphatic spread, however such spheres showed infrequent expression of signature genes associated with spheres from EpCAM‐positive samples and displayed neither typical lung cancer mutations (KRAS, TP53, ERBB1) nor copy number variations, but might be linked to disease progression >5 years post curative surgery. We conclude that EpCAM identifies relevant disease‐driving DCCs, that such cells can be expanded for model generation and that further research is needed to clarify the functional and prognostic role of rare EpCAM‐negative sphere forming cells.


| INTRODUCTION
Lung cancer is one of the most common malignancies in Europe and the leading cause of cancer-related deaths worldwide. 1With a prevalence of approximately 80%, non-small cell lung cancer (NSCLC) is the most frequent lung cancer subtype.The overall 5-year relative survival rate for all stages combined is 18% 2 and even 20-25% of patients with a complete resection in stage IA/B will relapse, 3,4 because metastatic dissemination has obviously occurred prior to surgery.This apparent aggressiveness of lung cancer is poorly understood since tumor volume doubling times of NSCLC are not lower (i.e., growth is not faster) than those for other epithelial cancers such as breast or colorectal cancers. 5,6nce, answers might be found by detection and analysis of metastasis founder cells.Prime candidates are disseminated cancer cells (DCCs) that can be detected in regional lymph nodes (LN) or bone marrow (BM) samples long before manifestation of distant metastasis. 7,8Several studies have linked detection of DCCs in LN or BM of NSCLC patients with poor outcome, however, data have been conflicting.Some observed an association between the presence of DCCs and poor prognosis, [9][10][11][12][13] whereas others did not. 14,15We recently analyzed the prognostic impact of DCCs of NSCLC patients in more detail.First, we compared EpCAM and cytokeratin (CK)-positive cells from BM of the same patient and found that the prognostic impact of EpCAM outperformed information provided by CK-positive cells. 16en, we tested whether EpCAM-positive cells from LNs confer prognostic information.Here, we found detection of EpCAM-positive DCCs to be the most informative variable for outcome prediction in multivariable analysis. 17though EpCAM-positive cells have thus become prime candidates for metastasis founder cells, we set out to investigate the following questions.First, we asked whether sphere formation under stem-like-cell selecting conditions is associated with disease progression; second, whether sphere formation is associated with EpCAM-positivity in immunocytologically analyzed samples and third, if spheres are formed, which molecular features were associated with expansion and possibly with disease progression.To address these issues, we collected LN samples from lung cancer patients, quantified DCCs and subjected LN suspensions to culture conditions promoting propagation of stem-like DCCs.We molecularly characterized single spheres to more specifically identify potential metastatic founder cells.

| Patients
Between 2011 and 2016, we collected lymph node (LN) samples from 131 consecutive patients diagnosed with lung cancer at the University Hospital Regensburg or at the Krankenhaus der Barmherzigen Brüder Regensburg.The baseline characteristics of the patient cohort are summarized in Table S1.Tumor stage and grading were determined according to the international system for the staging of lung cancer. 18

| Lymph node sample processing and detection of disseminated cancer cells
Lymph nodes were bisected by the surgeon directly after removal, resulting in half LNs that were processed in our laboratory for immunocytology staining 19 and sphere culture as described below.In total, we received 199 LN samples from 131 patients.This included patients of whom we received either one bisected LN (64 patients), two bisected LNs (66 patients) and three bisected LNs (1 patient).LNs were mechanically disaggregated using a Medimax machine (Syntec International), and cells were subsequently enriched via Percoll (GE Healthcare) gradient centrifugation. 20Cell numbers were determined with a Neubauer counting chamber.A total number of 2 Â 10 6 cells (at a density of 10 6 cells/ml PBS) were transferred onto adhesion slides (Thermo Scientific).After 1 h sedimentation, PBS was removed and slides were air-dried overnight at room temperature.For detection and quantification of DCCs, LN samples were stained with anti-EpCAM antibody (Ber-EP4, Dako) and visualized via Alkaline Phosphatase-anti-Alkaline Phosphatase antibody (APAAP, Bio-Rad) or an AP conjugate substrate kit (Bio-Rad). 21,22The number of EpCAM stained cells was counted and expressed as disseminated cancer cell density (DCCD = number of DCC per million LN cells).
Single positive cells from selected patients were manually picked from adhesion slides using a micromanipulator and subjected to whole genome amplification and molecular analysis as described below.

