Finding. Torres, Ali, Lee, Wan and de la Fuente-Nunez mined 2,025 ancient human mitochondrial genomes for encrypted peptides and named the hits mitochondrins. Sixty-five candidates came out of the computational pipe. Thirty-eight were made. Fourteen killed Gram-negative and Gram-positive bacteria. Activity lived in hydrophobic-cationic patterning — including motif switches that turned families on or off — and one peptide lowered bacterial load in a mouse skin abscess.
Why this paper matters
Mitochondria already behave like bacteria that the immune system never fully forgave: formyl peptides, naked mtDNA, cardiolipin, ROS. The usual papers ask how those DAMPs are sensed. This one asks whether the mitochondrial genome still is an antimicrobial sequence archive.
That is a different claim. If true even as chemistry — never mind as physiology — mtDNA is not only a 13-protein OXPHOS cassette plus tRNAs. It is also a library of host-defense peptides that deep learning can decrypt from ancient molecules. César de la Fuente-Nunez’s group has run this style of genome-to-peptide pipeline on other sequence stores. Applying it to ancient human mtDNA is the mitochondrial news.
What they actually measured
ORF extraction plus deep learning on 2,025 ancient human mitochondrial genomes produced 65 mitochondrin candidates. Synthesis of 38, and antibacterial testing, left 14 actives against clinically relevant Gram-negative and Gram-positive species. The authors say length, net charge, and helicity were not sufficient to explain potency. Patterning was: nested families showed that a single motif-level change could switch killing on or off.
Mechanistic assays did not collapse to one mode. Some peptides wrecked membranes. Others killed hard with little membrane damage, which the authors read as noncanonical or multi-step mechanisms. Several peptides were low-cytotoxicity on human cells. One representative reduced bacterial burden in a murine skin-abscess model.
How to read the score
The wet-lab half is strong for a peptide-discovery paper: synthesis, panels, mechanism split, a mouse experiment. Confidence is high that some mitochondrin sequences are antibacterial. Confidence is medium that mitochondrial genomes “encode antimicrobial functions” in a living human. Encrypted and translated are not synonyms. Ancient haplogroup collections are not a random sample of human mtDNA. One abscess model is not innate immunity.
Score 86 is for a new mitochondrial sequence class with teeth, not for a rewritten textbook chapter on host defense.
What to do with it
If you mine genomes for AMPs, add ancient mtDNA and keep hydrophobic-cationic motif grammar in the ranker. If you follow mitochondrial innate immunity, file this beside formyl peptides and cGAS — and keep the physiology bar high. The experiment that would move the claim from chemistry to biology is detection of a mitochondrin peptide, or its fragment, from human cells under bacterial challenge. Until that exists, treat mitochondrins as engineered sequences the organelle happened to store.