Somatic CRISPR tumorigenesis and multiomic analysis reveal a pentose phosphate pathway disruption vulnerability in MPNSTs

Abstract

Malignant peripheral nerve sheath tumors (MPNSTs) are aggressive and chemo-resistant sarcomas with poor survival rates. Loss of CDKN2A or P53 following NF1 disruption is a key event in MPNST development. Here, we used CRISPR-Cas9 somatic tumorigenesis in mice to identify transcriptomic and metabolomic features distinguishing CDKN2A- versus P53-deleted MPNSTs. Convergent, multiomic analyses revealed that CDKN2A-deleted MPNSTs are especially dependent on the pentose phosphate pathway (PPP) and NADPH metabolism for growth and viability. Disruption of glucose-6-phosphate dehydrogenase (G6PD), the PPP rate-limiting enzyme, slowed CDKN2A-deleted MPNST growth and sensitized MPNSTs to standard-of-care chemotherapy. Knockdown of the redox-regulated transcription factor NRF2 slowed MPNST growth and decreased G6PD transcription. Analysis of patient MPNSTs identified a NRF2 gene signature correlating with tumor transformation. Furthermore, G6PD and NRF2 expression in PanCancer TCGA samples correlates with patient survival. This work identifies NRF2-PPP dependency as a targetable vulnerability in these difficult-to-treat MPNSTs, particularly in the NF1/CDKN2A-deleted majority.

Fig. 1.. Derivation of primary tumor cell lines for direct comparison of tumor suppressor roles in MPNST.

(A) Schematic of development of Nf1/p53-deleted (Nf1/p53 dKO) and Nf1/Cdkn2a-deleted (Nf1/Cdkn2a dKO) murine orthotopic CRISPR-Cas9 primary MPNST cell lines for multiomic analysis. Schematic made using artwork from www.vecteezy.com (https://www.vecteezy.com/free-vector/mouse). (B) Ad5 CRISPR-Cas9 sgRNA viral constructs injected into sciatic nerve for primary MPNST tumorigenesis (A). Schematic made using Inkscape software. (C) Kaplan-Meier curve comparing primary tumor initiation from time of adenovirus injection (day 0) of 129 Sv/Jae murine strain injected with Nf1/p53 AdV5-CMV-Cas9 (n = 18 biological replicates) and Nf1/Cdkn2a AdV5-CMV-Cas9 (n = 20 biological replicates). (D) Days for tumors [in (C)] to triple in volume postinitiation. Data presented as mean ± SEM. P value was determined by log-rank test [in (C)] and two-tailed unpaired t test [in (D)].

Fig. 2.. Differential tumor suppressor loss alters transcriptomes of MPNST.

(A) RNA-seq analysis of primary cell lines: GSEA pathway enrichment analysis, top cancer hallmark pathways, (blue) Nf1/p53 dKO (n = 3) versus (red) Nf1/Cdkn2a dKO (n = 3), each cell line run in triplicate. (B) Leading edge analysis of RNA-seq data. UV, ultraviolet; TNF-α, tumor necrosis factor–α; NF-κB, nuclear factor κB. (C) Top enriched genes in GSEA hallmark ROS gene set in Nf1/Cdkn2a dKO cell lines (n = 3). (D and E) RT-qPCR validation of RNA-seq (D) tumor-derived cell lines, Nf1/p53 dKO (n = 3) and Nf1/Cdkn2a dKO (n = 3), and (E) whole tumor lysates, Nf1/p53 dKO (n = 3) and Nf1/Cdkn2a dKO (n = 3). Data presented as geometric mean and geometric SD [in (D) and (E)], multiple unpaired t tests followed by Holm-Sidak correction [in (D) and (E)]. P = 0.05.

Fig. 3.. Loss of Cdkn2a results in enrichment of the PPP.

