Cytidine diphosphate diacylglycerol synthase 2 is a synthetic lethal target in mesenchymal-like cancers

Abstract

Synthetic lethal interactions (SLIs) based on genomic alterations in cancer have been therapeutically explored. We investigated the SLI space as a function of differential RNA expression in cancer and normal tissue. Computational analyses of functional genomic and gene expression resources uncovered a cancer-specific SLI between the paralogs cytidine diphosphate diacylglycerol synthase 1 (CDS1) and CDS2. The essentiality of CDS2 for cell survival is observed for mesenchymal-like cancers, which have low or absent CDS1 expression and account for roughly half of all cancers. Mechanistically, the CDS1-2 SLI is accompanied by disruption of lipid homeostasis, including accumulation of cholesterol esters and triglycerides, and apoptosis. Genome-wide CRISPR-Cas9 knockout screens in CDS1-negative cancer cells identify no common escape mechanism of death caused by CDS2 ablation, indicating the robustness of the SLI. Synthetic lethality is driven by CDS2 dosage and depends on catalytic activity. Thus, CDS2 may serve as a pharmacologically tractable target in mesenchymal-like cancers.

Fig. 1. CDS1–2 form a common cancer-associated SLI pair.

a, Diagram depicting synthetic lethality based on lack of expression in cancer. b, Schematic depicting the computational analysis to identify cancer-specific SLIs. c, The correlation (Pearson’s r) of RNA, proteomic, mutation or methylation anchor gene data and gene dependency data in cancer cell lines for previously functionally validated synthetic lethal pairs. The rank of the known target gene partners was plotted for each data type. The significance cut-off was the top 5% enrichment (rank-based). n (cancer cell lines) = 913 (RNA), 299 (protein), 986 (damaging mutation) and 513 (methylation). d, Top predicted synthetic lethal pairs, named by their target gene and anchor gene, respectively. The y axis presents the correlation (Pearson’s r) between the anchor gene expression and target gene dependency (DepMap). The x axis presents the average reduction in anchor expression in patient tumor samples compared with patient healthy tissue samples (TCGA). Bonferroni-corrected P values from Pearson correlation tests were used to determine significance for synthetic lethality (P < 0.05). For the top synthetic lethal interactions, the significance of cancer specificity was assessed in Extended Data Fig. 1d,e. n = 913 cancer cell lines, 10,005 patient samples (9,264 tumor and 741 healthy). e, For each cancer lineage, the percentage of ΔCDS2 lethal cancer cell lines plotted against the percentage of low CDS1-expressing cancer cell lines (DepMap). n = 906 cancer cell lines. f, Probability of survival for patients with the highest and lowest quartiles of CDS1 expression in patient tumor samples (TCGA). The hazard ratio for the CDS1-low quartile was 1.90 [1.715–2.106] compared with the CDS1-high quartile. The log-rank Mantel–Cox test was used. n = 10,163 patients. g, Calibration gene-normalized RNA expression in healthy tissue samples (GTEx), cancer cell lines (DepMap), ΔCDS2 lethal cancer cell lines (DepMap) and ΔCDS2 nonlethal cancer cell lines (DepMap), presented in violin plots for the genes B2M, E2F1, CDKN2B and CDS1. The three dotted or dashed lines in each violin plot represent the quartile cut-offs. For significant differences, the percentage change in average TPM was indicated, with one overall average for controls. The Kruskal–Wallis test followed by Dunn’s test was used. n = 17,382 healthy tissue samples and 913 cancer cell lines (657 ΔCDS2 nonlethal and 256 ΔCDS2 lethal). h, Calibration gene-normalized CDS1 protein levels in matched healthy lung and lung tumor samples. The three dotted or dashed lines in each violin plot represent the quartile cut-offs. The Mann–Whitney U test was used. n = 110 tumor samples and 102 adjacent healthy lung samples. In f, g and h: ***P < 0.001. NS, not significant. Source data

Fig. 2. CDS1 and CDS2 are synthetic lethal across cancer types.

