The COVID-19 pandemic affects humans, but also many of the animals we interact with. So far, humans have transmitted the SARS-CoV-2 virus to pet dogs and cats, a wide range of zoo animals, and even wildlife. Transmission of SARS-CoV-2 from humans to animals can lead to outbreaks amongst certain species, which can endanger animal populations and create new sources of human infections. Thus, careful monitoring of animal infections may help protect both animals and humans.

Identifying which animals are susceptible to SARS-CoV-2 would help scientists monitor these species for outbreaks and viral circulation. Unfortunately, testing whether SARS-CoV-2 can infect different species in the laboratory is both time-consuming and expensive. To overcome this obstacle, researchers have used computational methods and existing data about the structure and genetic sequences of ACE2 receptors – the proteins on the cell surface that SARS-CoV-2 uses to enter the cell – to predict SARS-COV-2 susceptibility in different species. However, it remained unclear how accurate this approach was at predicting susceptibility in different animals, or whether their correct predictions indicated causal links between ACE2 variability and SARS-CoV-2 susceptibility.

To assess the usefulness of this approach, Mollentze et al. started by using data on the ACE2 receptors from 96 different species and building a machine learning model to predict how susceptible those species might be to SARS-CoV-2. The susceptibility of these species had either been observed in natural infections – in zoos, for example – or had been assessed in the laboratory, so Mollentze et al. were able to use this information to determine how good both their model and previous approaches based on the sequence of ACE2 receptors were.

The results showed that while the model was quite accurate (it correctly predicted susceptibility to SARS-CoV-2 about 80% of the time), its predictions were based on regions of the ACE2 receptors that were not known to interact with the virus. Instead, the regions that the machine learning model relied on were ones that tend to vary more the more distantly related two species are. This indicates that existing computational approaches are likely not relying on information about how ACE2 receptors interact with SARS-CoV-2 to predict susceptibility. Instead, they are simply using information on how closely related the different animal species are, which is much easier to source than data about ACE2 receptors.

Indeed, the sequences of the ACE2 receptors in many species are unknown and the species for which this information is available come only from a few geographic areas. Mollentze et al. also showed that limiting the predictions about susceptibility to these species could mislead scientists when deciding which species and geographic areas to surveil for possible viral circulation.

Instead, it may be more effective and cost-efficient to use animal relatedness to predict susceptibility to SARS-CoV-2. This makes it possible to make predictions for nearly all mammals, while being just as accurate as models based on ACE2 receptor data. However, Mollentze et al. point out that this approach would still fail to narrow down the number of animals that need to be monitored enough for it to be practical. Considering additional factors like how often the animals interact with humans or how prone they are to transmit the virus among themselves may help narrow it down more. Further research is therefore needed to identify the best multifactor approaches to identifying which animal populations should be monitored.

The global spread and high incidence of severe acute respiratory syndrome-associated virus 2 (SARS-CoV-2) has enabled human-to-animal transmission (hereafter, ‘reverse zoonotic’ transmission) in a variety of taxa and environmental contexts. For example, reports have described household transmission to domestic dogs and cats, transmission to lions and tigers in zoos, transmission to farmed mink, and transmission to free-living white-tailed deer (Barrs et al., 2020; Kuchipudi et al., 2022; McAloose et al., 2020; Sit et al., 2020). While the majority of such spillovers are likely to be dead-ends for the virus, repeated introductions followed by onward transmission have been reported in both European mink and North American deer (Kuchipudi et al., 2022; Oreshkova et al., 2020; Oude Munnink et al., 2021). The potential for reverse zoonotic transmission to both threaten wild and domestic animal health and to foster viral evolutionary diversification that could ultimately be re-introduced into humans (Hammer et al., 2021; Oude Munnink et al., 2021) has led to a surge of interest in pro-actively identifying potentially susceptible species in which surveillance could more efficiently detect reverse zoonotic spillover or sustained transmission of SARS-CoV-2.

