Diagnostic leukapheresis reveals distinct phenotypes of NSCLC circulating tumor cells

DLA was performed on four NSCLC patients with adenocarcinoma and two NSCLC patients with squamous cell histology. At the time of the DLA, three of those patients had active disease (initial diagnosis or progression from previous treatment), while tumors from the remaining patients were controlled (Supplementary Table 1).

A mean blood volume of 6.0 L per patient was screened during DLA and a mean number of 64.6 × 10⁸ WBCs were collected. MNCs were sufficiently collected via DLA (mean MNC collection efficiency was 41% (range: 26 - 68%, Supplementary Table 2)). As described before, DLAs were generally well tolerated [6]. Blood samples taken before and after DLA indicated minor decreases in peripheral differential blood counts, but all numbers were within the normal range (Supplementary Fig. 1).

For comparison, we also investigated CTCs from a 7.5 ml PB sample taken immediately prior to DLA and from 2 × 10⁸ cells of each DLA, using the “goldstandard”, FDA-cleared CellSearch^® system. In our cohort of advanced stage NSCLC patients, CTCs were detected in 2 out of 6 patients (33%), which is consistent with other studies using the EpCAM-based CellSearch^® system for detection of CTCs in patients with advanced stage NSCLC [7]. Using DLA increased CTC numbers in these patients (mean CTCs PB 9 (range 0–41) vs. mean CTCs DLA 54 (range 0–216) (Fig. 1B).

Because many NSCLC-CTCs lack expression of epithelial markers such as EpCAM [4], we purposely chose an unbiased approach based on a negative selection of CTCs by targeting hematopoietic cells (HPCs) for CTC enrichment from DLAs prior to sc RNA sequencing (scRNAseq) (Fig. 1A). First leukocytes (CD45⁺), T-cells (CD3⁺), endothelial cells (CD31⁺), monocytes/neutrophils (CD16⁺) and erythrocytes (CD235a⁺) were depleted using magnetic beads. Next, the remaining fraction was FACS sorted for live (DAPI⁻) and CD45⁻ cells. This population was then subjected to whole transcriptome scRNAseq via 10X Genomics technology.

For identification of CTCs, we pooled sc transcriptomes from all six patients. Shared nearest neighbor modularity clustering revealed 14 distinct clusters containing a total of 9,659 cells (Fig. 1C). High cluster stability and consistency, especially of the CTC cluster, was confirmed by Jaccard Index analysis and Silhouette width (Supplementary Fig. 2A and B, Supplementary Table 3). We utilized reference-based cell type annotation comparing to a reference data set with 'SingleR' and canonical marker gene expression and identified 7 distinct epithelial cell clusters; other identified cell types included megakaryocytes, neutrophils, natural killer (NK) cells, megakaryocyte/erythroid progenitors (MEPs), common myeloid progenitors (CMPs)/pro-myelocytes (CMPs) and granulocyte-monocyte progenitors (GMPs) (Fig. 1D & E, Supplementary Fig. 3A & B, Supplementary Table 4). Trajectory analysis, which enables the study of dynamic changes in gene expression, revealed a separate branch of epithelial cells compared to HPC, confirming a different origin for these cells (Supplementary Fig. 3C & D). Additionally, comparison of differentially expressed genes (DEGs) between HPCs and CTCs indicated substantially higher levels of epithelial markers, including keratins as well as increased expression of cell cycle (CCND1) and proliferation-related genes (SFN) in epithelial cells. S100A2, NQO1, and ID1 were also amongst the top DEGs. These genes are known to be expressed in epithelial cells, including respiratory cells and are associated with cancer progression [9, 10]. In contrast, HPCs exhibited higher expression of genes associated with hematopoietic lineages, such as MPO (myeloperoxidase), DEFA3 (neutrophil defensin 3), and HBA1 and HBA2 (hemoglobin subunits alpha 1 and 2, Supplementary Fig. 3E, Supplementary Table 5). We specifically also investigated endothelial and fibroblast maker genes, as these cells can be found in the peripheral blood. None of the marker genes were identified in either HPC clusters or CTC clusters, with the exception of COL1A1, which was partially expressed in CTC cluster 6 (Supplementary Fig. 3F). Since cells in CTC cluster 6 also expressed EPCAM and KRT19, these cells are of epithelial origin (Fig. 1J). Comparing inferred copy number variations (CNVs) from epithelial cells to reference HPCs revealed notable evidence for CNVs in epithelial cells. Thus, these cells are henceforth referred to as CTCs (Fig. 1F). Overall, a total number of 3,363 NSCLC-CTCs was identified. As a note, inferCNV analyses from CTCs with healthy lung epithelial cells as reference confirmed increased CNV in CTCs (Supplementary Fig. 3G). 

