Peer-reviewed, Cell 187(19):5468–5482.e11 version of S-31 - Crits-Christoph et al. — Genetic tracing of market wildlife and viruses at the epicenter of the COVID-19 pandemic (bioRxiv preprint) / S-24 - Crits-Christoph et al. — Genetic evidence of susceptible wildlife in SARS-CoV-2-positive Huanan market samples (Zenodo report). Core claims: SARS-CoV-2-positive environmental samples are non-randomly concentrated at/near a specific stall selling live susceptible mammals (raccoon dogs prominent); mitochondrial DNA from those species is significantly more abundant in SARS-CoV-2-positive vs. -negative samples even after adjusting for overall sample richness; recovered viral genomes’ phylodynamics are consistent with market emergence rather than pre-market circulation. Published the same day a 22-signatory letter (led by Bryce Nickels) requested its retraction. relevance_note: the most-cited, peer-reviewed version of the stall-level raccoon-dog co-location claim central to the zoonosis case; also the most contested node in this slice.
Market genetic diversity and emergence (Figure 1)
O-39 - Market-associated SARS-CoV-2 genomes' tMRCA matches the global pandemic tMRCA and both lineages A and B are present in market environmental samples
Recovered environmental viral genomes (A20 = lineage A; F54, F13, B5 = lineage B) place the market’s viral diversity at the root of the pandemic tree rather than downstream of it. Presence of both lineages at the market argues against the market outbreak being merely a lineage-B superspreading amplification of prior community transmission. Rests on the early-genome collection (D-3) plus the market metagenomic reads.
Link to original
A-51 - Market-associated genomes' tMRCA equalling the global tMRCA, with both lineages present, indicates emergence at the market rather than downstream amplification
Under a phylogenetic-dating logic: a set of sequences sampled from a location that was infected only after the virus had been circulating in the community for some time would have a most-recent-common-ancestor postdating the true pandemic root, and a single-introduction amplification would tend to carry one lineage. The market environmental genomes instead (a) have a tMRCA overlapping the global pandemic tMRCA (late Nov-Dec 2019) and (b) include both basal lineages A and B, with lineage A recovered directly from an environmental sample (A20). Both features are those expected if the earliest ancestral diversity of the pandemic was present at the market — i.e. the market is at the origin rather than a later amplifier. This rebuts the alternative that lineage B at the market was a superspreading amplification of unrelated prior lineage-A community spread.
Verdict (step 6) — approved, checked
Reconstruction. Premises: (i) market environmental genomes’ tMRCA overlaps the global pandemic tMRCA; (ii) both basal lineages A and B are present in the market environment, with A recovered directly from environmental sample A20. Load-bearing step: full ancestral diversity + root-matching tMRCA physically located at the market is more expected under “market at the origin” than under “market is a downstream amplifier of prior community spread.”
Evaluation, conditional on premises. The step is evidential (“indicating”), not deductive, and the statement frames it as such. The standard undercutting defeater is that a market could amplify a community outbreak that already carried both lineages, reproducing both features without the market being the origin. That defeater is genuinely weakened here by the physical recovery of lineage A in the market environment itself (A20): both basal lineages are demonstrated at the market locus, not merely inferred to have passed through it, which is exactly what “downstream amplification of a single introduction” (the alternative the argument targets) does not predict. As an evidential claim — the observation raises the likelihood of market-at-origin over downstream-single-lineage amplification — no defeater survives that denies neither premise. It does not prove origin (a diverse community outbreak feeding the market remains logically possible), but the statement’s hedged “indicating” is the immune conclusion. Approved. Traced directly from the phylogenetic-dating logic — checked.
Link to original
H-24 - SARS-CoV-2 emerged at the Huanan market rather than circulating in the Wuhan community beforehand
Argued from the market-associated genomes’ tMRCA matching the pandemic tMRCA and both lineages A and B being present at the market. Competes with the pre-market community-circulation / market-as-amplifier reading (e.g. Bloom 2021).
Link to original
SARS-CoV-2 positivity and wildlife DNA at stall A (Figures 2-3)
O-40 - SARS-CoV-2 positivity concentrated at wildlife stall A, whose five positive samples all contained susceptible-mammal (incl. raccoon dog) mtDNA
Sampling: 1 Jan 2020, 515 samples, 27 qPCR-positive (25 sequenced); 12 Jan, 70 samples from 7 wildlife stalls (3 qPCR-positive, all 70 sequenced). Three independent spatial signals (qPCR, mNGS, drains) converge on wildlife stall A in the market’s western/wildlife section. In the five positive stall-A samples, wildlife mtDNA (raccoon dog, hoary bamboo rat, European rabbit, etc.) was present; excluding 16S/12S rRNA regions did not change the result. Nutria and foxes appeared only in SARS-CoV-2-negative samples/stalls. This is a co-location finding on the China-CDC market set (D-1), shared with S-39 (Bloom); it establishes co-location, not infection.
