Development of NSCLC CDX mouse models with ptPDX-derived CTCs
Primary tumor tissues from non-metastatic NSCLC patients were collected at the time of lung resection (Table S1). Ten ptPDX models were generated by growing primary tumor fragments following subcutaneous (s.c.) implantation in immunodeficient NOD scid gamma (NSG) mice (Fig. 1A; upper panels). CTCs from ptPDX mice blood were isolated, enriched, and immunostained with traditional CTC identification criteria (Fig. 1A; lower panels). CTCs from ptPDX phenotypically matched CTCs isolated from patients (Fig. S1). CTCs isolated from ptPDX were then injected subcutaneously in naïve NSG mice to establish CDX. ptPDX-derived CTCs developed into stable CDX tumors in two out of ten (20%) patients (MU150, MU197) (Fig. 1B; Table S1). High numbers of CTC clusters present in the ptPDX mice may be a driving factor for successful CDX development (Table S2). CTC clusters have enhanced metastatic potential [11], and in PDX models CTC clusters correlated with metastatic development [12]. Pathology-reported biomarker expressions in resected patient primary tumors were conserved in matched ptPDX and CDX tumors (Fig. S2 A/B).
Due to the rarity and instability of micrometastatic CTCs, stable CTC expansion models from non-metastatic NSCLC patients are lacking. Although xenograft models represent a clonal cell selection, ptPDX-derived CDX liquid biopsy models can still be a valuable tool to molecularly study and predict the risk of future recurrences and metastases after curative resection of localized NSCLCs in individual patients.
Single cell analysis revealed an additional regenerative AT2-like population in CDX tumors and human patients’ NSCLC metastases
To generate global atlases of the transcriptomic landscape, ptPDX and CDX tumors were profiled by single nuclear RNA-sequencing (snRNA-seq) (Figs. 1C, S3, S4). Nine cell type clusters across all samples were visualized, except for MU197 ptPDX that had ten clusters (Fig. 1C; upper panels). Cell types were identified using differentially expressed genes (DEGs) expressing canonical, cell-specific markers (Fig. S5). As expected in xenograft tumors, most of the cell clusters were of epithelial origin. CDX tumors had two AT2-like cell populations, in contrast to ptPDX tumors that had just one. Feature plots of AT2-specific marker surfactant protein B (SFTPB) confirmed the presence of these populations (Fig. 1C; middle panels). Heatmaps of AT2 canonical marker expression in DEGs of all clusters also showed presence of an additional AT2-like cluster in CDX tumors (Fig. 1C; lower panels).
To validate our findings, we utilized an external single cell sequencing data set consisting of eight NSCLC primary tumors and five metastases [10]. This analysis of non-xenografted tumors showed 14 clusters in the primary tumor and 12 clusters in metastatic tissues (Fig. 1D; upper panels) that were annotated, as above (Fig. S6). Similar to the observation in CDX tumors, an additional AT2-like cluster was present in metastases, but not in primary tumors. Feature plots of the AT2-specific marker SFTPB verified an additional AT2-like population in metastatic NSCLC tissues (Fig. 1D; middle panels). Additionally, heatmaps showing AT2 canonical marker expression in DEGs of all clusters of primary and metastatic patient tumor tissue samples recapitulated the findings observed in ptPDX and CDX tumors, respectively (Fig. 1D; lower panels).
Single cell sequencing demonstrated the existence of an additional AT2-like cell population in metastatic CDX and, importantly, also in non-xenografted NSCLC metastases tissues. It is well reported that AT2 cells, apart from their regenerative stem cell capacity in non-cancerous lung, are found in lung cancers and have been linked to tumor initiation demonstrated in preclinical models [13, 14]. A recent single cell sequencing analysis on advanced-stage NSCLC tissues identified an AT2-like population expressing cell proliferation and migration genes [15]. These findings support our results suggesting that AT2-like cells may have a role in metastasis development.