a, Plot depicting synthetic lethality upon CRISPR perturbation of CDS2 (ΔCDS2 lethality) and CDS1 RNA expression for cancer cell lines (DepMap). Cancer cell lines were categorized by CDS1 RNA levels and cell lines used for validation are marked. The data distribution is depicted by a ΔCDS2 lethality histogram (top) and a CDS1 expression histogram (right). n = 913 cancer cell lines. b, Plot depicting CDS1 RNA expression as measured by RNA sequencing (RNA-seq; DepMap) or qPCR analysis. The correlation (Pearson’s r) and associated P value were added. There was an average of two independent experiments with each n = 4 replicates for 8 cancer cell lines. c, Diagram depicting the method for quantifying lethality on CRISPR perturbation using fluorescent tracker cells. The mCherry-positive percentage was quantified by flow cytometry. d, Graph depicting cumulative lethality 14 d after CRISPR perturbation of CDS2 using tracker cells (c). CDS2 sgRNAs, positive control sgRNA (RPL19) and negative control sgRNA (sgControl) were included. Two-way ANOVA followed up by Dunnett’s test was used. n = 3 replicates for 7 cancer cell lines with 2 CDS2 sgRNAs; see the 8-d measurement in Extended Data Fig. 2b. There is an independent experiment repeat in e (three cancer cell lines) and Fig. 3b (four cancer cell lines); the latter also includes an additional CDS1-high cancer cell line. e, For two CDS1-negative cell lines (NCI-H2030 and SK-MEL-2), CDS1 or GFP (control) was ectopically expressed and the 14-d ΔCDS2 lethality was determined using tracker cells (c). For one CDS1-high cell line CDS1 (sgCDS1 versus sgControl) was perturbed by CRISPR and the 8-d ΔCDS2 lethality was determined using tracker cells (c). Two-way ANOVA followed by Tukey’s test was used. n = 3 replicates for 3 cancer cell lines. A431 and NCI-H2030/SK-MEL-2 are separate experiments. f, Left: diagram depicting method for quantifying synthetic lethality using ectopic CDS1 or GFP expression. The GFP-positive percentage was quantified by flow cytometry. Right: graph depicting quantification of the 14-d synthetic lethality in melanoma. Two-way ANOVA followed by Šidák’s test was used. n = 3 replicates for 4 cancer cell lines; see the 7-d measurement in Extended Data Fig. 2d. g, Graph depicting snapshot of the total protein-normalized level of cleaved caspase-3 in ΔCDS2 or control samples of BLM and SK-MEL-2 cell lines, as determined by quantitative western blotting. Two-way ANOVA followed by Šidák’s test was used. n = 3 replicates from independent lentiviral transductions for 2 cancer cell lines. h, Left: diagram depicting the in vivo variant of the method for quantifying synthetic lethality. Right: graph depicting quantification of the in vivo synthetic lethality in two melanoma cell-line models 17 d after subcutaneous melanoma tumor inoculation. Tumors were analyzed by flow cytometry. Two-way ANOVA followed by Šidák’s test was used. n = 6 NOD-scid Il2rγ-null mice per group each for 2 cancer cell lines. In b and d–h: ***P < 0.001. Source data

Fig. 3. No common escape mechanism for ΔCDS2 lethality.

a, Diagram depicting the method used to screen for ΔCDS2 lethality escape mechanisms. One CDS1-high cancer cell line and four CDS1-negative cancer cell lines were screened. b, The lethality of the screen cells between 15 d and 25 d. Lethality was quantified using fluorescent tracker cells as depicted in Fig. 2c. Two-way ANOVA followed by Šídák’s test was used. n = 3 replicates for 5 cancer cell lines. c, Graph depicting the screen hits identified in three of four or all four CDS1-negative cancer cell lines. The effect size was calculated using the average ΔCDS2 lethality in the lethality assays with screen cells (b) and the fold-change enrichment of the second-best guides for each gene. Negative binomial modeling per sgRNA, followed by alpha-Robust Rank Aggregation per gene and Benjamini–Hochberg correction for the screens, was used to determine significance (P < 0.05). n = 1 for 4 cancer cell lines at 200× coverage with 4 sgRNAs per gene and 1,000 sgControls in the CRISPR library. d, Graph depicting Pearson’s r between ΔCDS2 lethality and other features of cancer cell lines corrected for CDS1 expression (DepMap). The correlation between ΔCDS2 lethality and CDS1 expression was added (red line). The features included were RNA expression, damaging mutations and hotspot mutations. Bonferroni-corrected P values from Pearson correlation tests were used. The gray box indicates the nonsignificant features. n (cancer cell lines) = 1,021. In b: ***P < 0.001. Source data

