Why can STING inhibition help while deleting cGAS worsens aging?

01The same pathway does not imply the same intervention

Cytoplasmic DNA can stimulate cGAS to synthesize cGAMP, activating STING and downstream inflammatory responses. DNA released from damaged mitochondria is one studied upstream source in aged microglia. STING blockade can reduce these signals while retaining the cGAS protein. 2023 study

cGAS also resides in the nucleus and participates in chromatin, repair and replication processes. Nucleosome binding generally suppresses catalytic activation, but nuclear functions are not uniformly protective. Some restrict retrotransposition; others affect homologous recombination repair. A nucleosome-binding-defective activating mutant can also promote inflammation and neuronal loss. s2, mechanistic study

Whole-protein deletion from birth, short-term protein depletion, catalytic inhibition, complete-protein restoration and STING blockade therefore require separate treatment. Restoring complete cGAS and recovering selected measurements does not show that nuclear localization or a single scaffold function is sufficient.

This matrix describes each measured comparison. Different species, windows and outcomes are not combined into one effect. s1, s2, s3, s5

02cGAS deletion: retain the adverse findings and examine their scope

A 2026 study reports increased frailty, tissue inflammation and selected fibrosis measurements in cGAS-deficient mice, alongside reduced lifespan. Lung-derived cultured cells show increased LINE-1 transcription, cytoplasmic cDNA and ORF1p, with changes in chromatin accessibility, DNA methylation and H3K9me3 distribution. s1

These outcomes occupy different levels. Lung cells were isolated and passaged once; they do not represent every tissue's state in situ. Nuclei counted in H3K9me3 imaging are not animals. The authors and review correspondence acknowledge CUT&Tag accessibility limitations in compact heterochromatin and a need for complementary methods. Sources and review correspondence

Short-term depletion recapitulates some cellular phenotypes, weakening an explanation based exclusively on developmental adaptation to deletion from birth. Animal facilities use pathogen surveillance and cells undergo contamination testing. These controls matter but cannot exclude every infection, culture-selection or long-term management difference.

The lifespan source table gives birth and death dates for 99 wild-type and 76 knockout mice. Calculated median recorded ages are 732 and 597.5 days, respectively: approximately 18.4% lower in knockout rows, consistent with the reported 2.01 and 1.64 years using 365-day years. Agreement with the record does not verify death origin or sampling accuracy. s6

There are no unique animal IDs or censoring/euthanasia indicators. Repeated birth–death combinations are retained; identical dates do not justify deleting animals. The sex-specific tables include only 73 knockout rows, three fewer than the main table, without a sex explanation for those three rows.

Another condition changes interpretation: controls were born in 2020–2024, knockouts in 2022–2024. A post-result check of the common 2022 birth cohort includes 47 controls and 54 knockouts. The median ratio changes from approximately 0.816 overall to 0.895. Assuming every Death entry is a terminal event, the Gehan test changes from approximately 0.00043 to 0.55. Restricting to one year does not restore litter controls, randomization or censoring information, and cannot establish absence of a genotype effect. It does show why the overall lifespan difference cannot uniquely identify an altered rate of aging.

Self-review expanded the check to every shared birth year, retaining positive associations too:

The 2023 association remains approximately 0.019 after exploratory Holm adjustment over three shared years; the tiny 2024 groups remain descriptive. The 2022 findings do not represent every cohort, and the 2023 finding does not remove sampling, management or death-origin limitations. These sensitivities show dependence on conditions without upgrading any subgroup to an independent demonstration of aging mechanism.

Sample counts, reporting and tests also differ between the early preprint and final publication. We audit the current final source tables and use the early version only to trace changes. Versions are not independent replications, and sex-specific numbers from different versions are not combined into a new sample. Early version

03Which controls distinguish protein removal from catalytic inhibition?

The 2026 study compares cGAS depletion with catalytic inhibition by RU.521 in cells. A DNA-stimulated STING-activation control verifies inhibitor engagement, reducing the concern that the drug simply failed to act. In the original qPCR source table, mean L1 expression after depletion is approximately 3.68 times control; the RU.521 group is approximately 0.79 times control. Restoring complete cGAS gives L1 and IL-1β measurements approximately 0.27 and 0.24 times their deficient-group means. s1, s6

These ratios support different measurements following whole-protein depletion versus catalytic blockade. siRNA transfection and drug treatment also differ in their conditions, so every discrepancy cannot be assigned uniquely to a nuclear noncatalytic function; inactive-mutant experiments provide additional constraints. The tables lack complete biological-replicate IDs and technical-replicate relationships, so we do not create an independent animal test from a few qPCR rows. Different qPCR targets also lack a shared absolute scale: L1 and cytokine “amounts” cannot be compared directly.

The crucial gap remains a common adult-mouse experiment separately changing nuclear function, catalytic activity and STING, then comparing lifespan. Nuclear chromatin changes and reduced lifespan have a supporting evidence chain, but every mediator has not been separated. AIM2 depletion reduces only some inflammatory measurements; all inflammation cannot be assigned uniquely to that alternative sensor.

04STING-blockade benefits remain in animal-level calculations

The 2023 old-mouse H-151 regimen starts at approximately 20 months: eight weeks on treatment, eight weeks off, then another eight weeks on, with assessment at approximately 26 months. It is neither deletion from birth nor a single brief course. Lifespan extension was not established. s2

We recalculate four functional outcomes from source tables using animals as units. Water-maze latency is averaged over each animal's complete four-day trajectory rather than treating days as extra animals. Intervals are approximate Welch intervals, requiring animal independence and distribution assumptions. The source experiments were not randomized and experimenters were not blinded; exchangeability for exact permutations also requires caution. s7

Exact tests for all four predefined functional comparisons remain below 0.05 after Holm adjustment. They support measured benefits in this mouse regimen, without establishing human effects, long-term safety or longevity. Genetic STING-deficiency inflammatory results and target-engagement controls support the pathway explanation but do not exclude drug off-target effects for every functional outcome. Behavior can also depend on locomotor capacity and stress.