Evidence-based prioritisation of SARS-CoV-2 surveillance in animals has proven intractable through traditional approaches. Animal infection experiments have demonstrated that a broad range of species spanning multiple mammalian orders are susceptible (Freuling et al., 2020; Munster et al., 2020; Mykytyn et al., 2021; Schlottau et al., 2020; Shi et al., 2020; Zhao et al., 2020). Yet, infection experiments also suggest fine-scaled variation in susceptibility within mammalian orders. For example, Egyptian fruit bats (Rousettus aegyptiacus) can be infected experimentally with SARS-CoV-2, while attempts to establish infection in big brown bats (Eptesicus fuscus) have thus far failed (Hall et al., 2020; Schlottau et al., 2020). This combination of wide host range with variable susceptibility within groups creates logistical and financial challenges to establishing susceptibility using experimental infections, since large numbers of taxa would need to be investigated. Cell culture-based compatibility assays, an alternative to in vivo infections for evaluating potential host range, are more scalable, but how well they map to susceptibility remains unclear.

A large number of computational heuristics have been developed to address the challenge of predicting the host range of SARS-CoV-2 in the absence of comprehensive experimental evidence. Most utilise the amino acid sequence of angiotensin-converting enzyme 2 (ACE2, the host receptor involved in entry of SARS-CoV-2) and base predictions on measures of similarity to human ACE2, reasoning that changes relative to this sequence (particularly at amino acid positions known to interact with the SARS-CoV-2 spike protein) decrease the probability of successful infection (Ahmed et al., 2021; Alexander et al., 2020; Damas et al., 2020; Frank et al., 2020; Kumar et al., 2021; Luan et al., 2020; Melin et al., 2020; Qiu et al., 2020). A related structural approach has compared computational estimates of binding strength (here grouped together as ‘binding affinity’) between the modelled ACE2 protein structures of different species and the SARS-CoV-2 spike protein, reasoning that stronger binding supports susceptibility (Huang et al., 2020; Lam et al., 2020; Rodrigues et al., 2020). A final approach predicted whether binding strength was above a fixed threshold which was assumed to represent susceptibility and capacity for onward transmission, using ecological and physiological traits of host species (Fischhoff et al., 2021). A limitation common to all efforts to date is that they have focused on untested surrogates of susceptibility, namely ACE2 similarity or binding affinity. This has two immediate drawbacks. First, how these surrogates map to susceptibility is unknown, leading to the development of largely arbitrary thresholds for deciding risk of infection. Second, since no model has been trained or validated using observed data on susceptibility (the outcome of interest), standard metrics of predictive value (e.g., sensitivity, specificity, accuracy) have not been quantified.

The recent accumulation of reports of SARS-CoV-2 spillover to various species, alongside ongoing efforts to characterise the host range of sarbecoviruses experimentally and via virus discovery, provides an exciting opportunity to develop verifiable models of sarbecovirus host range. Here, we assembled a database of compatibility across 96 species and trained sets of machine learning models to distinguish susceptible from non-susceptible species. We sought to understand the relative predictive power of commonly employed sequence and structural binding-based representations of ACE2 and to retrospectively assess the accuracy of previously published prediction methods against our compatibility database. By incorporating new metrics into model training, we also tested whether the exclusive focus on ACE2 is warranted, and whether predictive performance could be improved by considering evolutionary relationships among hosts. Finally, we used the best-performing models to assess geographic and taxonomic biases in the distribution of predicted sarbecovirus hosts that may arise depending on input data available and explore whether sarbecovirus host range prediction can be used to generate actionable insights for ongoing SARS-CoV-2 surveillance efforts in non-human animals.

Our final dataset contained records of interactions between sarbecoviruses and 96 mammalian and avian host species. Reports included natural infections (N=25) and the outcomes of experimental inoculations of captive animals (N=28), where in both cases subsequent detection of viral RNA in, or virus isolation from, nasal or rectal swabs or faeces was considered evidence of susceptibility. We also separately recorded whether species were shown to shed infectious virus (either through virus isolation or infection of other animals), but available data were limited to 23 species, of which only 17 shed infectious virus (Figure 1). To investigate the utility of in vitro data in the absence of natural infections or in vivo data, we also collected evidence of compatibility in cell culture experiments which involved either direct inoculation of cell cultures from each species (N=4) or inoculation of cells expressing heterologous ACE2 (N=39). We treated these different sources of information hierarchically, considering the best available evidence for compatibility or incompatibility in each host species (natural infection > experimental infection > cell culture > heterologous ACE2 cell culture experiments). Importantly, we included records from any strain of Severe Acute Respiratory Syndrome Coronavirus, a species which includes SARS-CoV, SARS-CoV-2, and several related strains. Therefore, a key assumption of our work is that susceptibility to different sarbecoviruses overlaps sufficiently to be modelled jointly. In support of this assumption, most susceptible species in our data (88%) have been linked to at least one sarbecovirus known to use the ACE2 receptor, and among the 10 included host species with records of natural SARS-CoV-2 infection, many were also reported to be susceptible to other sarbecoviruses (N=6), primarily SARS-CoV (N=5, Supplementary file 1).