Fig. 4. Mesenchymal-like cancers depend on CDS2 for PI synthesis.

a, Method for phenotyping cancer cell lines that are ΔCDS2 lethal and CDS1-low compared to those that are ΔCDS2 nonlethal and CDS1-high (DepMap) (illustrative). b, Genome-wide calculated Pearson correlations using the method depicted in a (DepMap). Representative genes enriched on either side are marked. Bonferroni-corrected P values from Pearson correlation tests were used to determine the significance cut-off. n = 913 cancer cell lines. c, Plot depicting CDS1 RNA expression on ZEB1 CRISPR perturbation as measured by qPCR analysis. The dashed line indicates the baseline of no change (1-fold change). Two-way ANOVA followed by Dunnett’s test was used for each cancer cell line. n = 4 replicates for 2 cancer cell lines in 2 independent experiments with each 2 sgZEB1 and 1 sgControl. d, For each cancer lineage, the percentage of ZEB1-high cancer cell lines (DepMap) was plotted against the percentage of CDS1-high or CDH1-high cancer cell lines. A linear regression, the associated Pearson correlation and associated P values were added. n = 906 cancer cell lines. e, Diagram depicting the characteristics of cancer cell lines with and without the CDS1–CDS2 SLI. CDH1 and VIM are representative epithelial and mesenchymal markers, respectively. f, Diagram depicting the core CDS pathway. The lethality on perturbation (color) and lineage-specific expression patterns (gray boxes on the left) of each enzyme in the pathway are indicated (DepMap). Gene dependency is represented by CERES. n = 913 cancer cell lines. g, Correlation between ΔCDS2 lethality and ΔCDIPT lethality, ΔPIK3CA lethality, PIK3CA inhibitor lethality (alpelisib) or the pan-PI3K inhibitor lethality (copanlisib) in solid cancer cell lines (DepMap). Bonferroni-corrected P values from Pearson correlation tests were used. n (cancer cell lines) = 865 (CDIPT and PIK3CA), 428 (copanlisib) and 441 (alpelisib). h, Rescue of synthetic lethality on supplementation with PI or phosphatidylcholine (PC, control). The method to quantify synthetic lethality was introduced in Fig. 2f. PI and PC were complexed with lipid-free bovine serum albumin (BSA). Treatment was with 0.25 mM lipid for 7 d, started 7 d after transduction. Two-way ANOVA followed by Dunnett’s test was used. n = 3 replicates for 2 cancer cell lines. In c, d, g and h: ***P < 0.01, ***P < 0.001. Source data

Fig. 5. Cholesterol production is induced on CDS1+CDS2 ablation.

a, Diagram depicting the samples generated for quantifying proteins and lipids (Fig. 6). One CDS1-high cancer cell line and four CDS1-negative cancer cell lines were analyzed. b, Average ΔCDS2 protein changes for the four CDS1-negative cancer cell lines (y axis) plotted against the ΔCDS2 protein changes in the CDS1-high cancer cell line (x axis). The significance was determined using the Student’s t-test. n = 4 replicates for 5 cancer cell lines. An independent experiment repeat using a second independent instrument is shown in Extended Data Fig. 5c for SK-MEL-2. c, CDS1 protein levels plotted against CDS2 protein levels for a panel of ten cancer cell lines. CDS1 expression from public RNA-seq was added as in Fig. 2b (DepMap) with additional data for BLM and SK-MEL-147 (accession no. SRP132830). For each axis, the relative abundance to the highest cancer cell line included was plotted. Data were normalized to the median levels of calibration proteins and the mean of the experiments was used. CDS1 was detected in all MCF7 and A431 replicates, but no other replicates. n = 3 and 4 replicates from 2 independent experiments. Source data

Fig. 6. Lipid homeostasis requires either CDS1 or CDS2 expression.