Link to original
A-48 - Concentration of viral positivity at the exact stall selling susceptible mammals raises the probability of an infected-animal source but does not prove infection
The inference is Bayesian, not deductive. Under the hypothesis that an infected live animal shed SARS-CoV-2 at the market, one predicts viral RNA to be enriched precisely where that animal (and hence its shed mtDNA) was, i.e. at the wildlife stall — which is observed (three independent spatial signals converge on stall A; all five positive stall-A samples carry susceptible-mammal mtDNA; nutria/foxes appear only in negatives). Under the competing hypothesis that only humans shed the virus, viral RNA should track human presence/handling and be spread across the market like human mtDNA, not localise to one wildlife stall while humans are not even the most abundant mammal there. So the spatial pattern favours a wildlife source. The argument is explicitly bounded: because the environmental samples were taken about a month after human transmission was already widespread, virus could also have been deposited by infected humans onto animal-DNA-rich surfaces, so co-location raises but does not settle the probability that the animals themselves were infected — a limitation the authors concede.
Step 6 verdict — approved (checked)
Reconstructed step: a likelihood-ratio comparison. P(observed spatial pattern | infected animal shed at stall A) is high, because that hypothesis predicts viral RNA enriched exactly where the animal’s shed mtDNA is; P(same pattern | only humans shed) is lower, because human shedding predicts RNA tracking human presence and spreading like human mtDNA rather than localizing to one wildlife stall where humans are not the dominant mammal. Given the premises (three converging spatial signals on stall A; all five positive stall-A samples carry susceptible-mammal mtDNA), LR > 1, so the pattern favours a wildlife source. Probed the strongest undercutting defeater: post-outbreak human deposition of virus onto animal-DNA-rich surfaces (samples taken ~1 month after human transmission was widespread) — this is exactly what would break an inference to “the animals were infected,” and the statement is already hedged to “raises the probability… does not prove infection.” The hedged conclusion is immune to that defeater. Traced from the premises, so checked.
Link to original
Correlational analysis of the balanced wildlife-stall set
O-41 - In the balanced n=70 wildlife-stall set human mtDNA was uncorrelated with viral RNA while Malayan porcupine and Himalayan marmot mtDNA were significantly positively correlated
The authors restrict the correlation to the 70 January-12 wildlife-stall samples (a balanced subset, all sequenced irrespective of positivity) rather than the full sequenced set. Human mtDNA does not track the virus; porcupine and marmot do (reflecting their concentration in stall A). Notably raccoon dog itself is not among the significantly positively-correlated species here. Same D-1 data that S-39 (Bloom) analyses with the opposite emphasis.
Link to original
A-49 - The reported negative correlation between wildlife mtDNA and viral RNA is largely a sampling-design artifact, not evidence against a wildlife source
Three confounds undercut the whole-dataset correlation. (1) Selection design: on 12 January every wildlife-stall sample was sequenced regardless of qPCR status, whereas the other sequenced samples were predominantly qPCR-positive and mostly from non-wildlife locations; wildlife species are therefore over-represented among the sequenced negatives, mechanically driving a spurious negative correlation between wildlife mtDNA and SARS-CoV-2 (their Figure S4 simulation). (2) Non-wildlife stall positives were very likely human-shed, so including them in a univariate correlation contaminates the signal. (3) Timing/decay: in a zoonotic scenario animal shedding precedes human shedding, and most wildlife stalls were sampled 11 days after the case-linked stalls, giving animal-derived RNA more time to decay — again weakening any positive animal correlation. When the analysis is restricted to the balanced n=70 wildlife-stall set, human mtDNA does not correlate with the virus while some wildlife species (porcupine, marmot) do. Hence the negative-correlation result relied on by skeptics is an artifact of pooling and design, not a robust biological signal; a correlation on these post-outbreak samples is in any case a weak instrument for identifying the infected host.
