Reanalysis of the same China CDC raw sequencing data (see S-21 - China CDC raw metagenomic sequencing data release — Huanan market samples (NGDC GSA - NCBI BioProject)) used by S-24 - Crits-Christoph et al. — Genetic evidence of susceptible wildlife in SARS-CoV-2-positive Huanan market samples (Zenodo report). Quantifies per-sample mitochondrial-DNA read counts against SARS-CoV-2 read counts across species and finds the samples with the most SARS-CoV-2 material correlate at least as strongly with fish (catfish, largemouth bass) as with any mammal; of 14 samples with substantial raccoon-dog mtDNA, only 1 had any SARS-CoV-2 reads. Argues co-location of animal DNA and viral RNA at a market, sampled a month after human transmission was already widespread, cannot reliably indicate which animals (if any) were infected. relevance_note: the central methodological counter to the raccoon-dog co-location claim; argues the correlational evidence is too weak/confounded to identify an infected intermediate host.
Raccoon-dog-dominated samples and viral reads (Section 2.3)
O-51 - Of 14 samples dominated by raccoon-dog mtDNA only one contained any SARS-CoV-2 read (1 of ~2.1x10^8); all six bamboo-rat-dominated samples had zero
The samples richest in raccoon-dog genetic material are almost all SARS-CoV-2-negative, the opposite of what a simple raccoon-dog-shed-the-virus picture would predict. This is the pointed count underlying Bloom’s skeptical reading, on the same D-1 market data.
Link to original
Correlation of SARS-CoV-2 with species mtDNA (Section 2.4)
O-50 - Across market samples SARS-CoV-2 abundance correlates most with fish and livestock mtDNA (bass, catfish, cow, carp, snakehead) and negatively with raccoon dog and bamboo rat
Bloom aligned reads from 176 environmental samples to the SARS-CoV-2 genome and a panel of chordate mitochondrial genomes, then correlated per-sample SARS-CoV-2 read counts against each species’ mtDNA read counts. The virus tracks fish/livestock/human material (species handled or deposited by people throughout the market) at least as strongly as any susceptible mammal, and shows a negative association with raccoon dog and hoary bamboo rat. The ranking is analysis-choice-sensitive (excluding the first sampling timepoint makes Oriental rat snake most-correlated; linear vs log scale and non-zero-only subsets shift the ordering), but raccoon dog stays negative. Same China-CDC market set (D-1) that S-35 analyses toward the opposite emphasis.
Link to original
A-54 - SARS-CoV-2 tracking fish and livestock rather than raccoon dogs is the pattern expected from human deposition, undercutting the specific raccoon-dog-shedding claim
The argument is the direct animal-shedding prediction. In a scenario where a live animal at a stall is productively infected and shedding SARS-CoV-2, the environmental RNA it deposits should be spatially and quantitatively tied to that animal’s own biological traces (its mtDNA): samples richer in raccoon-dog material should tend to be richer in virus. Bloom’s quantification shows the opposite — the virus is most abundant in samples dominated by fish (largemouth bass, catfish, carp, snakehead) and livestock (cow), and is if anything negatively associated with raccoon dog and hoary bamboo rat, with 13 of the 14 raccoon-dog-dominated samples carrying no viral reads at all. Fish and cattle are not plausible SARS-CoV-2 hosts, so their strong association indicates the viral RNA is not indexing an infected animal but rather tracks surfaces, products, and drainage where infected humans deposited virus during a market that was, by the sampling date, already a human-transmission hotspot. On this reading the metagenomic co-location fails to single out raccoon dogs (or any susceptible mammal) as the shedding host. (Crits-Christoph counter that this whole-dataset correlation is confounded by the sampling design and by animal-before-human shedding plus RNA decay — a validity dispute step 6 must weigh.)