a, Five ΔCDS2 lipid class changes presented (full results in Extended Data Fig. 6a). The asterisks indicate the number of CDS1-negative cell lines with at least P < 0.01 significant lipid alterations compared with the CDS1-high cell line. Two-way ANOVA followed by Šidák’s test was used. n = 4 replicates for 5 cancer cell lines. b, Cell areas staining positive for the lipid compartment dye BODIPY in live-cell imaging. Two-way ANOVA followed by Šidák’s test was used. n = 4 replicates for 2 cancer cell lines; see d for an independent experiment for one cancer cell line. c, Representative combined light and transmission electron microscopy image of an sgControl and an sgCDS2 SK-MEL-2 cell stained for lipids (Nile Red) and DNA (Hoechst). n = 1 for 1 cancer cell line. Two independent light microscopy experiments were performed for quantification (b and d). d, Lipid droplet quantification on supplementation with 0.25 mM PI or PA in SK-MEL-2. PI and PA were complexed with lipid-free BSA. Treatment was started at 8 d and refreshed at 11 d. Two-way ANOVA followed by Šidák’s test was used. n = 4 replicates for 1 cancer cell line. See b an for independent experiment repeat without lipid treatments, also covering one additional cancer cell line. e, Diagram depicting lipid compartments and the CDS2 pathway. The ΔCDS2 lipid class changes in the four CDS1-negative cell lines are marked by arrows. DGKs, diacylglycerol kinases. In a the significance was indicated by one asterisk for each cell line with P < 0.01. In b and d: ***P < 0.001. Source data

Fig. 7. CDS2 dosage and catalytic activity govern synthetic lethality.

a, The relationship between CDS1 expression and ΔCDS2 lethality. Deciles for cancer cell line CDS1 expression were plotted against ΔCDS2 lethality using DepMap data, with variation in ΔCDS2 lethality indicated by the interquartile range. A sigmoid curve was fitted through the CDS1 expression deciles with r = 0.998. Hill’s equation yielded a log₂(TPM + 0.0625) of 1.273 as the level of CDS1 expression, with 50% of maximum synthetic lethality among the cancer cell line collection (dashed line). For the totals above and below this level, numbers and percentages are shown on top, together with a histogram of the cancer cell line distribution. n = 913 cancer cell lines (DepMap). b, Graph depicting quantification of the 10-d synthetic lethality in BLM using the method for quantifying synthetic lethality with ectopic CDS1 or GFP expression described in Fig. 2f. Two-way ANOVA followed by Šidák’s test was used. n = 3 replicates for 1 cancer cell line for 3 shCDS2. c, Tumor growth after subcutaneous inoculation of control and CDS2-perturbed BLM melanoma cells. Two-way ANOVA was used (n = 9 NOD-scid Il2rγ-null mice per group for 1 cancer cell line). d, For two CDS1-negative melanoma cancer cell lines CDS1, GFP (control), CDS2–GFP or CDS2–GFP Asp384Ala was ectopically expressed and the 14-d ΔCDS2 lethality was determined using tracker cells (Fig. 2c). Two-way ANOVA followed by Šidák’s test was used. n = 3 replicates for 2 cancer cell lines. The CDS2–GFP and CDS2–GFP Asp384Ala cell line variants were each generated in duplicate via independent transductions (no. 1, 2). ΔCDS2 lethality in CDS1 or GFP variants was replicated in one independent experiment for SK-MEL-2 (Fig. 2e) and two independent experiments for BLM (Fig. 2f: in vitro; Fig. 2h: in vivo). e, A diagram summarizing the mechanistic aspects of CDS1–CDS2 synthetic lethality. On the left, upregulated cholesterol proteins are listed. On the right, buildup of the substrate lipids (PA, DAG) and of the lipid droplet lipids (CEs, TAGs) and depletion of the product PI is indicated. In b–d: ***P < 0.001. Source data

Extended Data Fig. 1. CDS1-2 form a common cancer-associated SLI pair.

