In vivo functional screens reveal KEAP1 loss as a driver of chemoresistance in small cell lung cancer

Abstract

Exquisitely chemosensitive initially, small cell lung cancer (SCLC) exhibits dismal outcomes owing to rapid transition to chemoresistance. Elucidating the genetic underpinnings has been challenging owing to limitations with cellular models. As SCLC patient-derived xenograft (PDX) models mimic therapeutic responses, we perform genetic screens in chemosensitive PDX models to identify drivers of chemoresistance. cDNA overexpression screens identify MYC, MYCN, and MYCL, while CRISPR deletion screens identify KEAP1 loss as driving chemoresistance. Deletion of KEAP1 switched a chemosensitive SCLC PDX model to become chemoresistant and resulted in sensitivity to inhibition of glutamine metabolism. Data from the IMpower133 clinical trial revealed ~6% of patients with extensive-stage SCLC exhibit KEAP1 genetic alterations, with activation of a KEAP1/NRF2 transcriptional signature associated with reduced survival upon chemotherapy treatment. While roles for KEAP1/NRF2 have been unappreciated in SCLC, our genetic screens revealed KEAP1 loss as a driver of chemoresistance, while patient genomic analyses demonstrate clinical importance.

Fig. 1.. In vivo cDNA screening in a chemosensitive SCLC PDX model identifies drivers of chemoresistance.

(A) Schematic of in vivo cDNA screen design in the SCLC PDX model SC151. (B) cDNA-level volcano plots of edgeR analysis results showing average log₂ fold change (log₂FC) enrichment versus P0 and −log₁₀ false discovery rate (−log₁₀FDR) for each cDNA across tumors for the saline (SAL) (n = 13) and cis-eto (n = 45) treatment (trx) arms. Each dot depicts one cDNA. MYC family members are displayed in orange, while additional top hits are displayed in blue (log₂FC > 2 and FDR < 0.05 in the cis-eto versus P0 comparison). cDNAs that were also enriched in cis-eto versus saline are denoted by * (FDR < 0.2) or ** (FDR < 0.05). The cis-eto versus saline comparisons are shown in more detail in fig. S1B. (C) Waterfall plots depicting top six hits (log₂FC > 2 and FDR < 0.05 in the cis-eto versus P0 comparison) at the individual tumor level for both the saline and cis-eto versus P0. Each vertical bar depicts one tumor.

Fig. 2.. In vivo CRISPR-Cas9 deletion screens in an SCLC PDX model identifies drivers of chemoresistance.

(A) Schematic of in vivo sgRNA screen. NT, nontargeting. (B) FHSC14 tumor volume kinetics with 3 to 4 weekly cycles of cis-eto. (C) Computational pooling of tumors and percentage of library sgRNAs per pool. Each point depicts one pool generated by randomly selecting the number of tumors specified on the x axis for pooling. For each number of tumors per pool, tumors were randomly pooled 10 times to generate distributions shown. Six tumors per pool was selected for subsequent analyses based on high sgRNA coverage in saline group. Boxes indicate 25th percentile (bottom), median (horizontal line), and 75th percentile (top) of individual points. Vertical lines attached to boxes extend to minimum or maximum values or up to 1.5 times the interquartile range (IQR) from box edges. (D) Gene-level volcano plots of MAGeCK analyses results. Gene-level log₂FC was calculated as median of sgRNA-level log₂FC values. SAGA complex members displayed in blue. FDR < 0.05 for genes with right-side up triangles, 0.05 to 0.20 for genes with upside down triangles, and >0.2 for genes with filled circles (E) Waterfall plots depicting log₂FC enrichment versus P0 of sgRNAs targeting CREBBP and KEAP1, showing top three sgRNAs per gene for all FHSC14 tumor pools. Each vertical bar depicts one sgRNA from one pooled tumor. (F) Relative abundance of KEAP1 sgRNAs in each pooled tumor in cis-eto and saline groups, expressed as log₂FC versus mean abundance in P0. Counts of top three sgRNAs for KEAP1 were summed. Relative abundance of KEAP1 sgRNAs was significantly higher in the cis-eto group [log₂FC = 4.1, FDR < 0.05, based on the MAGeCK analysis from (D)]. Horizontal lines indicate median abundance, while box indicates the IQR. Vertical lines indicate data range, up to 1.5× IQR from the upper or lower quartile.