| Sphere cultures
To promote non-adherent sphere formation, 10 6 pre-enriched LN cells were cultured in polyHEMA-coated 6 cm plates (12 mg/ml, Sigma Aldrich) under low-oxygen conditions (7% O 2 ) at 37 C in 5 ml defined sphere medium.Since no information on tumor-type specific growth conditions for DCCs from lung cancer patients was available, we compared two non-adherent, serum-free sphere culture media conditions.First, a basic sphere culture medium (medium A), which was previously shown to support sphere formation of melanoma DCCs 23 or circulating tumor cells in metastatic breast cancer. 24,25Medium A contained DMEM/F12 culture medium supplemented with 0.5% penicillin/ streptomycin, 10 μg/ml insulin (all from PAN-Biotech), 10 nM HEPES (Sigma-Aldrich), 1Â B27 supplement (Gibco), 10 ng/ml epithelial growth factor, 10 ng/ml basic fibroblast growth factor, 4 μg/ml heparin (all from Sigma-Aldrich), 5 ng/ml GRO-α (R&D Systems), 20 ng/ml hyper interleukin-6 (kindly provided by S. Rose-John) and 0.2% Methylcellulose (Sigma-Aldrich).Medium B was comprised of medium A supplemented with 20% mouse embryonic fibroblast conditioned medium (R&D Systems), which was shown to support cell growth by secretion of growth factors, cytokines and nutrients. 26Since we observed no significant difference in sphere-formation between medium A and B within DCCD = 0 (P = .3125)and DCCD>0 groups (P = .5885)the type of medium was neglected for evaluation (Figure S1).Cultures were monitored at least once a week up to 6 weeks and complemented with 500 μl fresh medium once a week.Compact cell aggregates and well-rounded 3D structures bigger than 40 μm were counted and henceforth referred to as spheres.Photographic documentation was performed for each sphere before analysis.

| Combined whole transcriptome-and genome amplification (WTA and WGA)
Reverse transcription of mRNA and amplification of cDNA was performed as described previously. 27,26The quality of amplified cDNA was evaluated with a set of 3 control primers using end-point PCR (quality control; QC). 28Genomic DNA of spheres was collected during WTA procedure, precipitated and subjected to WGA. 29,30 DNA from isolated DCCs or supernatants from spheres were amplified using Ampli1 WGA Kit (Menarini Silicon Biosystems).The quality of amplified DNA was evaluated with a set of 4 control primers using end-point PCR and assigned to genomic integrity index 0 to 4 (GII0-GII4). 31

| Detection of copy number variations (CNVs) using array comparative genomic hybridization (aCGH) and low-pass whole genome sequencing (WGS)
Genome-wide copy number analysis of DCCs and spheres was conducted using a previously published array comparative genomic hybridization (aCGH) protocol. 32Low-pass is a sequencing approach targeting the whole genome with average coverage <1Â.Libraries were prepared using Ampli1 LowPass kit for Illumina platforms according to the manufacturer's instructions (Menarini Silicon Biosystems).Libraries were quantified using Qubit dsDNA HS Assay kit and Qubit 2.0 Fluorometer (Thermo Fisher Scientific) and average fragment sizes were assessed using Agilent High Sensitivity DNA Kit on the Agilent 2100 Bioanalyzer System (Agilent Technologies).Libraries were mixed in equimolar concentrations to obtain a 4 nM final pool ready for direct sequencing.
Sequencing was performed in single read (SR) mode on a MiSeq System with MiSeq Reagent Kit v3 (150-cycle) (Illumina).For CNV profile analysis the human genome was divided into non-overlapping bins, each with a size of 500 kb.After removing adapter sequences and poor-quality bases (BBDuk 38.84 33 ) as well as decontamination of non-human reads originating from microbial and/or fungal fresh water and/or reagents (BioBloom Tools 2.0.13 34 ), for each sample, reads aligned with bwa mem 0.7.17 35 to the reference genome hg19 were counted per bin, corrected, filtered, normalized and segmented with the bioconductor package QDNAseq 1.26.0. 36The same package was used to create the log2 (ratio) CNV profiles.Samples with <100.000mapped reads were rejected from CNV-analysis.The sequencing coverage and quality statistics for each low-pass WGS sample are summarized in Table S2.

