Between May and August of 2019, agents of the Maranhão state agricultural defense agency went into caves in northeastern Brazil to kill vampire bats. This is routine. Rabies control in cattle country means culling Desmodus rotundus, and it happens under a legal framework that has been in place for decades. The agents netted seventy bats. Two of them were vampire bats. The other sixty-eight were not.
Among the bystanders were nine mustached bats, and one of those nine was carrying a coronavirus that nobody would look at for another six years. Its intestinal tissue went into RNAlater, then into a freezer at minus eighty, where it sat through the entire pandemic. The paper reporting what was in it was submitted to Nature Communications in November 2025 and published on Friday.
What was in it is a 30,386 nucleotide betacoronavirus genome, divergent enough from all five recognized subgenera that the authors propose a sixth. Its closest known relatives are two 816 nucleotide fragments from the same bat species in Costa Rica, sequenced in 2011 and 2012 and never followed up. And its spike protein carries a furin cleavage site at the S1/S2 junction that actually works.
The motif is RDAR. The one in SARS-CoV-2 is RRAR. The paper says, correctly, that these differ by a single amino acid. Within about ten days that sentence had been turned into a headline about a new Covid-like virus stoking fears of spillover, and, more interestingly, into an error inside the peer-reviewed literature itself.
This is a good paper doing careful work at the exact spot where careful work gets conscripted. Both sides of the origins fight will find something here, and both will have to bend it slightly to make it fit.
BRZ batCoV shares 66.99% amino acid identity with its closest ORF1ab relative, a hedgehog betacoronavirus from Germany. Across the standard six gene marker set the International Committee on Taxonomy of Viruses recommends, its best match is MERS-related CoV at 65.8%. Genetic distance to every existing subgenus runs 0.541 to 0.704 substitutions per site, against a within-subgenus range of 0.002 to 0.201. The spike is the most divergent part of it, topping out at 25.7% identity to Pipistrellus bat CoV HKU5.
South America has been a hole in the coronavirus map. More than 180 bat coronaviruses had already been described across 43 species in the Americas, and none of them had a furin cleavage site. That is a real gap being filled with a complete genome rather than another RdRp fragment.
The caveatNew subgenus is a proposal, not a ruling. ICTV decides that, and it has not.
This is the part of the paper that earns its place. Most claims about furin sites in wildlife viruses are ProP predictions, which is to say a neural network's opinion. Here they built the spike, expressed it in HEK293T cells, and blotted it. A cleaved band appeared. They swapped RDAR for GSAS and the band vanished. They knocked furin down with siRNA and cleavage dropped. They re-expressed furin, human or bat, and cleavage came back. They ran RaTG13 spike, which has no cleavage site, as a negative control and got nothing.
That is a clean experiment with the right controls. The site is not a prediction. It is a functioning furin substrate.
The scopeWhat was demonstrated is that a plasmid-expressed, C-terminally truncated spike gets cut by furin in a human kidney cell line. That is what "functional" means in the title. It is not a statement about the virus.
RDAR and RRAR differ at one position. Both satisfy the minimal furin recognition motif, arginine at P4 and arginine at P1 with anything in between. Both get cut.
What that sentence carries that it should notThe substituted position is P3, and the substitution is arginine to aspartate. That is a charge reversal, positive to negative, at the residue that does most of the work distinguishing an ordinary furin site from a polybasic one. RRAR has three basic residues in four. RDAR has two, with an acidic residue wedged between them. The polybasic character of the SARS-CoV-2 site is the specific feature that started the argument in 2020, and RDAR does not have it.
The nucleotide distance matters more. Aspartate is encoded by GAU or GAC. Arginine is encoded by six codons, and the nearest of them is two substitutions away. So "one amino acid substitution from SARS-CoV-2" cashes out as at least two nucleotide changes, at a position where the change has to flip the charge and then be retained. Evolution does this. It does not do it casually, and the phrase invites readers to imagine a single typo.
| Virus | S1/S2 motif | Basic residues | Polybasic |
|---|---|---|---|
| SARS-CoV-2 | RRAR | 3 of 4 | Yes |
| BRZ batCoV | RDAR | 2 of 4, one acidic | No |
| MERS-CoV | RSVR | 2 of 4 | No |
The honest version is that BRZ batCoV is closer to MERS in cleavage site architecture than to SARS-CoV-2, and the RRAR comparison is chosen because it is the one people care about.
No live virus. No pseudovirus. No receptor identified, so no ACE2 or DPP4 binding data. No infectivity assay in any cell type. No animal challenge. The only work in a living animal was immunizing four BALB/c mice with purified spike to raise an antibody for the western blots.
The structural claims run on AlphaFold3 with an average pLDDT of 58, which is low confidence by any reading. The cleavage site region scores better, 78.89 in the monomer and 75.69 in the trimer, and they cross-checked with AlphaFold2 and got agreement within 1.89 Å. They also ran Boltz-2, which disagreed with the experimental HKU5 reference structure, and dropped it. Reporting the excluded model is good practice and worth saying out loud.