(A) GC and LC metabolomics unsupervised clustered heatmap, scaling (z-score). IMP, inosine monophosphate; CDP, cytidine diphosphate; HMB, beta-hydroxy-beta-methylbutyrate; NMN, β-nicotinamide mononucleotide. (B) GC and LC metabolomics enriched metabolite set analysis. (C) Schematic of oxidative PPP:G6PD (glucose-6-phosphate dehydrogenase), PGLS (6-phosphogluconolactonase), PGD (6-phosphogluconate dehydrogenase), NADP⁺ (oxidized nicotinamide adenine dinucleotide phosphate), NADPH (reduced nicotinamide adenine dinucleotide phosphate), GSH (reduced glutathione), and GSSG (glutathione disulfide). TNBC, triple-negative breast cancer; AGAT, L-arginine:glycine amidinotransferase; GAMT, Guanidinoacetate methyltransferase; HHH, hyperornithinemia-hyperammonemia-homocitrullinuria. (D) Volcano plot showing interaction of fold changes (FC) and P values of metabolites, fold changes represent Nf1/Cdkn2a dKO mean relative to Nf1/p53 dKO mean for each metabolite, G6P (glucose 6-phosphate), P = 0.05 indicated by a dotted line. (E and G) PPP and redox metabolites relative abundance, normalized to Nf1/p53 dKO average for each metabolite. 6PG, 6-phosphogluconate; R5P, ribose 5-phosphate; S7P, sedoheptulose 7-phosphate; NADP, oxidized nicotinamide adenine dinucleotide phosphate. (F) PPP transcripts, RNA-seq read count values normalized to Nf1/p53 dKO average for each transcript. Nf1/p53 dKO (n = 3) and Nf1/Cdkn2a dKO (n = 3), each cell line run in quadruplicate (independent samples) [in (A) to (C), (E), and (G)] and triplicate (independent samples) [in (F)]. Data presented as mean ± SEM [in (E) to (G)]. Unpaired t tests [in (E) and (G)]. Adjusted P values were determined by differential expression analysis followed by Benjamini and Hochberg’s approach for false discovery rate (FDR) [in (F)]. P = 0.05.

Fig. 4.. Loss of Cdkn2a results in increased vulnerability to PPP inhibition.

(A) Representative dose-response curves of Nf1/p53 dKO (n = 3) and Nf1/Cdkn2a dKO (n = 3) cell lines treated with DHEA for 48 hours. (B) Clonogenic survival assay post-DHEA (75 μM) treatment, each data point is the average of one cell line in triplicate, n = 3 per genotype, each cell line normalized to its own DMSO control. (C) Western blot of Nf1/p53 dKO and Nf1/Cdkn2a dKO cell lines 72 hours posttransfection G6pdx targeted siRNA transfection, β-actin used as loading control. (D) Representative relative viability of cell lines 96 hours post-G6pdx targeted siRNA transfection. NT, nontargeting. (E) Representative relative viability of Nf1/Cdkn2a dKO cell lines 72 hours posttransfection with G6pdx targeted siRNA +/− 5 mM NAC. (F) Representative doxorubicin IC₅₀ curves, Nf1/p53 dKO (n = 2) and Nf1/Cdkn2a dKO (n = 2) cell lines treated for 48 hours. (G and H) Representative relative viability of (G) Nf1/p53 dKO and (H) Nf1/Cdkn2a dKO treated with 100 μM DHEA, 1 μM Dox (doxorubicin), or combination for 48 hours. (I) Representative dose-response curves of Nf1/Cdkn2a dKO cells transfected with G6pdx targeted siRNAs and treated with doxorubicin 72 hours posttransfection for 48 hours. Curves fitted using nonlinear regression, option [inhibitor] vs normalized response-variable slope in GraphPad Prism. Shaded areas represent 95% confidence intervals [in (A), (F), and (I)], independently derived (different tumors) primary cell lines within each genotype represented by line patterns [in (A) and (F)], representing specific siRNA treatments [in (I)]. Genotype average IC₅₀ and individual cell line IC₅₀s denoted in (A), (E), and (I), IC₅₀ confidence intervals shown in fig. S5 [for (A) and (F)]. Data presented as mean ± SEM, two-tailed unpaired t test [in (B)]. Data presented as mean ± SD, one-way analysis of variance (ANOVA) with Holm-Sidak correction [in (D), (E), (G), and (H)]. P = 0.05.