| RNA-sequencing (RNA-Seq)
Good quality WTA products from 25 spheres were processed for RNA sequencing as previously described. 37Sequencing libraries were prepared according to the TruSeq DNA PCR-Free Library Prep Kit (Illumina).Libraries were quantified with KAPA Library Quantification Kit for Illumina Platforms (Roche), pooled in equal molar ratios and sequenced using Illumina NextSeq 2000.Fastq reads were processed using inhouse RNA-Seq workflow.The sequencing coverage and quality statistics for each RNA-Seq sample are summarized in Table S3.Initial read processing including quality control, trimming and decontamination were performed as described above in the lowpass WGS section.Following this, reads were aligned to the hg38 reference genome with STAR 2.7.10b. 38 FeatureCounts from Subread 2.0.0 39 was used to extract counts per gene per sample.Only uniquely mapping reads to exonic regions were used for counting.
Library complexity (using Preseq 2.0.3 40 ), genomic origin of the reads and the 5 0 -3 0 -bias (both using QualiMap 2.2.2d 41 ) were evaluated from the mapping results.Final counts table were utilized to uncover the cellular heterogeneity of the spheres.Following steps were performed in R programming language 4.2.1.As first steps, gene counts were log normalized and scaled.Log (counts) as well as log (cpm) and top 500 differential expressed genes were visualized using Principle Component Analysis (PCA) and t-distributed stochastic neighbor embedding (t-SNE) clustering techniques.RunPCA and RunTSNE from Seurat 4.2.0 42 were used to construct PCA and t-SNE, respectively.Finally, using the nonparametric ssgsea method in gene set variation analysis (GSVA) package 1.44.5 43 the spheres were deconvoluted to identify different cell types.Subset of cell types from lung from CellMarker database, 44 a manually curated resource, was used as gene set in gsva against which the count table was evaluated.All the plots were made using ggplot2 3.4.2. 45

| DNA fingerprinting
Patient origin was confirmed using short tandem repeat (STR) analysis (Geneprint10, Promega).Due to WGA of DCC and sphere samples, which includes restriction digestion by Mse I prior to STR analysis, in some cases only STR loci TH01, D21S11, D5S818, D13S317, D16S538 and vWA were accessible for analysis.Amplified fragments were detected using 3130-Avant Genetic Analyzer (Applied Biosystems).Fragment sizes were determined and evaluated using Gene-MapperSoftware Version 4.1.STR-profile of cultured spheres were compared with the profile of the corresponding DCC.All analyzed patient samples matched each other.The obtained STR profiles were additionally analyzed using a public database (https://web.expasy.org/cellosaurus-str-search/).This comparison showed no match with other cell lines.Patient samples are therefore revealed as unique and not cross-contaminated or misidentified with cells of other origin.

| Statistical analysis, follow up and survival endpoints
Statistical analysis was performed with GraphPad Prism Software, version 9, using unpaired, non-parametric Mann-Whitney test, 1-way ANOVA or fisher's exact test to determine associations between two categorical variables.For survival analysis Log-rank (Mantel-Cox) test for comparing survival curves.Asterisks represent the significance as follows: ***P < .001,**P < .01,*P < .05 and ns: P ≥ .05.
Overall survival (OS) was calculated as a period of time from the date of surgery to the date of death from any cause or the date of last follow-up.Tumor-specific survival (TSS) was defined as the duration from the date of surgery until death that was documented to be caused by lung cancer.

| Detection of EpCAM-positive cells in lymph nodes and ex vivo cultivation
We investigated the presence of EpCAM-positive DCCs by immunocytological quantification and concomitantly cultivated cell suspensions of 199 lymph node samples from 131 lung cancer patients (Figure 1A, patient demographics listed in Table S1).DCCs were detected in 49 of 199 analyzed LNs (24.6%), with a calculated disseminated cancer cell density (DCCD = EpCAM-positive DCCs per one million LN cells) ranging from 0.5 to 750,000 (median = 3; Figure S2).
In order to investigate the proliferative potential of DCCs from LNs, LN-derived cell suspensions were cultivated under non-adherent, serum-free sphere culture conditions using two different media.We observed morphologically heterogeneous sphere structures among different patients, but also within individual patient samples.Spheres differed in size (range 40-800 μm in diameter), color (ranging from light gray to deep black, which may indicate inclusion of lung macrophages filled with tar) and compactness (ranging from clusters to compact spheroids; representative pictures in Figure 1B).Irrespective of their morphological appearance, all spherical structures were counted and documented microscopically for each LN culture.
Finally, we tested whether sphere formation in LN cultures correlates with clinical outcome of the patients.We found that the presence of spheres in LN-cultures was associated with a reduced overall survival (OS) as well as tumor-specific survival (TSS) within the lung cancer patient cohort (Log-rank test, OS: P = .0253,TSS: 0.0391; Figure 1E).