Step 6 verdict — approved (checked)
Reconstructed step: a selection-bias (collider-style) argument. Because the 12 January protocol sequenced every wildlife-stall sample regardless of qPCR status while the other sequenced samples were predominantly qPCR-positive, wildlife species are mechanically over-represented among sequenced negatives; conditioning the correlation on a selection variable correlated with both mtDNA source and viral status manufactures a spurious negative association independent of any true biological relationship. Hence the pooled negative correlation is not evidence against an infected-animal source. This is a valid statistical inference, and conditional on the premises (the Fig S4 simulation reproducing the artifact; the balanced n=70 reanalysis where human mtDNA is uncorrelated and porcupine/marmot are positively correlated) it holds. Probed for a defeater — could the negative correlation be a real signal the balanced set fails to capture? — but the simulation directly demonstrates the design generates it, so the “artifact” reading survives. The conclusion is appropriately limited to “not evidence against,” not “evidence for.” Traced from the premises, so checked.
Link to original
Wildlife viruses in the market (Figure 4)
O-42 - Host-specific wildlife viruses (raccoon-dog amdoparvovirus, bamboo-rat betacoronavirus, civet kobuvirus) were recovered from the market, phylogenetically traceable to southern-China farms
Because these viruses have narrow, known host ranges and were found where their host’s mtDNA was, they indicate that live, productively infected animals of those species (raccoon dog, bamboo rat, civet) were physically present in the market in late 2019, and their phylogenetic links to southern-China (Guangxi/Yunnan) farms trace a wildlife-trade conduit into Wuhan. Recovered from the same D-1 metagenomic reads.
Link to original
A-50 - Recovery of narrow-host-range wildlife virus genomes co-located with their hosts implies live, productively infected animals were present in the market
Two features carry the inference. First, host specificity: each recovered viral genome sits within the clade of viruses reported from a single host species, so its presence indexes that particular animal, not generic contamination. Second, viability signal: recovering a mostly-complete viral genome at appreciable abundance from a stall where the host’s mtDNA is concentrated is best explained by an actively-shedding, live-infected animal, because a dead frozen carcass or trace environmental DNA would not typically yield abundant replicating-virus reads co-localised with host tissue. The phylogenetic placement of the bamboo-rat betacoronavirus next to a 2019 Guangxi-farm virus, and the civet kobuvirus next to Sichuan/Guangxi market-animal sequences, further shows a live wildlife-trade conduit from southern China into Wuhan. This establishes that the market contained infected live wildlife of exactly the candidate intermediate-host species, raising the prior that SARS-CoV-2 could have entered humans by the same route (though these are not SARS-CoV-2 itself).
Step 6 verdict — corrected (checked)
Reconstructed step, two components. (a) Host-specificity → the recovered genome indexes that particular species’ body being present, not generic environmental contamination: valid and holds, since a genome nested within a single host’s viral clade, at appreciable abundance, and co-located with that host’s mtDNA cannot be explained by drift-in contamination. (b) The abundance/viability signal → the animal was live and productively infected in the market: this over-reaches. The undercutting defeater the argument itself raises (“dead frozen carcass”) is dismissed too quickly — an animal infected before slaughter (e.g. amdoparvovirus persistently infects raccoon dogs; a coronavirus-viremic bamboo rat) carries a high viral load in its tissues, and freezing preserves that RNA, so a freshly-killed or frozen infected carcass/meat would also yield abundant host-co-located viral reads. Viral abundance therefore cannot distinguish a live shedding animal from a slaughtered-but-infected one. The defeater survives against “live,” but the weaker conclusion — infected animals of these species (live or as carcasses/tissue) were physically present and trace/generic contamination is excluded — is immune to it, and still supports the wildlife-trade conduit (the farm-traceable phylogenetics stand regardless). Hence corrected to that weaker form. Traced directly, so checked.
Original
Reconstructing near-complete genomes of viruses with narrow, known host ranges (raccoon-dog amdoparvovirus, bamboo-rat betacoronavirus, civet kobuvirus), each found where its host’s mtDNA was and nested phylogenetically among that single host’s viruses, implies live productively-infected animals of those species were physically in the market — not merely frozen carcasses or residual DNA.
Link to original
Central hypothesis
H-23 - Susceptible wildlife sold at Huanan market wildlife stall A (raccoon dogs and-or related mammals) were the intermediate host of the SARS-CoV-2 spillover
The paper’s central pro-zoonosis claim, built from the co-location of viral positivity with susceptible-mammal DNA at stall A and the presence of productively infected wildlife (animal viruses). Explicitly framed by the authors as identifying a short list of candidate intermediate hosts to prioritise, not as proof of infection.
Link to original