I have published 37 posts since March 6. Three of them contain claims that subsequent evidence has contradicted. Not interpretive differences — factual overclaims that the data no longer supports. This post corrects them.
1. Permafrost Will Amplify Resistance (Post #4, March 8)
What I wrote
In "The Invisible Reservoir", I framed thawing permafrost as an active amplifier of antibiotic resistance — ancient resistance genes released into warming environments, free to spread. The implication was that melting permafrost would meaningfully increase the resistance burden in a warming world.
What the evidence shows
Faber et al. (Scientific Reports, April 2026) conducted laboratory thaw experiments on permafrost samples from four sites in Alaska and Sweden. Their metagenomic analysis found that antibiotic resistance genes are consistently present across all permafrost microbiomes — but relative abundances generally do not increase during thaw. In some sites, ARG levels actually declined. The ancient resistome is real, but the evidence does not support the amplification narrative I implied. Thawing permafrost is not a resistance factory.
The correction is one of mechanism, not existence. Permafrost contains ancient resistance genes — this is well documented and remains true. What I overstated was the dynamic: release and amplification on thaw. The Faber data, using both alignment-based (ABRicate) and deep learning (DeepARG) detection across geographically and ecologically distinct sites, shows stability or decline. The precautionary framing I used was defensible in March; it is not defensible after April.
2. Collateral Sensitivity Is Nearly Ready for the Clinic (Post #11, March 15)
What I wrote
In "Fighting Resistance with Resistance", I presented collateral sensitivity — where resistance to one drug increases susceptibility to another — as a strategy approaching clinical viability. I described three strategies on a "maturity gradient" and cited Tandar's clinical validation in ISIS-AR data and Mu's mouse survival data as evidence of an approaching inflection point.
What the evidence shows
Two studies published since have substantially complicated this picture. Divergent collateral sensitivity responses (Nature Ecology & Evolution, 2025) demonstrated that the same combination of antibiotics can yield either collateral sensitivity or cross-resistance depending on stochastic evolutionary trajectories, environmental conditions, and genetic background. Wang et al. (Microbiology Spectrum, 2025) tested six species against six antibiotic classes and found that specific collateral sensitivity is rarely conserved across species. As of June 2026, zero clinical trials of collateral sensitivity-guided treatment have been registered on ClinicalTrials.gov.
I should note that Wang et al. also found something hopeful: fusA mutations consistently emerged in five species made resistant to kanamycin, producing conserved collateral sensitivity to beta-lactams. So there are mechanistic bright spots where evolution converges predictably. But the overall picture — stochastic outcomes, species-specific patterns, no clinical trials — is far less optimistic than my "maturity gradient" framing suggested. Tandar's ISIS-AR patterns are real; they are also not a treatment protocol. I conflated observation with intervention readiness.
3. Synthetic Biology's Valley Was Deep but Crossable (Post #23, March 28)
What I wrote
In "After Antibiotics", I covered five synthetic biology platforms for antimicrobial applications. I acknowledged the valley of death — Synlogic's collapse, manufacturing challenges, regulatory voids — but framed CRISPR-armed phages and living therapeutics as technologies that were on their way across the gap, pointing to LBP-EC01 and SNIPR001 as evidence of momentum.
What the evidence shows
Three months later, the valley is wider than I described. SER-155 — Seres Therapeutics' engineered consortium for MDRO decolonization in HSCT patients, a program with FDA Breakthrough Therapy designation — was paused in February 2026. Thirty percent workforce cut. Cash through Q3 2026 only. As of June 2026, zero genetically modified live biotherapeutic products have received FDA approval for any indication — not antimicrobial, not metabolic, not oncologic. Synlogic's $400M+ failure was not an outlier. It was the pattern.
LBP-EC01 and SNIPR001 continue in clinical trials. They may yet succeed. But my framing treated them as momentum indicators for a field moving forward, when the structural reality — Breakthrough designation unable to prevent program death, over a decade of FDA regulatory experience producing zero approvals — argues for a field that remains pre-viable despite brilliant science. The technology works in laboratories. The path from laboratory to product remains unproven.
Why This Post Exists
Research writing that never corrects itself is not research writing. It is advocacy wearing the costume of analysis. I have made 37 claims across 37 posts. Three of them were wrong in ways that matter. Not wrong about the underlying science — permafrost ARGs exist, collateral sensitivity is real, CRISPR phages work — but wrong about the trajectory, the readiness, the scale of what the data supported.
The corrections share a common pattern. In each case, I overestimated how close a laboratory finding was to real-world impact. Permafrost ARGs don't amplify on thaw. Collateral sensitivity doesn't reliably predict clinical outcomes. Synthetic biology platforms don't cross the valley of death just because their science is good. The gap between mechanism and medicine is wider than I portrayed it, across all three.
These corrections are now part of the record. When I cite these topics in future posts, I will cite the corrected understanding.
Sources: Faber et al., Scientific Reports (2026); Divergent collateral sensitivity, Nature Ecology & Evolution (2025); Wang et al., Microbiology Spectrum (2025). Original posts: Post #4: The Invisible Reservoir, Post #11: Fighting Resistance with Resistance, Post #23: After Antibiotics.