De Novo Purine Metabolism is a Metabolic Vulnerability of Cancers with Low p16 Expression

Abstract

p16 is a tumor suppressor encoded by the CDKN2A gene whose expression is lost in approximately 50% of all human cancers. In its canonical role, p16 inhibits the G1-S-phase cell cycle progression through suppression of cyclin-dependent kinases. Interestingly, p16 also has roles in metabolic reprogramming, and we previously published that loss of p16 promotes nucleotide synthesis via the pentose phosphate pathway. However, the broader impact of p16/CDKN2A loss on other nucleotide metabolic pathways and potential therapeutic targets remains unexplored. Using CRISPR knockout libraries in isogenic human and mouse melanoma cell lines, we determined several nucleotide metabolism genes essential for the survival of cells with loss of p16/CDKN2A. Consistently, many of these genes are upregulated in melanoma cells with p16 knockdown or endogenously low CDKN2A expression. We determined that cells with low p16/CDKN2A expression are sensitive to multiple inhibitors of de novo purine synthesis, including antifolates. Finally, tumors with p16 knockdown were more sensitive to the antifolate methotrexate in vivo than control tumors. Together, our data provide evidence to reevaluate the utility of these drugs in patients with p16/CDKN2Alow tumors as loss of p16/CDKN2A may provide a therapeutic window for these agents.

Significance: Antimetabolites were the first chemotherapies, yet many have failed in the clinic due to toxicity and poor patient selection. Our data suggest that p16 loss provides a therapeutic window to kill cancer cells with widely-used antifolates with relatively little toxicity.

FIGURE 1

Multiple CRISPR KO screens identify nucleotide metabolism genes that are selectively depleted in p16/CDKN2A^low cells. A, Schematic of our CRISPR screens. p16/Cdkn2a wildtype cells were infected with lentiviruses expressing shGFP control (shCont), shp16 (human), or shCdkn2a (mouse). Human and mouse isogenic cell pairs were infected with nucleotide-focused or whole metabolism-focused CRISPR gRNA libraries, respectively, at an MOI of <0.3. After 14 days in culture, gDNA was harvested and sequenced. Analysis of genes included in the “nucleotide metabolism signature” (Supplementary Table S1) identified multiple genes that are negatively enriched in shp16/shCdkn2a vs. shCont in human SKMEL28 (B) and mouse Yumm5.2 (C) melanoma cells. Raw data can be found in Supplementary Tables S3 and S4. D, Comparison of datasets and list of 31 common genes negatively enriched (log₂ fold change <0) in the indicated analyses.

FIGURE 2

p16/CDKN2A negatively correlates with multiple nucleotide metabolism genes, proteins, and metabolites. A–D, SKMEL28 human melanoma cells were infected with lentivirus expressing a shRNA targeting p16 (shp16). shGFP was used as a control (shCont). A, Expression of the 128 nucleotide metabolism gene signature from RNA-seq. Raw data can be found in Supplementary Table S6. B, Polysome fractionation was performed and both the heavy fraction (>2 ribosomes) and total mRNA were sequenced. The ratio of heavy to total was used to assess transcripts with increased translation. Raw data can be found in Supplementary Table S7. C, Genes that are transcriptionally or translationally upregulated in shp16 SKMEL28 cells. D, Expression of the indicated proteins by proteomics. E and F, DepMap data of cutaneous melanoma cell lines. E, mRNA expression of 23 genes identified in the CRISPR screens. F, Protein expression of genes identified in the CRISPR screens. Note only 20 proteins were found in the DepMap data. G, Steady-state metabolite profile of one carbon metabolites and purines.

FIGURE 3

p16/CDKN2A^low cells are more sensitive to inhibitors of nucleotide metabolism. A, Table of inhibitors used in in vitro cell line studies. 1C metabolism = one carbon metabolism. B, SKMEL28 human melanoma cells were infected with lentivirus expressing a shRNA targeting p16 (shp16). shGFP was used as a control (shCont). Cells were treated with the indicated inhibitors and proliferation was assessed by crystal violet staining. IC₅₀ and fold change (shp16 vs. shCont, FC) are indicated. Data from one of 2–3 independent experimental replicates are shown (n = 6). C, Increased drug sensitivity from DepMap data of cutaneous melanoma cell lines with high or low CDKN2A expression. Data are mean ± SD. t test. *, P < 0.05; **, P < 0.01.

FIGURE 4

Multiple antifolates induce apoptosis in p16 knockdown cells. A–D, SKMEL28 human melanoma cells were infected with lentivirus expressing a shRNA targeting p16 (shp16). shGFP was used as a control (shCont). A, Cells were treated with the indicated inhibitors (MTX – 0.17 µmol/L; LTX – 0.12 µmol/L) and cytotoxicity was assessed using IncuCyte Cytotox Green reagent. Data from one of three independent experimental replicates are shown (n = 6). Data are mean ± SD. One-way ANOVA at endpoint. ****, P < 0.0001. B, Cells were treated with the indicated inhibitors (MTX – 0.17 µmol/L; LTX – 0.12 µmol/L) for 72 hours, and apoptosis was assessed using Annexin V/PI (propidium iodide) staining by flow cytometry. Data from one of three independent experimental replicates are shown (n = 6). One-way ANOVA of live cells. ****, P < 0.0001. C, Purine metabolite abundance by mass spectrometry. Cells were treated with methotrexate (MTX; 0.17 µmol/L, 72 hours). Data represent one independent experimental replicate (n = 8). Controls are the same data as shown in Fig. 2G. D, Cells were treated with lometrexol (LTX; 0.17 µmol/L) or methotrexate (MTX; 0.17 µmol/L) for 72 hours, and immunofluorescence analysis for γH2AX foci was performed. Data from one of two independent experimental replicates are shown (n = 3). Data are mean ± SD. One-way ANOVA. ****, P < 0.0001; ns = not significant.

FIGURE 5

shp16 tumors are more sensitive to the antifolate methotrexate. A and B, SKMEL28 human melanoma cells were infected with lentivirus expressing a shRNA targeting GFP (shCont) or p16 (shp16). A total of 10⁷ cells were subcutaneously implanted into athymic nude mice. Mice were treated with vehicle controls or methotrexate (MTX). Individual tumor growth curves in shp16 tumors (A) and shCont tumors (B). Shown are growth rates ± SE. Linear mixed-model group comparisons: shp16 versus shCont P < 0.001; shCont: MTX versus vehicle P = 0.734; shp16: MTX versus vehicle P = 0.031; (shp16: MTX vs. vehicle) versus (Control: vehicle vs. MTX) P = 0.002.

FIGURE 6

De novo purine synthesis and one carbon metabolism genes are associated with worse overall survival in metastatic melanomas. Data from TCGA Skin Cutaneous Melanoma PanCancer Atlas (367 metastatic melanomas). Raw data can be found in Supplementary Table S8. A,De novo purine synthesis and one carbon metabolism genes in TCGA metastatic melanoma samples. Red indicates increased mRNA expression. Blue indicates decreased mRNA expression. Overall survival probability of patients with alterations in de novo purine synthesis and one carbon metabolism genes (B), de novo pyrimidine genes (C), overall nucleotide biosynthesis genes (D), and nucleotide salvage genes (E). log-rank P-value and 95% confidence interval.