| Molecular characterization of single spheres
Therefore, we further characterized single spheres on a molecular level.We manually isolated 325 single spheres derived from DCCD = 0 and DCCD > 0 LN samples from 55 patients (median = 4, range of 1-22 per LN).In addition to phenotypical categorization of size, shape and color, we isolated RNA (for WTA) and DNA (for WGA) from each sphere and assessed WTA and WGA quality.We identified 215 LN spheres with sufficient WTA quality (QC2-3) and 224 spheres with good DNA quality (GII3-4) (Figure 2A).We observed that poor WTA and WGA quality was visually strongly associated with cellular particles smaller than 40 μm in size or very dark colored aggregates, possibly representing necrotic spheres or clusters, which were therefore excluded for additional downstream analysis.

| Expression profiling of selected transcripts in single spheres
To  which are commonly described and investigated for many cancer entities including lung cancer cells. 46,47In addition, we tested PROM1 (CD133), a stem cell marker described to be expressed by a rare tumor cell population within lung cancer patients with the potential to form spheres, 48,49 and the leukocyte common antigen PTPRC (CD45) as common leukocyte marker to uncover the cellular composition of the spheres.
Moreover, the stem cell marker PROM1 was exclusively expressed in  2C).Henceforth, patients with at least one sphere expressing either KRT19, EpCAM and/or PROM1 will be referred to as DCC-signature positive.Of note, in most of the analyzed spheres expression of epithelial markers KRT8 and KRT18 was frequently observed, to a similar extent in samples derived from DCCD = 0 patients (CK8: 75%, CK18: 65%) as well as from DCCD > 0 patients (CK8: 87%, CK18: 87%; Figure S3A).Remarkably, we found heterogenous transcript expression in spheres of individual patients, including spheres expressing the DCC-signature genes as well as CD45 (exemplary data are shown for five patients in Figure S3B), indicating that in some cases non-cancer cells support sphere formation.

| Mutational analysis of single spheres
We then asked whether spheres derived from DCCD = 0 and DCCC > 0 LNs differ with regard to mutations typically found in lung cancer and analyzed 131 LN-spheres from 59 LN of 51 patients for the presence of mutations in TP53, EGFR and KRAS.For this, we amplified TP53 exons 5, 6, 7 and 8, EGFR exons 18, 20 and 21 and KRAS exon 2. In total, we identified four patients with spheres harboring mutations, all derived from DCCD > 0 LNs (Figure 2D).Of note, TP53, KRAS or EGFR mutated spheres from these patients were also DCC-signature positive (Table 1).

| Copy number variations (CNVs) of single spheres
We then tested sphere-derived DNA for copy number variations by aCGH as well as low-coverage whole genome sequencing (low-pass sequencing) of selected 35 patients.We analyzed spheres of 12 patients with DCCD = 0, comprising six patients expressing DCCsignature genes and additional six patients with a documented disease-related death and/or early disease progression, as well as spheres of 23 DCCD > 0 patients.Chromosomal aneuploidies and CNVs could not be detected in any of the patients with DCCD = 0 (Table 1A and Figure S4A).In contrast, CNV analysis of DCCD > 0 patients revealed genomic aberrations in 10 of 23 patients (43.5%,Table 1B, see Figure 3 for aberrant and Figure S4B for balanced spheres).
When comparing the genomic profiles of the analyzed spheres with the corresponding single DCCs isolated directly from the LN suspension of the same patient, we found that the profiles of ex vivo generated spheres represent matched DCC-derived profiles (Figure 3).oncogenes like EGFR and BRAF, and chromosome 8, containing MYC.
On chromosomes 13 and 17 we frequently observed common deletions of genomic regions containing well-described tumor suppressor genes like BRCA1 and 2, RB1 and NF1 and oncogenes like ERBB2 (Figure S5).