a, Enrichment of previously functionally validated synthetic lethal interactions (FAM50A-FAM50B, RPP25L-RPP25, EIF1AX-EIF1AY, DDX3X-DDX3Y, DNAJC19-DNAJC15, TTC7A-TTC7B, MCL1-BCL2L1) for different sample sizes. The top 100 Pearson correlations between RNA and dependency data were plotted using all DepMap cancer cell lines or cancer type subsets. n (cancer cell lines) = 913 (all), 122 (lung), 70 (CNS), 63 (skin), 40 (breast). b, Top: plot depicting ΔCDS2 lethality and CDS1 RNA expression for cancer cell lines (DepMap). Bottom: plot depicting ΔFAM50A lethality and FAM50B RNA expression for cancer cell lines (DepMap). n = 913 cancer cell lines. c, Replication of the analysis and method of plotting presented in Fig. 1d which used 21Q2 CERES data using 23Q4 Chronos data (DepMap). n = 1,014 cancer cell lines, 10,005 patient samples (9,264 tumor). d, The average fold change in patient tumor samples compared to patient healthy samples (TCGA) for the anchor genes of the top ten predicted synthetic lethal pairs in Fig. 1d. n = 10,005 patient samples (9,264 tumor). e, The RNA expression in patient tumor and healthy samples for downregulated anchor genes in panel d (TCGA). Dotted lines represent the quartile cut-offs. Kruskal-Wallis test followed by Dunn’s test was used. n = 10,005 patient samples (9,264 tumor). f, For each cancer type the relative cancer mortality (SEER) was plotted against the frequency of ΔCDS2 lethal cell lines (DepMap). The fraction of ΔCDS2 lethal cell lines within each type is categorically depicted. n (cancer cell lines) = 906 (40 breast, 122 lung, 97 blood, 92 brain, 64 skin, 45 soft tissue). g, For the top lineages from panel f, the calibration gene-normalized CDS1 expression in healthy tissue samples (GTEx), cancer cell lines (DepMap), ΔCDS2 lethal cancer cell lines and ΔCDS2 nonlethal cancer cell lines is presented. Dotted lines represent the quartile cut-offs. Kruskal-Wallis test followed by Dunn’s test was used. n (healthy) = 578 (lung), 2,642 (brain), 929 (blood), 1,809 (skin). n (cancer) = 122 (lung, 35 ΔCDS2 lethal), 75 (brain, 25 ΔCDS2 lethal), 48 (blood, 35 ΔCDS2 lethal), 63 (skin, 24 ΔCDS2 lethal). In e and g: ^***P<0.001. ns, not significant. Source data

Extended Data Fig. 2. CDS1 and CDS2 are synthetic lethal across cancer types.

a, As an example, representative histograms for the flow cytometry data used to calculate lethality for NCI-H2030 in Fig. 2d are presented. b, The 8-d results for the experiment in Fig. 2d. Two-way ANOVA followed up by Dunnett’s test was used. n = 3 replicates for 7 cancer cell lines with 2 CDS2 sgRNAs. c, The percentage of cancer cells with frameshift upon sgCDS2 compared to sgControl determined using TIDE. n = 3 replicates for 5 cancer cell lines. d, The 7-d results for the experiment in Fig. 2f. Two-way ANOVA followed by Šidák’s test was used. n = 3 replicates for 4 cancer cell lines. e, Graph depicting a snapshot of the total protein normalized level of cleaved caspase-3 in apoptotic or control samples of BLM and SK-MEL-2 cell lines, as determined by quantitative western blotting (Abby). Apoptosis was induced by TPCA-1 + TNF for BLM cells and staurosporine for SK-MEL-2 cells. Two-way ANOVA followed by Šidák’s test was used. n = 3 replicates for 2 cancer cell lines. f, Representative histograms for the flow cytometry data used to calculate synthetic lethality in Fig. 2h. g, Growth curves for the experiment in Fig. 2h. Note that the tumor growth curves are derived from cell mixes including both GFP-expressing and ectopically CDS1-restored cells. Two-way ANOVA was used. Mean is presented by dark lines. n = 6 NOD-scid Il2rγ-null mice per group for 2 cancer cell lines. In b-e and g: ^*P<0.05, ^**P<0.01, ^***P<0.001. Source data

Extended Data Fig. 3. No common escape mechanism for ΔCDS2 lethality.

