αKG, Carnitine Synthesis and DNA Repair

αKG-mediated carnitine synthesis drives DNA repair via histone acetylation (Uboveja et al., 2026), published in Nature, takes on a problem for many patients. PARP inhibitors and other DNA-damaging drugs work best against tumors that cannot repair DNA by homologous recombination. Tumors that repair well, including many with high levels of the cell-cycle gene CCNE1, resist them. The authors found a metabolic reason those cells repair so well, and a way to take it away.

First author Apoorva Uboveja, PhD, received an OCRA Mentored Investigator Grant in 2023 for this line of work in the laboratory of Katherine Aird, PhD.

Explore the data: Open the dataset in the Discovery Lab TRE to run reproducible analyses where the data resides.

Dataset page · Open in the Discovery Lab TRE

Key Findings

• Cells that repair DNA well need TMLHE. A CRISPR screen of 64 enzymes that depend on the metabolite αKG found that cells under a PARP inhibitor need TMLHE, the first enzyme in making carnitine, to survive.
• αKG works through carnitine. Depleting αKG or blocking carnitine synthesis makes the drug far more lethal, and adding back carnitine alone restores survival.
• Carnitine feeds histone acetylation at specific sites. Carnitine made this way supplies the acetylation of particular histone sites that helps cells repair DNA, a role not covered by the cell's other sources of acetyl groups.
• The link appears in patients. In tumor samples, TMLHE tracked histone acetylation, and high TMLHE or acetylcarnitine went with worse progression-free survival after treatment with DNA-damaging drugs.

The result points to a metabolic way to make repair-proficient tumors vulnerable to drugs that already exist.

The Data

• Experiments: the paper's supplementary source data, in CCNE1-high FT282 (fallopian tube) and Ovcar8 (ovarian cancer) cells.
• RNA-seq: GEO GSE289735. Histone proteomics: PRIDE PXD060690.

Paper: Uboveja A, Yang B, Buj R, et al. αKG-mediated carnitine synthesis drives DNA repair via histone acetylation. Nature 655, 790–800 (2026). doi:10.1038/s41586-026-10584-7

Cells Under a PARP Inhibitor Need TMLHE to Survive

The authors knocked out, one at a time, each of 64 enzymes that depend on αKG (63 appear in the published results), then treated the cells with the PARP inhibitor olaparib. A gene whose loss kills cells under the drug drops out of the population. Eight genes dropped out strongly in both cell lines, and TMLHE, the first enzyme in making carnitine, was among the strongest. Points toward the lower left are the genes the cells needed most.

αKG Protects Cells by Making Carnitine

If αKG matters because it feeds carnitine synthesis, then carnitine alone should undo the damage of losing it. That is what happens. Blocking carnitine synthesis with mildronate, together with olaparib, cuts viability to 38–41%. Adding αKG does not help, because the block sits downstream of it, but adding carnitine restores viability to 76–77%. Lowering αKG itself with an IDH1 inhibitor gives the same picture: 44–48% viable with olaparib, and 85–92% once carnitine is added.

Carnitine Synthesis Feeds Histone Acetylation at Specific Sites

Carnitine helps move acetyl groups, which cells use to acetylate histones and open chromatin for repair. CCNE1-high cells carry more histone acetylation than their parental cells; inhibiting IDH1 lowers it, and giving carnitine to cells without TMLHE raises it. The effect is site specific: knocking down IDH1 cuts acetylation most at H3K23, H3K9 and H4K5, while sites such as H3K56 and H3K79 barely change.

Losing TMLHE Reshapes Gene Expression

Knocking down TMLHE in CCNE1-high Ovcar8 cells shifts the expression of thousands of genes. The shift is robust: an independent reanalysis of the raw sequencing reads in the Discovery Lab TRE matches the authors' fold changes closely (correlation 0.88 across 12,346 genes), so the expression changes do not depend on one analysis pipeline.

About These Charts

Read duplication in the RNA-seq tracks the experimental group (CCNE1 OE 55.6%, EV 43.7%, shTMLHE#1 36.4%, shTMLHE#2 32.9%). Both hairpins sit on the same side of that gradient, so agreement between them does not control for it.

The reanalysis and the published table are compared, not merged. The reanalysis is nf-core/rnaseq followed by DESeq2 with CCNE1 OE as the reference group.

These charts were built in the Discovery Lab TRE from the paper's supplementary files (ab71949b; figure tables e444bc88) and the GSE289735 reads (9ae00f0d; differential expression 0e3ade70).