| Gene expression analysis of spheres
To further decipher the cell type composition within single spheres, we performed RNA-Seq of 25 selected spheres of four different categories (including 8 signatureÀ/balanced, 9 signature+/balanced, 1 signatureÀ/ aberrant, 7 signature+/aberrant).Dimensionality reduction plots based on the top 500 most variable genes revealed that signature+/aberrant spheres were clearly separated from all other groups (Figure 4A).Furthermore, we observed a tendency that balanced spheres grouped together in three sub-clusters without a clear separation based on signature gene expression.Overall, the clustering identified clear differences in gene expression between balanced vs aberrant spheres.
To track the individual cell types within single spheres we additionally performed deconvolution analysis (Figure 4B).Aberrant spheres (with exception of the one signatureÀ/aberrant sphere, 019) most prominently classified into specific cancer cell related categories such as "cancer cell, lung epithelial cell, malignant mesothelioma or migrating cancer stem cell."However, expression of non-cancer gene categories such as "plasmacytoid dendritic cell or fibroblast" reveal that aberrant spheres also contain additional stromal cell types.In contrast, signature-and signature+ balanced spheres scored highest in "fibroblast, myofibroblast and plasmacytoid dendritic cell" categories as well as in categories related to innate immune cells (e.g., "M1/M2 macrophage, natural killer cell, dendritic cell").Remarkably, all spheres possibly contained (hidden) epithelial DCCs as indicated by the identified expression of "alveolar epithelial progenitor cell or epithelial cell" gene sets.Epithelial cell content of spheres did not correlate with histological subtypes.

| Identification of high-risk patients with sphere formation
All our analyses revealed that characteristic traits (e.g., signature genes, marker mutations, CNVs) were more frequently or exclusively found in spheres from patients with DCCD > 0 (Figure 5A).As DCCD > 0 patients with aberrant spheres also displayed a significantly Here, we show that sphere formation under stem-like-cell selecting conditions is associated with patient outcome.Spheres were characterized by expression of selected signature genes, cancer-related gene sets, CNVs and more easily generated from samples with a DCCD > 18.We also noted that sphere forming samples of DCCD = 0 patients tended to reduced survival.
Metastasis founder cells must be comprised among DCCs, however currently used markers fail to unambiguously identify them.
DCCs are detected by staining cells from mesenchymal organs via antibodies directed against CK (for bone marrow DCCs) and EpCAM (for lymph node DCCs), commonly used in lung cancer and a large number of other carcinomas to identify patients at risk of relapse. 50wever, since in all tested cancers a significant proportion of marker-negative patients progress, it has long been argued that marker-negative cancer cells exist.We therefore used sphere formation under stem-like-cell selecting conditions and then compared it to the immunocytological result.
The observed sphere formation rate of immunocytological-positive LNs was higher than for published lung cancer tissue biopsies (49% vs 40%). 51Sphere numbers were considerably lower compared with the corresponding DCC numbers, indicating that only a subpopulation of DCCs was competent to survive and proliferate in the applied culture conditions.Interestingly, sphere formation was also documented in 30% of DCCD = 0 samples.None of the previous studies using comparable sphere culture conditions reported generation of non-tumor cellderived spheres, however, only few validated the malignant origin of spheres generated from patient-derived material. 51en we compared spheres from patients with DCCD = 0 vs DCCD > 0 LNs, spheres of both groups expressed CD45 at a comparable frequency, precluding DCC-origin confirmation based on this exclusion marker.However, LN-spheres from patients with a DCCD > 0 more frequently expressed epithelial markers EpCAM, CK19 and the stem cell marker CD133, indicating that those spheres may comprise cells of epithelial origin and stem cell-like properties.Since the absence of EpCAM transcript in spheres derived from EpCAM-negative samples was somehow expected, we also searched for genetic hints of malignancy in spheres derived from DCCD = 0 samples, however, none were found.In contrast, low-pass sequencing to detect CNVs confirmed tumorigenic origin in 43% of tested DCC-positive patients, with four of them also harboring typical gene mutations (TP53, KRAS and EGFR).
Some DCC-negative and -positive samples expressed one or several of the signature genes but did not exhibit CNVs that would provide unequivocal evidence of malignancy.Gene expression analysis of such spheres revealed that they mainly consist of fibroblasts or innate immune cells but we also found genes characteristic like "epithelial or alveolar epithelial progenitor cells".This might indicate the existence of sphere-forming cancer cells that lack typical mutations and/or are present in low abundance within spheres, so that CNV are not detected.If present, such DCCs left behind in patients may have the potential to progress after years of clinical latency-possibly after they acquired disease-driving genetic alterations.This reasoning appears to be supported by the borderline significant association (OS: P = .065,Figure 5D) of sphere forming samples and disease progression in early stage (pN0) patients.In this group, Kaplan-Meier survival analysis revealed separation of the survival curves rather late, that is, only after 65 months, consistent with ongoing cellular evolution.However, we cannot conclusively decide whether the detection of the epithelial gene sets in samples negative by DCC immunocytology indicates the presence of hidden DCCs in the tested spheres or not, since we did not obtain thoracic lymph nodes from non-cancer patients as controls.
To clarify these issues further studies such as single cell RNA-Seq and more extensive follow-up studies will be needed.
These observations support our starting hypothesis that detection of marker-negative DCC may be uncovered by growth in culture, since currently used markers fail to unambiguously identify distantdisease driving cells in all patient samples.Comparing sphere formation under stem-like-cell selecting conditions with immunocytological results side-by-side provides further support for this reasoning.
In the case of more extensive lymph node involvement (DCCD > 18), sphere formation with expression of gene sets reflecting epithelial and cancer stem cell profiles and detection of CNVs place patients at highest risk.In addition, samples with mutations (n = 4, 17% of all DCCD > 0 sphere forming patients) also displayed CNVs, and 8/ 10 samples with CNVs also expressed the DCC signature genes.Apparently, a high number of DCCs in LNs reflects a proliferative phenotype and genotype that also enables sphere formation.In addition, specific genomic aberrations are acquired during proliferation within the LNs, 17 which may drive sphere formation.This is consistent with observations in melanoma, where we previously found that xenograft formation and acquisition of specific alterations within the LN are intimately linked. 23 the other hand, not all samples with high DCCD formed spheres and therefore more investigations are needed to identify the cellular and (epi)genetic changes required for metastatic colony formation and ex vivo expansion.
Our conclusions are inherently linked to the methods applied, that is, non-adherent sphere culture conditions.Alternative approaches like organoid culture 52 may be tested in the future.For the time being, our sphere assay, when applied to samples with a DCCD > 18 generates cell models that (i) reflect the genotype of the DCCs detected by immu-