Integrating a publicly available healthy donor PBMC scRNA dataset into the DLA scRNAseq dataset, demonstrated that the seven previously identified distinct CTC clusters prevailed and exclusively consisted of cells from DLA patients, while HPCs from the DLA scRNAseq dataset formed clusters with HPCs from the healthy PBMC dataset (Supplementary Fig. 4). This further corroborates that our pipeline can be used for the specific identification of CTCs.

Notably, the Human Primary Cell Atlas (HPCA) reference within SingleR does not classify megakaryocytes per se, but platelets (Supplementary Fig. 3A). Since most platelets were excluded during enrichment, we tested whether cluster 8 (Fig. 1C) was composed of platelet-coated CTCs. The DEGs between cluster 8 (‘platelets’) and HPCs did not indicate any expression of epithelial marker genes such as keratins or EPCAM, thus contradicting this hypothesis (Supplementary Table 6). Hence, cluster 8 was classified as megakaryocytes.

Surprisingly, CTCs clustered independently of individual patient, primary tumor histology or disease state (Fig. 1G-I), indicating a cellular state-based CTC-intrinsic clustering. To further analyze the different CTC clusters, we performed in-depth characterization of their transcriptomic profiles by analyzing mean gene expression in combination with pathway- and trajectory analyses (Fig. 1J-L, Supplementary Fig. 3H & I, Supplementary Tables 7–10). A heatmap showing the top 10 positive marker genes of each CTC cluster revealed that genes associated with epithelial phenotypes (KRT5, KRT6A), stemness (ALDH1A1) and extracellular matrix remodeling (VTN, MMP7, FBLN1) were distinctive of different CTC clusters (Supplementary Fig. 3I, Supplementary Table 7).

This was further specified by comparison of canonical marker genes demonstrating varied expression among CTC clusters (Fig. 1J). While each cluster showed some expression of different epithelial genes, mesenchymal markers were predominantly expressed in CTC clusters 4, 5 and 6, with CTC cluster 4 exhibiting a cancer stem cell (CSC)-like phenotype and cluster 6 displaying markers of migration and hypoxia. CTC cluster 1 showed immune response marker upregulation and marker indicating increased proliferation. CTC cluster 9, 10 and 13 exhibit diverse phenotypes. They showed enriched epithelial markers, but also some upregulation of makers associated with cancer stemness, proliferation and hypoxia.

To enhance understanding of these phenotypes and their relationships, trajectory analysis was performed, elucidating that CTC clusters 1, 5 and 6 represent the endpoints of three distinct branches (Fig. 1K, Supplementary Fig. 3H). Next, we utilized GSEA to understand the distinctions between the three branches and clusters (Fig. 1L, Supplementary Tables 8–10). In comparison to CTC clusters 5 and 6, CTC cluster 1 enriched fewer genes related to the EMT pathway. However, there was an upregulation of genes associated with mitotic spindles, DNA repair, and E2F targets, coupled with increased activity in cell adhesion via apical junctions in CTC cluster 1. Based on the greater expression of genes involved in the interferon-α/γ-response pathways, CTC cluster 1 could further be characterized as being more immune responsive than CTC cluster 5. The immune response pathways were also more highly expressed in CTC cluster 6 than in CTC cluster 5. GSEA confirmed the mesenchymal phenotype of CTC clusters 5 and 6. Remarkably, despite both CTC clusters exhibiting a mesenchymal-like phenotype in comparison to CTC clusters 1, 10, 9 and 13, they were located at the two ends of the trajectory analysis, indicating substantial heterogeneity of mesenchymal CTCs. According to the GSEA, CTC cluster 5 enriched genes involved in oxidative phosphorylation, adipogenesis and fatty acid metabolism, while CTC cluster 6 encompassed genes related to hypoxia, glycolysis and ROS pathways. Additionally, compared with CTC cluster 1 and 6 cells, CTC cluster 5 cells exhibited decreased expression of genes involved in inflammatory response, interferon response α and γ, indicating a rather immune-evasive phenotype (Fig. 1L).