Verdict (step 6) — approved, checked
Reconstruction. Premise (the observed correlations): viral RNA abundance is highest in fish/livestock/human-dominated samples and null-to-negative with raccoon-dog and bamboo-rat mtDNA (13/14 raccoon-dog-dominated samples had zero reads). Shedding prediction: a productively infected live animal deposits RNA co-located with its own mtDNA, so a shedding host would produce a positive viral–mtDNA association. Load-bearing step: the observed absence of that positive association for raccoon dog → the metagenomic co-location does not support (undercuts) the raccoon-dog-as-shedding-host claim.
Evaluation, conditional on premises. The relevant undercutting defeater is the Crits-Christoph confound: with sampling weeks-to-months after shedding, plus decay and stall turnover, even a genuine shedding animal could fail to show a positive correlation, so absence of correlation is a weak instrument. Critically, this defeater cuts against a stronger claim — that the null proves raccoon dogs were uninfected (A-55’s territory) — but it does not rescue positive support for raccoon dogs from a null result: a null/confounded result simply fails to provide affirmative support. The statement’s operative verb is “undercutting the specific raccoon-dog-shedding claim” (i.e. removing support), and “expected if… human deposition” is consistency language, not proof. That epistemically-safe residual survives the defeater. Approved (the inference does not license the stronger affirmative “human deposition is proven,” but the statement does not assert that). Traced directly — checked.
Link to original
What co-location can and cannot show (Discussion)
A-55 - Because the samples were collected a month into widespread human transmission, their viral-animal co-location is uninformative about the outbreak source in either direction
This is the meta-level caveat that bounds Bloom’s own negative result. By the time the market was sampled (January-March 2020), the market had been an active human-transmission setting, so viral RNA on surfaces overwhelmingly reflects human deposition and post-hoc contamination rather than the original spillover event; a correlation (or its absence) between viral reads and a species’ mtDNA in such samples is a weak instrument. Hence: (i) a positive co-location would not prove the co-located animal was the source (it could be human-deposited virus on animal-rich surfaces), and (ii) the observed absence of a raccoon-dog association does not prove raccoon dogs were uninfected — infected animals could have been removed before sampling, or their RNA could have decayed over the intervening weeks. The correct conclusion is symmetric agnosticism: this particular metagenomic co-location analysis does not settle whether an animal at the market was infected. This blocks the inference in S-35 from co-location to animal infection, while equally blocking any inference from Bloom’s null to animal non-infection.
Verdict (step 6) — corrected, checked
Reconstruction. Premise: sampling (Jan-Mar 2020) postdates widespread human transmission by ~a month, so surface viral RNA is dominated by human deposition and post-hoc contamination. Load-bearing step: therefore the viral–mtDNA co-location analysis is a confounded instrument in both directions — (i) a positive co-location need not indicate an infected animal (human-deposited virus can land on animal-rich surfaces), and (ii) a null association need not indicate uninfected animals (removal before sampling, RNA decay).
Evaluation, conditional on the premise. Both directional legs are valid: human-deposition confounding genuinely severs co-location→animal-source, and artifact/decay/removal genuinely severs null→animal-non-infection. The symmetric-agnosticism structure holds. The surviving defeater is only against the degree: “uninformative in either direction” asserts a likelihood ratio of exactly 1, which is too strong — a sufficiently strong, spatially-resolved co-location (e.g. virus concentrated at specific susceptible-animal stalls, as Crits-Christoph argue) would still shift beliefs, so the instrument is weak, not null. The argument’s own body says “weak instrument” and “does not settle,” which is the immune form. Corrected the statement from “uninformative either way” to “weak, non-decisive instrument in both directions (rather than strictly zero information)”; the “cannot reliably identify” and “does not disprove” clauses were already valid and are retained. Traced directly — checked.
Original
Because the samples were collected around a month after human-to-human SARS-CoV-2 transmission was already widespread in Wuhan, so co-mingling of viral RNA and animal mtDNA cannot reliably identify an infected intermediate host — and, symmetrically, the lack of viral-animal association does not disprove that infected animals were present; the data are uninformative about the source either way.
Link to original