a, Top: Colony formation assays with adherent cell lines used for the screen (crystal violets). Bottom: Cell survival assay with the suspension screen cell line (cell counts). n = 1 for 5 cancer cell lines, an independent experiment was performed for quantification (Fig. 3b). b, The average ΔCDS2 changes for knockouts of each other gene were plotted for the four CDS1-negative cancer cell lines against the CDS1-high cell line. The dotted line corresponds to the ΔCDS2 change required to compensate for the average lethality quantified in Fig. 3b. n = 1 for 5 cancer cell lines at 200x coverage with 4 sgRNA per gene and 1,000 sgControls in the CRISPR library. c-f, Lethality rescue plotted against the Benjamini-Hochberg corrected P values from MAGeCK for each gene in the separate screens. Each of these panels presents the results for a different cancer cell line. The rescue was calculated using average lethality quantified in Fig. 3b. The two genes significant in three cell lines (LCMT1 & EP300) and the two significant genes in two cell lines with the strongest rescue (PPP6C & PTPN11) were marked. Note that outliers above 100% were not plotted. n = 1 for 4 cancer cell lines at 200x coverage with 4 sgRNA per gene and 1,000 sgControls in the CRISPR library. g, As a quality control, each control arm was compared with the original library to detect essential genes (true positives) over non-essential genes (true negatives). The resulting area under the receiver operator curve (ROC-AUC) values for each screen are presented. A ROC-AUC above 90% of maximum is considered outstanding for CRISPR screens. n = 1 for 5 cancer cell lines at 200x coverage with 4 sgRNA per gene and 1,000 sgControls in the CRISPR library. h, Left: Plot depicting ΔCDS2 lethality and CDS1 RNA expression for cancer cell lines (DepMap). Cancer cell lines used for validation are marked. Right: The 14-d lethality of the marked cancer cell lines. Lethality was quantified using fluorescent tracker cells as depicted in Fig. 2c. Two-way ANOVA followed by Šidák’s test was used. n = 1,014 cancer cell lines (DepMap, left) or 4 replicates for 4 cancer cell lines (right). In h: ^***P<0.001. Source data

Extended Data Fig. 4. Mesenchymal-like cancers depend on CDS2 for PI synthesis.

a, CDH1 and VIM expression in ΔCDS2 lethal, ΔCDS2 nonlethal, CDS1-low (bottom quartile) and CDS1-high (top quartile) cancer cell lines (DepMap). Mann-Whitney U test was used. n = 913 cancer cell lines (657 nonlethal, 256 lethal, 345 CDS1 low, 352 CDS1 high). b, The percentage of CDS1-high cancer cell lines (DepMap) was plotted against the percentage of VIM-high or CDH1-high cancer cell lines by type. A linear regression and Pearson correlation test were added. n = 906 cancer cell lines. c, CDS1 promoter methylation in all cancer cell lines and the blood subtype (DepMap). Histograms show the distribution of CDS1 promoter methylation (top). Mann-Whitney U test was used. n = 792 cancer cell lines (72 blood). d, The gene set enrichment analysis (GSEA) results for Hallmark epithelial to mesenchymal transition. Results are from a GSEA on genome-wide CDS1 coexpression (Pearson’s r) using DepMap and TCGA data. The Benjamini-Hochberg corrected GSEA P values were used. n = 913 cancer cell lines, 11,069 patient samples. e, For 16 cancer types the calibration gene-normalized RNA expression in healthy tissue samples (GTEx) and cancer cell lines (DepMap) was compared. The cut-off for low expression of CDS1 or CDH1 expression was the bottom-tertile boundary for expression in healthy tissue samples of the same type (right). CDH1 and VIM expression were plotted as a percentage of the highest expressing cancer type included (left). Kruskal-Wallis test followed by Dunn’s test was used by type. The Pearson correlation test was used across types. n (healthy/cancer) = 2,642/70 (brain), 929/48 (blood), 653/8 (thyroid), 89/25 (kidney), 1,809/63 (skin), 226/22 (liver), 180/53 (ovarian), 578/122 (lung), 142/32 (uterus), 19/13 (cervix), 359/29 (stomach), 328/42 (pancreatic), 459/40 (breast), 1,445/28 (esophagus), 966/50 (colorectal), 245/5 (prostate). f, Left: the method for quantifying the dead cell fraction using the DNA-intercalating dye DRAQ7. Right: rescue of ΔCDS2 cell death upon overnight supplementation with 2 mM phosphatidylinositol (PI) started 11 d after transduction of BLM cells. One-way ANOVA followed by Šidák’s test was used. n = 3 replicates for 1 cancer cell line. In a-f: ^*P<0.05, ^**P<0.01, ^***P<0.001. Source data

Extended Data Fig. 5. Cholesterol production is induced upon CDS1+CDS2 ablation.