Gene expression similarities and differences within and across patient-matched PDX/CDX models correlate with aggressive tumor growth
To explore the similarities and differences of the patient-matched PDX/CDX models, we aggregated snRNA-seq of PDX and CDX samples. MU150 PDX was strikingly different than MU150 CDX, as determined by DEGs following aggregation (Fig. S7 A). Gene set enrichment analysis (GSEA) [16] and pathway enrichment analysis [17] performed using top DEGs showed that cell adhesion/migration genes were commonly upregulated in MU150 PDX/CDX, whereas MU150 CDX had higher enrichment of ERBB signaling and Rho GTPase pathway genes (Fig. S7 B). Unlike MU150 PDX/CDX, MU197 PDX DEGs were not very different from MU197 CDX (Fig. S7 D). However, MU197 CDX showed enrichment of cell adhesion/migration ERBB signaling and Rho GTPase pathway genes, similar to MU150 CDX (Fig. S7 E). It is well reported that cell adhesion/migration, ERBB signaling, and Rho GTPase pathway gene upregulation leads to increased aggressiveness in cancer, including NSCLC [18,19,20]. MU150/197 CDXs were more aggressive in tumor growth compared to their parental PDXs (Fig. S7 C/F). To explore the similarities and variabilities of PDX/CDX models across patients MU150 and MU197, we compared aggregated snRNA-seq of MU150 (PDX + CDX) with MU197 (PDX + CDX). DEGs were strikingly different (Fig. S8 A); GSEA and pathway enrichment analysis showed upregulation of cell adhesion/migration and EMT genes in MU150 PDX/CDX (Fig. S8 B/C) which may be the reason for aggressive tumor growth compared to MU197 PDX/CDX models as discussed above (Fig. S8 G/H). Further, aggregation of CDX models across patients showed that MU150 CDX is different than MU197 CDX (Fig. S8 D). MU150 CDX showed enrichment of know cancer aggressiveness causing genes/pathways, such as cell adhesion/migration genes and ERBB pathway (Fig. S8 E/F), which may be the reason for the aggressive growth kinetics of MU150 CDX versus MU197 CDX (Fig. S8 H).
Chemosensitivity testing of CDX models with carboplatin/paclitaxel and MYC blockade to overcome drug resistance
Post-surgery, patients at higher risk receive chemotherapy to eradicate micrometastatic, minimal residual disease. While PDX models represent the primary tumor, CDX models are derived from CTCs and represent micrometastatic disease targeted by adjuvant chemotherapy. The few CDX models that were established so far have been shown to be valuable tools for drug response testing [8, 9, 21]. Hence, the clinical utility of CDX models as drug testing platforms was determined by administering standard-of-care doublet paclitaxel/carboplatin chemotherapy intraperitoneally to CDX tumor-bearing mice. MU150 CDX tumor growth was not altered by carboplatin/paclitaxel in comparison to vehicle-treated controls, consistent with chemoresistance, whereas MU197 CDX tumors were chemosensitive as demonstrated by significant reduction in tumor growth (p < 0.01; Student’s t-test) (Fig. 2A). As observed in human patients, we noted differential chemotherapy responses further supporting the potential for these models in translational studies.
MYC is a known contributor to drug resistance in cancer [6]. snRNA-seq DEG analysis showed MYC target genes were enriched in chemoresistant MU150 CDX tumors (Fig. 2B; upper panel). Western blot and immunohistochemistry also confirmed overexpression of MYC protein (Fig. 2B; middle and lower panels). In in vitro studies, carboplatin/paclitaxel-resistant MU150 CDX tumor-derived spheroids became chemosensitive when a MYC/MAX dimerization blocker (10058-F4) was added (Fig. 2C/D). MYC blockade led to a significantly higher dead cell percentage and inhibited cell proliferation (p < 0.001, multiple t-test) (Fig. 2E/F). Blockade of MYC/MAX dimerization with 10058-F4 was confirmed by downregulated protein expression of telomerase reverse transcriptase (TERT), a direct downstream target of MYC/MAX (Fig. 2G) [22]. Based on these in vitro results, MU150 CDX tumor-bearing mice were treated in vivo with carboplatin/paclitaxel and the MYC blocker. MYC blockade overcame chemoresistance in MU150 CDX tumors, whereas MYC blockade alone had no effect on tumor growth (Fig. 2H). A high degree of necrosis (Fig. 2I), overexpression of apoptosis markers cleaved Poly (ADP-ribose) polymerase (cPARP) and cleaved Caspase-3 (cCASP3) in the carboplatin/paclitaxel plus MYC blocker group, and reduced expression of TERT was observed in tumors treated with MYC blocker (Fig. 2J).
Liquid biopsy CDX mouse models generated from non-metastatic NSCLC ptPDX offer valuable drug sensitivity testing platforms, possibly in a time window before patients develop incurable recurrence. They can also be used to study strategies of chemoresistance reversal, such as by MYC inhibition. CDX models also allow identification of distinct single cell heterogeneity that matches non-xenografted NSCLC metastases.