Fig. 5.. The NRF2/G6PD axis drives Nf1/Cdkn2a-deleted MPNST.

(A) Nfe2l2 transcript levels, RNA-seq read count values normalized to Nf1/p53 dKO average. (B) RT-qPCR Nfe2l2 transcript levels in Nf1/p53 dKO and Nf1/Cdkn2a dKO cell lines and whole tumor lysates, normalized to Nf1/p53 dKO average. (C) Immunofluorescence validation of NRF2 protein knockdown 72 hours posttransfection with Nfe2l2 targeted siRNAs, representative images. Scale bars, 25 μm. DAPI, 4′,6-diamidino-2-phenylindole. (D) Relative viability of Nf1/p53 dKO and Nf1/Cdkn2a dKO cell lines 96 hours posttransfection of Nfe2l2 targeted siRNAs. (E) RT-qPCR, G6pdx transcript levels in cells 48 hours posttransfection of Nfe2l2 targeted siRNA. (F) Relative viability of Nf1/p53 dKO and Nf1/Cdkn2a dKO cell lines 96 hours postcombination transfection of Nfe2l2 and G6pdx targeted siRNAs. (G) Model schematic of the proposed relationship between Cdkn2a gene products, ARF (p19) and p16, and the NRF2/PPP axis in regulating ROS levels in Nf1/Cdkn2a-deleted MPNSTs. Schematic made using Inkscape software. Data presented as mean ± SD, Nf1/p53 dKO (n = 3) and Nf1/Cdkn2a dKO (n = 3), each cell line run in triplicate (independent samples), adjusted P values were determined by differential expression analysis followed by Benjamini and Hochberg’s approach for FDR [in (A)]. Data presented as geometric mean and geometric SD, Multiple unpaired t tests followed by Holm-Sidak correction [in (B) and (E)]. Data presented as mean ± SD, one-way ANOVA followed by Holm-Sidak correction [in (D) and (F)]. P = 0.05.

Fig. 6.. Patient MPNSTs have an active NRF2/G6PD axis.

(A) Schematic: RNA-seq analysis of patient neurofibromas (n = 23) and MPNST (n = 23) samples. (B) NRF2 core target gene signature in human MPNST/neurofibroma RNA-seq dataset [from (A), NRF2 core target genes (orange), non-NRF2 core target genes (gray), oxidative PPP (highlighted in pink), nonoxidative PPP (highlighted in blue), and PPP recycling (highlighted in yellow)]. (C) Human MPNST cell line S462 treated with 100 μM DHEA, 1 μM doxorubicin, or combination for 48 hours. (D) NFE2L2 and G6PD mRNA expression in soft tissue sarcoma cell lines. TPM, transcripts per million. (E and F) TCGA PanCancer Atlas, effect of mRNA expression levels on overall survival, NFE2L2 high (n = 331) and low (n = 2426) [in (E)] and G6PD high (n = 1869) and low (n = 134) [in (F)]; high expression groups were determined by a z-score of >2, and low expression groups were determined by a z-score of <−2 [in (E) and (F)]. Data presented as mean log₂ fold change ± SD, analyzed by mean MPNST expression compared to mean neurofibroma expression [in (B)]. Data presented as mean ± SD, one-way ANOVA followed by Holm-Sidak correction [in (C)]. Data analyzed using the Broad Cancer Dependency Map Project portal, each data point represents an individual cell line (n = 44), Pearson correlation (r) [in (D)]. Log-rank tests were conducted in cBioportal [in (E) and (F)]. P = 0.05.