Figure 1 with 1 supplement see all

Download asset Open asset

To explore the host range of sarbecoviruses and how susceptibility varies within animal groups, we first quantified how susceptible and non-susceptible species were distributed across both a general time-scaled phylogeny based on the accepted patterns of evolutionary divergence among hosts (derived from TimeTree; Kumar et al., 2017) and a maximum likelihood phylogeny constructed from ACE2 orthologs (using the NCBI reference sequence for each species in most cases, see Methods). Sarbecovirus susceptibility was highly conserved within both host phylogenies (Pagel’s λ>0.686, likelihood ratio test p-values ≤0.003; Figure 1A–B). Clustering was particularly strong among non-susceptible species, which tended to be closer to other non-susceptible species than known susceptible host species, both in terms of divergence dates and ACE2 amino acid distances (Figure 1C–F). This remained true when focusing on mammals only (Wilcoxon rank sum test p-value <0.001), indicating that results were not driven by the large evolutionary divergence between birds (mostly non-susceptible) and mammals, which comprised the majority of susceptible records. Further, a phylogeny based on all available ACE2 ortholog sequences closely matched the topology and branch lengths of the time-scaled phylogeny (Baker’s γ correlation coefficient = 0.934, permutation test p-value <0.001; Spearman’s ρ correlation of pairwise distances = 0.870, Figure 1—figure supplement 1). Together, these results show that variation in ACE2 largely mirrors host evolutionary relationships, but it is still possible that current ACE2-based heuristics for predicting susceptibility to SARS-CoV-2 leverage areas of phylogenetic incongruence.

We therefore trained a series of machine learning models using metrics which captured the breadth of currently used representations of the information embedded in ACE2 (sequence dissimilarity and binding affinity with SARS-CoV-2 spike). We also tested whether representing information about individual ACE2 amino acid positions could outperform these commonly used representations. Individual positions were represented by either the observed amino acids (‘AA categorical’), the physicochemical properties of observed amino acids (‘AA properties’), or as a distance to the most common amino acid observed among susceptible species at the same position (‘AA consensus distance’). In terms of sensitivity, no ACE2 representation performed markedly better than a null model which randomly assigned species as ‘susceptible’ or ‘non-susceptible’ in proportion to the frequency of susceptible species in the training data (Figure 2). In contrast, specificity varied more dramatically among representations and all outperformed the null model, indicating that ACE2 can rule out non-susceptible species. However, since non-susceptible species were rare in our training data (N=21), only two models (using either the AA consensus distance or AA properties representations) clearly outperformed the null in terms of overall accuracy. The best-performing model (AA consensus distance) correctly predicted the susceptibility or non-susceptibility of 77 of 96 species (80.2%, 95% binomial confidence interval [CI]: 70.8–87.6%). Among other models, we found that encoding ACE2 amino acid distances relative to the closest susceptible species (‘distance to susceptible’) outperformed the widely used metric of distance to human ACE2 (‘distance to humans’) in terms of specificity while maintaining high sensitivity. The AA categorical model performed considerably worse than the null across all performance metrics (Figure 2). Combining all representations of ACE2 had little effect on performance (74.0% accurate, CI: 64.0–82.4%), suggesting that divergence at key ACE2 amino acids from the susceptible consensus (and most other possible representations of ACE2 sequences) and structural models of binding affinity supply equivalent information. However, mis-classified species differed between individual models, meaning performance was improved by averaging predictions from any two ACE2-based models (ensemble models, Figure 2). This included an ensemble averaging predictions from two iterations of a model trained on the same ACE2 representation, suggesting that these minor improvements in performance stem purely from random variation between models (likely driven by the large numbers of features available for selection in any one representation) and not from unique information in different sequence representations (AA consensus distance/AA consensus distance ensemble, Figure 2).