Taken together, these data indicated a high degree of phenotypic heterogeneity and a variety of CTC phenotypes were revealed: (i) epithelial-like, immune responsive and highly proliferative (CTC cluster 1), (ii) mesenchymal, oxidative phosphorylation, and immune evasive (CTC cluster 5), (iii) mesenchymal, invasive, and glycolytic (CTC cluster 6), and (iv) cancer-stem cell like (CTC cluster 4).

To elucidate to which degree CTCs may exhibit distinct phenotypes in comparison to primary tumor cells (PTC), we analyzed the sc CTC-RNAseq dataset alongside an independent scRNAseq dataset from 45 primary NSCLC tumor samples (Supplementary Table 1). We randomly selected 5000 PTCs from the sc PTC-RNAseq dataset set to match the size of the sc CTC-RNAseq dataset, before the two datasets were integrated using the 'harmony' R package for batch effect correction by incorporating NK cells ([11] and Supplementary Material and Methods). This revealed 22 distinct cell clusters in an unsupervised clustering (Fig. 2A and B). As a note, Jaccard Index and Silhouette width indicated good cluster stability and consistency (Supplementary Fig. 2C & D, Supplementary Table 11).

Remarkably, this analysis indicated the preservation of most CTC clusters (Fig. 2B - D). CTC clusters 12 and 13 from this combined dataset comprised more than 90% CTCs, and CTC cluster 3 and 8 contained more than 75% of CTCs, reinforcing a distinctive phenotype of these CTC clusters (Supplementary Table 12). Notably, only CTC cluster 6 (mesenchymal, invasive and hypoxic) was integrated with PTCs (cluster 2 in Fig. 2B). A trajectory analysis further affirmed the separation of CTC clusters from PTCs, suggesting a potential distinctive phenotype for CTCs (Fig. 2E, Supplementary Fig. 5).

GSEA unveiled enriched pathways encompassing proliferation, translation (G2M checkpoint, DNA repair, E2F and Myc targets) and oxidative phosphorylation, while exhibiting decreased expression in TNFα signaling via NFκB, apoptosis and p53 pathways compared to PTCs (Fig. 2F, Supplementary Table 13). Noteworthy, immune-evasive phenotypes were observed in most CTC clusters, evidenced by reduced enrichment in inflammatory and interferon response pathways. Differential gene expression analysis highlighted genes related to hemoglobin metabolism, with HBB and its partners HBA1 and HBA2 highly expressed in CTCs, emphasizing a CTC-specific function (Fig. 2G). Conversely, genes involved in surfactant metabolism (NAPSA, SFTPB, SFTPA2 and SFTA2) were downregulated in CTCs compared to PTCs. Interestingly, giantin (GOLGB1) which is associated with invasion and was shown to be higher expressed on breast cancer CTCs than on primary breast cancer cells, is also higher expressed in NSCLC-CTCs than primary NSCLC tumor cells (Fig. 2G, Supplementary Table 14).

Overall, these data corroborate the hypothesis that CTCs possess a distinct, immune-evasive phenotype, enabling them to adapt to the unique microenvironment and stresses in the bloodstream. On the other side, mesenchymal CTC clusters 5 and 6 exhibit increased expression of genes associated with EMT, while CTC clusters 9, 10 and 13 presented an overall lower expression of EMT genes compared to PTCs. CTC cluster 6, in particular, showcased a metabolically unique profile with elevated expression of anaerobic glycolysis genes despite oxygen availability in the bloodstream. Together, this also emphasizes the role of CTCs in representing heterogeneity of PTCs, particularly concerning EMT, metabolic disparities, and other relevant pathways.