Where the authors landThe discussion states plainly that because receptor binding and infectivity were not examined, any discussion of the virus's zoonotic potential should be limited. That is the correct sentence and it is in the paper. Hold onto it.
The host is identified as Pteronotus parnellii, on morphology alone. Pteronotus parnellii has been a species complex for a decade. Pavan and Marroig proposed eight species within it in 2016, the nominal parnellii is a Jamaican animal, and Brazilian populations are generally assigned to P. rubiginosus or to an undescribed Amazonian lineage. The literature on that split says explicitly that cranial characters cannot separate rubiginosus from its sympatric sister lineage. Morphological identification cannot resolve this bat to species.
The furinThe bat furin used in the rescue experiment came from Pteronotus mesoamericanus, a Central American member of the same complex, because that is the genome that exists in GenBank. The paper says "furin derived from a bat of the genus Pteronotus," which is accurate and also a tell. Furin is highly conserved and the result is almost certainly fine. But the experiment shows that a Central American bat's furin cleaves the spike, not that the host's does.
Neither of these breaks anything. Both are the kind of thing that gets sanded off when a finding travels.
The furin cleavage site has been the load-bearing artifact of the origins debate since February 2020, when Coutard and colleagues pointed out that SARS-CoV-2 had one and its closest known relatives did not. Everything downstream is an argument about what that absence means.
The natural-emergence case has been accumulating for six years and it is not thin. Wu and Zhao mapped furin motifs onto the coronavirus spike phylogeny in 2020 and found they had arisen independently many times across the family. Sander and colleagues, working with European horseshoe bat sarbecoviruses, found purine-rich sequence and RNA secondary structures at the S1/S2 junction resembling the ones that let avian influenza acquire polybasic cleavage sites, and reported that a single nucleotide change in two of their viruses would create a furin site. Zhu and colleagues described Bat CoV CD35 from Hainan in 2023, basal to the SARS lineage, carrying a polybasic furin-like site. Garry published a paper in PNAS in 2022 with the title "SARS-CoV-2 furin cleavage site was not engineered," which is not subtle but is the position.
The other case is a document. The 2018 DEFUSE proposal, submitted by EcoHealth Alliance with Ralph Baric and Shi Zhengli to DARPA's PREEMPT program, proposed screening bat SARS-related coronavirus spike genes for cleavage sites and then introducing human-specific ones at the S1/S2 junction and testing growth in Vero cells and human airway cultures. DARPA declined to fund it. It surfaced in 2021. Two years before a novel coronavirus with a novel S1/S2 cleavage site emerged in the city where one of the collaborators worked, someone wrote down a plan to make one. That is a real fact and no amount of phylogenetics dissolves it.
The Brazil paper does something specific to that argument, and it is narrower than either side will want. DEFUSE was about sarbecoviruses. BRZ batCoV is not a sarbecovirus, is not from the right hemisphere, and its cleavage site is not polybasic. What it adds is one more data point to the pile showing that the S1/S2 junction is a place where cleavage motifs keep appearing across the betacoronavirus tree, which is the paper's actual claim: the junction is a structurally permissive hotspot. Nine of fifteen betacoronaviruses they modeled had predicted sites landing in the same structural region.
That weakens "the site could only have been put there" without touching "the site could have been put there." Those are different sentences, and the whole fight is about which one people think is being contested.
The preprint went up in October 2025. In February 2026 a letter was submitted to the Revista da Sociedade Brasileira de Medicina Tropical titled "Alert on a Novel South American Bat Coronavirus with a Furin-Cleavage Site: A One Health Early Warning." It was accepted in March and it is peer-reviewed and open access.
It contains two errors, and they run in the same direction.
The first is that it expands the abbreviation. In the source paper, BRZ batCoV is named for Brazil. The letter renders it as "bat-related zoonotic (BRZ) batCoV," which inserts the word zoonotic into the virus's own name. Nothing in the paper supports that. The paper's discussion says the opposite.
The second is that the letter describes the finding as "a polybasic furin cleavage site (RRAR)." The motif is RDAR and it is not polybasic. The letter has replaced the actual finding with the thing the actual finding resembles.
I want to be careful here, because this is not a bad-faith document. It is a call for surveillance funding and One Health integration, written by people who are correct that the Americas are undersampled, and it accurately reports elsewhere that furin sites have natural evolutionary origins. The authors were not trying to mislead anyone. They compressed, and compression has a direction. Under time pressure, the version of a finding that survives is the version that matches the shape you already have in your head, and the shape everyone has in their head for "bat coronavirus furin cleavage site" is RRAR.
Four months, from preprint to peer-reviewed distortion, without a single bad actor in the chain.
Here is the part that makes the origins fight different from other scientific disagreements, and it is not about virology.