Figure 2 with 4 supplements see all

Download asset Open asset

High congruence between the overall host phylogeny and the phylogeny of ACE2 (Figure 1, Figure 1—figure supplement 1) raised the possibility that successful prediction of sarbecovirus hosts could arise through the correlation of variation in ACE2 with other evolutionarily conserved features of host biology which define susceptibility, rather than reflecting the disproportionate mechanistic importance of ACE2. Exploiting this correlation would be practically desirable since it would produce models which could be applied to all species, rather than only those with publicly available ACE2 sequences. We therefore tested models of sarbecovirus susceptibility which used phylogenetic eigenvectors to represent the time-scaled phylogeny for all amniotes (the taxonomic clade including both mammals and birds). The phylogeny-only model performed equivalently to the best ACE2-only models, correctly predicting susceptibility labels of 81 species (84.4%, CI: 75.5–91.0%) with a nearly identical sensitivity alongside improved specificity (‘phylogenetic eigenvectors’, Figure 2). Combining phylogeny with either the best-performing ACE2 representation or all ACE2 representations performed similarly to phylogeny alone (Figure 2), suggesting that ACE2 sequences and measures of binding affinity supply information that is redundant with host divergence patterns. Similar performance was also observed when averaging numeric predictions from individual ACE2-based and phylogeny-only models (‘All ACE2 representations/phylogeny ensemble’, Figure 2; 81.2% accurate, CI: 72.0–88.5%).

Deeper analysis of the features used by our ACE2-only models to classify host susceptibility supported the hypothesis that much of their accuracy derives from correlation with host phylogeny. The model trained with all ACE2-representations (but not phylogeny) retained 25 features, of which just two were clearly linked to virus interaction (both representations of binding affinity, Figure 3—figure supplement 1). The remaining 23 features described information about individual ACE2 positions, primarily in the form of amino acid properties (hydrophobicity, polarity, and Van der Waals volume). Several of these features represented the same positions (e.g., hydrophobicity and Van der Waals volume of position 752), meaning the model used just 18 positions in the ACE2 alignment (Figure 3—figure supplement 1). Notably, only one of these positions is known to interact with the SARS-CoV-2 spike protein (position 82), but included positions did contain more phylogenetic information than average (as measured by Shannon entropy, Figure 3; Wilcoxon rank sum test p-value <0.001), suggesting they largely reconstructed host phylogeny. Similar results were observed in the best-performing ACE2-only model (the individual site-based ‘AA consensus distance’ model), which retained no known spike-binding sites (Figure 3). This implies that the inclusion of binding affinity measurements cannot explain the ‘All ACE2 representation’ model’s selection of phylogenetically informative sites over known spike-binding sites. It is possible that the model-selected sites were randomly selected proxies for the same information in correlated spike-binding sites. However, it is unclear why the same sites would have been repeatedly selected from among the large number of correlated alternatives (Figure 3), and why they would be consistently preferred over spike-interacting sites themselves. Further, a model based only on spike-binding sites performed considerably worse than other ACE2-based models (Figure 3—figure supplement 2).

Figure 3 with 3 supplements see all

Download asset Open asset

To ensure that our machine learning approach adequately captured earlier representations of ACE2 sequences, we next compared predictions from our ACE2- and host phylogeny-based models to those of previously published heuristics. Strong correlations between predictions from our combined ACE2-based model and those from earlier studies confirmed that it was broadly representative of earlier approaches (Figure 4—figure supplement 1). Direct comparisons of model performance were difficult given the absence of performance metrics for previously published heuristics and wide variation in the number and identity of species for which prediction was attempted. However, performance appeared to be broadly similar across studies, with binding affinity-based predictions performing best (Huang et al., 2020), followed by distance-based and finally host trait-based approaches (Figure 4). Most distance-based approaches achieved their high accuracy by prioritising sensitivity (i.e., accurate prediction of susceptible species, which dominate current data), and had very low specificity (Figure 4B). The performance of our models based on fitted values (the most comparable approach to previous heuristics) was equivalent to that of the best previously published heuristic (Huang et al., 2020). Formal holdout-based measurements of performance for our ensemble and phylogeny-only models also overlapped with the best performing heuristics, further suggesting comparable performance, despite the more stringent evaluation. This identifies predictions based on host phylogeny as a viable alternative to ACE2 sequences for which availability remains limited.