Fig. 3.. Isogenic PDX model reveals KEAP1 loss as a driver of chemoresistance in SCLC.

(A) Schematic of isogenic PDX model generation and flow cytometry with lentiviral sgRNA expression showing purity of the resulting models. PE, phycoerythrin area; FITC, fluorescein isothiocyanate area. (B) Western blot confirming KEAP1 deletion with sgKEAP1 expression as well as corresponding NRF2 and NQO1 increase. (C) Tumor volume growth curves of nontargeting sgRNA versus sgKEAP1-expressing FHSC14 PDX models over a 21-day treatment period with saline or three weekly cycles of cis-eto. Data are means ± SEM; n = 7 animals total per treatment group. (D) Immunohistochemistry analyses and representative images for pH3 and CC3 to assess proliferation and cell death, respectively, in sgCtrl versus sgKEAP1 FHSC14 PDX models at a 48-hour time point after treatment with either saline or cis-eto. Scale bars, 10 μm. Statistical analyses in (C) and (D) were performed by two-way analysis of variance (ANOVA) with Tukey’s multiple comparisons test. *P < 0.05, **P < 0.01, and ***P < 0.001. GAPDH, glyceraldehyde phosphate dehydrogenase; n.s., not significant.

Fig. 4.. Glutamine metabolism represents a targetable vulnerability for KEAP1-deleted SCLC.

(A) Top six MSigDB Hallmark pathways that are up-regulated in sgKEAP1 FHSC14 tumors relative to sgCtrl tumors, based on GSEA of RNA-seq data after treatment for 48 hours with saline or cis-eto. Point size is scaled to FDR, and point color is scaled to normalized enrichment score (NES). (B) Heatmap showing select gene enrichment for the indicated MSigDB pathways in sgKEAP1 versus sgCtrl FHSC14 tumors treated for 48 hours with saline or cis-eto. (C) Top six KEGG pathways positively enriched in sgKEAP1 relative to sgCtrl tumors, based on Enrichr analysis of RNA-seq data after treatment for 48 hours with saline or cis-eto. The input list of genes was defined on the basis of applying a cutoff of log fold change of >2 and an FDR of <0.05 to the edgeR data. (D) Bar charts showing relative ion counts (normalized to sgCtrl) for reduced GSH and glutamate for FHSC14 sgCtrl and sgKEAP1. (E) Tumor volume growth curves of nontargeting sgRNA versus sgKEAP1-expressing FHSC14 PDX models over a 1-week treatment period with vehicle or twice daily with CB-839 (200 mg/kg). Data are means ± SEM; n = 5 animals total per treatment group. Statistical analysis in (E) was performed by two-way ANOVA with Tukey’s multiple comparisons test. Statistical analysis in (D) was performed by unpaired t test. *P < 0.05, **P < 0.01.

Fig. 5.. KEAP1/NRF2 pathway is altered in a subset of patients with SCLC and associated with SCLC clinical response to chemotherapy treatment.

(A) WES alteration data for patient tumors at baseline in the IMpower-133 trial. Mutations for TP53, RB1, KEAP1, and NFE2L2 are shown. Each vertical bar corresponds to one patient tumor. (B) KEAP1 signature score and WES alteration data for patient tumors at baseline in the IMpower-133 trial organized by the transcription factor (TF) subset. Each vertical bar corresponds to one patient tumor. Gene expression data for transcription factor subsets and genes comprising the KEAP1/NRF2 signature and MYC family members shown. NE, neuroendocrine. (C) PFS curves stratified by the median KEAP1/NRF2 pathway signature score. Dashed curve indicates less than the median KEAP1/NRF2 pathway signature, and solid curve indicates greater than the median KEAP1/NRF2 pathway signature. (D) OS curves stratified by the KEAP1/NRF2 pathway signal. Dashed curve indicates less than the median KEAP1/NRF2 pathway signature, and solid curve indicates greater than the median KEAP1/NRF2 pathway signature. Vertical dashed lines indicate median OS in each group. (E) Forest plot showing HR of IMpower133 patients in the placebo + carboplatin + etoposide arm stratified by KEAP1/NRF2 signature low versus high status. The expression percentile cutoff threshold used to determine signature high versus signature low are shown in the population column, and the number (N) and percentage of patients belonging to the signature-high group with the given cutoff are shown. Higher OS or PFS HR indicate increased likelihood of PFS or OS event in the signature-high population. WT, wild type; Inf, infinity.