Finding. Unstable mitochondrial DNA inside the melanoma cell speeds the tumor, but only if cGAS can hear it, and the immune neighborhood changes with time. Bashaw, Shadel, Bosenberg, West and colleagues silence TFAM in melanocytes of an autochthonous BrafV600E/Pten-null model. Growth and metastasis accelerate. Remove cGAS and that acceleration fails. Early tumors look type I interferon-rich. Later they license immunoregulatory programs. Macrophages are the cell that changes most: inflammatory first, then efferocytic and tissue-adaptive. Project those two macrophage signatures into human melanoma and they point opposite ways. Early looks like better survival and therapy response. Late looks like resistance. Anti-PD-1 delays the mtDNA-instable mouse tumors, which fits an early interferon window, not a late one.
Why this paper matters
mtDNA in the cytosol is usually told as a STING snapshot. This preprint adds a clock. The same tumor-intrinsic instability that starts as interferon can end as a macrophage state humans already know as bad news. If you biopsy a TFAM-low or mtDNA-unstable melanoma, the question is when, not only whether cGAS is on.
It also keeps the genetics clean: tumor-cell TFAM, host cGAS, autochthonous BRAF/PTEN. That is not a cell-line supernatant story.
What they actually measured
Conditional TFAM, cGAS dependence, early/late RNA, macrophage states, human cohort mapping, anti-PD-1. Metastasis is in the first sentence, so score it as progression, not only volume.
How to read the score
Low 90s. mtDNA instability, cGAS, timed TME, human mapping, checkpoint window. Confidence is high for the mouse axis.
Caveats
Signature transfer is not a TFAM-stratified trial. Anti-PD-1 delay is modest language (delayed progression). Do not start TFAM inhibitors as immunotherapy.
What to do with it
If you profile melanoma myeloid states, split them on an mtDNA-instability / IFN axis and keep time in the model. If you treat TFAM-low tumors, the paper says hit PD-1 early. Pair it with gasdermin-mitochondria briefs already in the corpus: different leak, same innate reader family.