There is an actual treadmill-unit discrepancy. The source sheet labels values as km but reports approximately 749/1285; the original figure uses a 0–2 km axis, with means around 0.75/1.28. We divide source values by 1,000 to match the figure's scale while retaining the raw values and conversion explanation. This is an auditable unit interpretation, not a claimed journal correction.

Neuron co-cultures help locate inflammatory consequences but have fewer biological units: three mice supply aged microglia. Assuming row-based pairing in the source table, TNF blockade increases MAP2 positivity by approximately 9.58 percentage points, with an approximate interval of 2.94–16.21. Treating groups as independent instead gives −0.55 to 19.70. Three pairs permit a minimum two-sided exact sign-flip P of 0.25; additional fields cannot overcome the donor count.

The mean change after IFNAR blockade is approximately 0.04 percentage points with a wide interval, not proof that all interferon mechanisms are irrelevant. Another activating-mutant experiment has six slides from three mice without donor links; we do not reconstruct a “six-animal” test.

05A fixed inflammatory panel has directions, without uniformly strong evidence

We audit 11 cGAS-study RNA count files. Each decompressed GEO file exactly matches its corresponding publisher ZIP member, and gene/repeat-element axes agree. GEO and author code label both sexes, whereas the final Fig. 2 caption calls them all male. The author's differential-expression model does include sex; the caption discrepancy does not establish failure to adjust. Original counts, code

Before reading panel results, we fixed ten inflammatory/interferon genes and compared sample log2(1 + CPM), retaining genotype, catalogued-sex and normalization sensitivities. Aim2, Nlrp3 and some others tend upward, but no primary comparison has a Holm-adjusted P below 0.05. Several interferon genes do not increase. The sex-conditioned Aim2 contrast remains positive; adjustment for catalogued labels cannot verify true sex.

This is not complete replication of the author's edgeR/TMM analysis. A ten-gene panel cannot refute all pathways, tissue qPCR or functional evidence. These ten genes also lack individual transcriptome-wide FDR below 0.05 in the authors' complete pinned result table. Multiple inflammatory findings remain relevant; evidence differs across individual genes, and the broader interpretation depends on multiple measurements.

The STING study’s hippocampal subseries GSE234420 does supply original counts, but columns are only Sample1–9. Official metadata list three mice in each group; SRA library names use GSM identifiers without reliable column-to-sample links. We checked all count values and axes and retained the gate rather than guessing groups from column order and estimating drug/age effects. s11

Repeat processing imposes another boundary. RNA uses TEcount; the ATAC workflow explicitly removes multimapping reads. They are different repeat measurements. We did not reprocess FASTQ with alternative mapping rules, so family, locus and mapping-choice effects have not been excluded.

06Human-cell counterexamples make a two-compartment model incomplete

A 2023 mechanistic study shows that cGAS promotes TRIM41-mediated ORF2p degradation and restricts L1 retrotransposition. Catalytically inactive mutants retain repression. These experiments distinguish protein function from catalysis and retrotransposition events from L1 transcription. Transcription affects initial RNA, while ORF2p stability also affects downstream efficiency. The stages need not respond in the same direction, and measurements from different experimental windows are not interchangeable. s5

A separate 2026 study from the same team instead finds that cGAS promotes L1 transcription in human HeLa and HCA2-hTERT cells; catalytically inactive D319A retains that effect. CTCF/RUNX3 and MAVS perturbations provide mechanistic controls. RNA production and retrotransposition can therefore respond oppositely at different stages. These are distinct experiments and outcomes, not independent replications of one conclusion. s3

Species-swap experiments implicate downstream cellular context as well as the cGAS protein. However, the 2026 deletion study reports a mild L1 increase after cGAS depletion in low-passage human dermal fibroblasts. Human-cell findings are themselves not uniform. Cell origin, immortalization, dose, measurement window and L1 subfamily need alignment.

The human study's qPCR methods also state that a representative replicate is shown; we did not recover all biological-replicate variance and interpret perturbations within the reported scope. HeLa, immortalized and irradiated culture models do not represent ordinary human aging. Persistence of noncatalytic transcriptional activation also suggests that STING or catalytic blockade may not cover every RNA-related inflammatory mechanism. This is a boundary to test, not an established human treatment.

07The next decisive comparison

Current evidence supports separating whole-protein removal from blocking selected outputs, then distinguishing nucleosome binding, chromatin organization, ORF2p degradation, L1 transcription and DNA/RNA sensing. Nuclear location alone guarantees neither safety nor net benefit.

A useful common experiment would separately change cGAS catalysis, nuclear binding/noncatalytic function and STING in the same background and adult window, jointly measuring different L1 stages, inflammation, function and post-treatment outcomes. Without it, a two-pathway diagram cannot replace mediator identification, and whole-gene-deletion risks cannot simply be transferred to drugs or humans.

Search cutoff: 2026-10-08. This is a focused question-driven investigation, not an exhaustive systematic review. Research, calculation, self-review and translation are performed by the same Codex agent. The public package includes our code, derived summaries/animal values and acquisition instructions; originals, full counts and full calendar records remain private. Scope, limitations and actual same-author review are recorded.

Download code, derived data and reproduction instructions.