In January 2025 the CIA released an assessment that a research-related origin is more likely than a natural one. Officials confirmed it was not based on new intelligence. It was a re-reading of the existing body of reporting. The FBI has held a lab-origin position at moderate confidence since 2021. The Department of Energy moved to a lab-origin position at low confidence in 2023. Four other agencies favor natural emergence, also at low confidence. The WHO's Scientific Advisory Group on the Origins of Novel Pathogens concluded in 2025 that the weight of available evidence suggests zoonotic spillover, while noting it could not close the question without data China has not provided.
Every one of those bodies is looking at the same genome. There is no classified sequence. The disagreement is not empirical in the way disagreements about, say, whether a drug lowers mortality are empirical. It is a disagreement about what a prior should be, and about what an absence of evidence licenses you to conclude, and about how much weight a proposal document should carry against a phylogeny.
The natural and engineered categories are also less clean than the argument assumes. Furin cleavage sites are conserved motifs that appear across unrelated RNA virus families because they are useful. A site that arises by recombination in a bat and a site inserted by a graduate student are the same four residues doing the same thing to the same enzyme. The sequence carries no provenance. This is why the question is answerable only by epidemiology, by sampling records, by lab notebooks, by the things that were not preserved.
Which is the actual scandal, and it is procedural rather than virological. We are seven years out and arguing about how to weight a document because the contact tracing was interrupted, the lab records were not shared, and the market sampling happened after the market was cleaned. A paper about a bat in Maranhão cannot fix that. It gets recruited anyway, because in the absence of the evidence that would settle the question, every new datum gets asked to do work it was never built to do.
The bat, for its part, was never the target. It was standing next to the vampire bats.
A working furin cleavage site in a South American bat coronavirus is a real and useful finding, and it makes the S1/S2 junction look more like a place where cleavage motifs keep arriving on their own. It does not resolve where SARS-CoV-2 came from, and it cannot, because the sequence of four amino acids carries no record of how it got there.
The thing worth watching is not the virus. It is how quickly "RDAR, not polybasic, in a merbecovirus-adjacent lineage from Brazil" became "polybasic RRAR" inside a peer-reviewed journal, with nobody lying.
Sources
The paper: Takada K, Yamahoki N, Mifsud JCO, et al. A divergent betacoronavirus with a functional furin cleavage site in South American bats. Nature Communications, published 5 September 2026. doi:10.1038/s41467-026-77599-6. Preprint: bioRxiv 2025.10.24.684489.
The letter with the errors: Alert on a Novel South American Bat Coronavirus with a Furin-Cleavage Site: A One Health Early Warning. Rev Soc Bras Med Trop 59 (2026). doi:10.1590/0037-8682-0583-2025.
Furin sites arise repeatedly: Wu Y, Zhao S. Furin cleavage sites naturally occur in coronaviruses. Stem Cell Research 50:102115 (2020). doi:10.1016/j.scr.2020.102115.
European bat sarbecoviruses: Sander AL, Moreira-Soto A, Yordanov S, et al. Genomic determinants of Furin cleavage in diverse European SARS-related bat coronaviruses. Communications Biology 5:491 (2022). doi:10.1038/s42003-022-03421-w.
Bat CoV CD35: Zhu W, Huang Y, Gong J, et al. A novel bat coronavirus with a polybasic furin-like cleavage site. Virologica Sinica 38:344-350 (2023). doi:10.1016/j.virs.2023.04.009.
The engineering question, both sides: Chan YA, Zhan SH. The Emergence of the Spike Furin Cleavage Site in SARS-CoV-2. Mol Biol Evol 39(1):msab327 (2022). doi:10.1093/molbev/msab327. Garry RF. SARS-CoV-2 furin cleavage site was not engineered. PNAS 119(40) (2022). doi:10.1073/pnas.2211107119.
The original observation: Coutard B, Valle C, de Lamballerie X, et al. The spike glycoprotein of the new coronavirus 2019-nCoV contains a furin-like cleavage site absent in CoV of the same clade. Antiviral Research 176:104742 (2020).
DEFUSE: Leaked Grant Proposal Details High-Risk Coronavirus Research. The Intercept, 23 September 2021. theintercept.com.
Bat taxonomy: Pavan AC, Marroig G. Integrating multiple evidences in taxonomy: species diversity and phylogeny of mustached bats (Mormoopidae: Pteronotus). Mol Phylogenet Evol (2016). De Thoisy B, Pavan AC, Delaval M, et al. Cryptic Diversity in Common Mustached Bats Pteronotus cf. parnellii. Acta Chiropterologica 16(1):1-13 (2014).
Costa Rican relatives: Corman VM, et al. Highly diversified coronaviruses in neotropical bats. J Gen Virol 94:1984-1994 (2013).
Intelligence assessments: CIA statement on COVID-19 origins, January 2025 (low confidence, no new intelligence). DOE assessment, February 2023 (low confidence). FBI assessment, 2021 (moderate confidence). WHO SAGO report, June 2025.