Frequency
U R E 2 Molecular characterization on RNA and DNA level.(A) Overview and summary of information of isolated, amplified and tested spheres analyzed for the diverse molecular assays.(B) Frequency of LNs with spheres expressing indicated transcripts derived from DCCD = 0 (black) or DCCD > 0 (green) samples (n = 63 LN of 53 patients).(C) Frequency of LN with spheres expressing DCCsignature of DCCD = 0 (black) compared with DCCD > 0 (green) samples (n = 63 LN of 53 patients), P = .0161,Fisher's exact test.(D) Mutational analysis of spheres.Exemplary electropherograms show wild type and mutated regions detected in TP53 exon7 and exon8, KRAS and EGFR exon18 and exon21.
Upon closer examination of the genomes, we recognized in several cases genomic gains on chromosome 7, containing lung cancer-related F I G U R E 3 CNV analyses.Genomic profiles of matched pairs of DCC and spheres.Histogram of genomic gains (red) and losses (blue) of 10 aberrant patients are shown.ADC, adenocarcinoma; NEC, neuroendocrinic carcinoma; SCC, squamous cell carcinoma.

F I G U R E 4
Cell composition of LN-derived spheres.(A) tSNE plot of 25 spheres of different categories (DCC-signature and CNV profile).Balanced spheres are indicated in black, aberrant in green.Signature-spheres are shown as squares, signature+ as triangles.(B) Heatmap from deconvolution analysis of 25 spheres using a subset of cell types from lung from CellMarker database.Patient ID is on x-axis, cell types are in yaxis.Phenotype categories (CNV profile, DCC-signature and subtypes) of the patient samples are highlighted on top.Heatmap colors correspond to the enrichment scores (es) of the cell type markers and range from À0.5 to 0.5 (blue to yellow).
nocytology and (ii) are associated with disease progression.They can therefore be considered representative models for drug testing experiments to uncover vulnerabilities in early systemic lung cancer.Whether spheres devoid of malignancy indicators comprise the earliest metastasis founders or reflect a cancer-associated epiphenomenon, is open for future studies.

1
Data of patients with spheres and their analyses