a, Volcano plots presenting the ΔCDS2-induced protein changes in five cancer cell lines. In addition to CDS2, the top and bottom five proteins significant in at least three out of four CDS1-negative cancer cell lines were marked. n = 4 replicates for 5 cancer cell lines. b, The top ten significantly enriched GO terms among ΔCDS2-changed proteins in the four CDS1-negative cancer cell lines. Benjamini-Hochberg corrected P values from Stringdb were used. n = 4 replicates for 4 cancer cell lines. c, Graph depicting a second independent experiment for SK-MEL-2 using a different instrument (y axis, Astral). Data is plotted against the experiment presented in Fig. 5b (x axis). All plotted changes were significant in both experiments. Only the 13 cholesterol-related proteins from Fig. 5b and CDS2 were plotted. Significance was determined by Student’s t-test. n = 4 replicates for 1 cancer cell line in two independent experiments. d, For one CDS1-high cell line CDS1 (sgCDS1, versus sgControl) was perturbed by CRISPR and the change in CDS1 protein level was plotted. Significance was determined by Student’s t-test. n = 4 replicates for 1 cancer cell line. In b and d: ^*P<0.05, ^**P<0.01, ^***P<0.001. Source data

Extended Data Fig. 6. Lipid homeostasis requires either CDS1 or CDS2 expression.

a, ΔCDS2 lipid changes, presented in alphabetic order. Only lipid classes detected in all samples were plotted. Stars indicate number of CDS1-negative cancer cell lines with at least two-star significant lipid alterations compared to the CDS1-high cancer cell line. For abbreviations see bottom of figure description. Two-way ANOVA followed up by Šidák’s test was used. n = 4 replicates for 5 cancer cell lines. b, CDS2 pathway with ΔCDS2-changed lipid classes in CDS1-negative cancer cell lines marked with arrows. For abbreviations see bottom of figure description. Included enzymes are annotated on UniProt. c, Graphs depicting sgCDS2 lipid changes in a second independent experiment for SK-MEL-2 (y axis). Key lipids from Fig. 6a were plotted. Data is plotted against the Fig. 6a data (x axis). All plotted changes were significant. Two-way ANOVA followed up by Šidák’s test was used. n = 4 replicates for 1 cancer cell line in 2 independent experiments. In a: Significance was indicated by one * for each cell line with P< 0.01. Lipids: CE = cholesterol esters, Cerd18:0 = ceramide d18:0, Cerd18:1 = ceramide d18:1, DAG = diacylglycerol, FFA = free fatty acids, HexCER = hexosylceramides, LPC = lysophosphatidylcholine, LPE = lysophosphatidylethanolamine, PA = phosphatidic acid, PC = phosphatidylcholine, PE = phosphatidylethanolamine, PG = phosphatidylglycerol, PI = phosphatidylinositol, PS = phosphatidylserine, SM = sphingomyelin, TAG = triglycerides. Enzymes: PEMT = phosphatidylethanolamine n-methyltransferase, CEPT1 = choline/ethanolamine phosphotransferase 1, DGATs = diacylglycerol o-acyltransferases, DGKs = diacylglycerol kinases, CDS = cytidine diphosphate diacylglycerol synthase, TAM41 = mitochondrial phosphatidate cytidylyltransferase, CDIPT = cytidine diphosphate diacylglycerol synthase inositol-3-phosphatidyltransferase, PGS1 = phosphatidylglycerophosphate synthase 1, PTPMT1 = protein tyrosine phosphatase mitochondrial 1. Source data

Extended Data Fig. 7. CDS2 dosage and catalytic activity govern synthetic lethality.

a, Graph depicting RNA quantification by qPCR of 10-d CDS2 knockdown in ectopic CDS1-restored BLM cells from the experiment depicted in Fig. 7b. Two-way ANOVA followed up by Šidák’s test was used. n = 3 replicates for 1 cancer cell line for 3 shCDS2. b, Plot depicting time until palpable tumors were formed for the experiment in Fig. 7c. Student’s t-test was used. n = 9 NOD-scid Il2rγ-null mice per group. c, Alignment of a part of the human CDS2 amino acid sequence with bacterial TmCdsA showing they both contain an aspartic acid (red box). Added on top is the conservation across 150 species of each of the presented amino acids as determined by ConSurf. d, The computationally predicted (Alphafill) ion binding to the aspartic acid highlighted in panel c for bacterial TmCdsA and human CDS2. The bacterial structure was validated in a previous report. In a: ^***P<0.001. Source data

Extended Data Fig. 8. Gating strategies.

a, The gating strategy for in vitro GFP or mCherry mixing experiments. b, The gating strategy for the in vivo mixing experiment. c, The gating strategy for dead cell quantification.