Figure 4 with 1 supplement see all

Download asset Open asset

Finally, we investigated how including host phylogeny when predicting host range would influence recommendations for targeted surveillance of sarbecovirus spillovers. We produced predictions for all mammals and birds with available ACE2 sequences or which were included in our time-scaled phylogeny, but focus our analysis on mammals since training data for birds were extremely limited (Supplementary file 2). Our best-performing ACE2-based model, the AA consensus distance/AA consensus distance ensemble, predicted 87.6% of mammals to be susceptible (N=178). This is broadly in line with currently available data, where 82.2% of mammals are reported to be susceptible (N=90; Figure 1). Mapping the distribution of terrestrial wild mammal species predicted to be susceptible by this model revealed an apparent hotspot in South-East Asia (Figure 5C), as previously observed by Fischhoff et al., 2021. However, species in this region were not more likely to be susceptible to sarbecoviruses than expected by chance (Figure 5E). Instead, the observed concentration of putatively susceptible species appears to be shaped largely by the increased availability of ACE2 sequences for species from this region (Figure 5A). Indeed, the only regions in which more susceptible species were predicted than expected under a uniform spatial distribution were areas containing very few species with available ACE2 sequences (e.g., the arctic and the Sahara desert).

Figure 5 with 4 supplements see all

Download asset Open asset

By allowing predictions on species without available ACE2 sequences, the phylogeny-only model leads to different conclusions on how surveillance might be targeted. This model reduces the proportion of susceptible mammals from 87.6% to 42.7% (N=4183), but this still implies the need for surveillance in 1785 mammal species (Figure 5—figure supplement 1). In contrast to the ACE2-based predictions, geographic risk was diffused across most of the tropics and subtropics with apparent hotspots limited to areas with very low species diversity (Figure 5D and F). Thus, this model – which performs comparably to or better than ACE2-based models – suggests that clearly defined geographic hotspots of host susceptibility suitable for targeted surveillance are unlikely to exist. Consistent with observed data, in most mammalian taxonomic orders that contained susceptible species, all included species were predicted to be susceptible (Figure 5—figure supplement 2). Interestingly however, primates, rodents, and bats deviated from this pattern, containing several families with no species predicted as susceptible (Figure 5—figure supplement 3, Figure 5—figure supplement 4). Thus, while the wide host range of sarbecoviruses makes it impractical to specifically target surveillance, it may be possible to exclude some groups from consideration.

Frequent spillovers of SARS-CoV-2 from human to non-human hosts has led to a surge in the development of heuristics that seek to predict which species are likely to be susceptible. However, the predictive power of these heuristics or their utility for guiding research and surveillance remained unclear. By cataloguing available data on sarbecovirus host range, we show that while a variety of ACE2-based approaches produce relatively sensitive and specific predictions, these predictions largely derive from strong correlations with host phylogeny, and limitations and biases in their input data limit their actionability. Models based on host phylogeny alone perform equivalently, enabling scalable prediction across nearly all mammals, but imply that in the absence of additional metrics of inter-species exposures, vast numbers of species would need to be surveyed with limited geographic or taxonomic focus.

ACE2 sequences showed surprising accuracy for predicting sarbecovirus host range, both in the formally trained models produced here and in earlier heuristics. However, various lines of evidence suggest the predictive power of ACE2-based models derives primarily from the correlation of ACE2 variation with host phylogeny. First, susceptibility to sarbecovirus infection was highly conserved, clustering in patterns consistent with the evolutionary history of potential host species. Second, ACE2 is also evolutionarily conserved, and a phylogeny derived from ACE2 sequences was highly congruent with one reflecting broader evolutionary history. Third, commonly used approaches for representing ACE2 sequence differences are largely equivalent in the amount of susceptibility information carried (Figure 2, Figure 4—figure supplement 1), despite measuring biologically distinct aspects of ACE2 sequence variation and its interaction with sarbecovirus spike proteins. Fourth, models based on host evolutionary history performed similarly to the best ACE2-based models, while considering ACE2 in addition to evolutionary history provided no detectable improvements. Finally, in ACE2-based models with access to information about individual amino acid positions, most predictive power was derived from phylogenetically-informative sites rather than virus-binding sites, even when measures of binding affinity and sequence similarity were removed. This held true for both rhinolophid bats (the putative ancestral hosts of sarbecoviruses which may have different host-virus interactions due to long-term co-evolutionary dynamics) and other species, with the same features and sites used across all species without affecting predictive accuracy (Figure 3—figure supplement 1, Figure 3—figure supplement 3). Taken together, these results question the exclusive focus on ACE2 as a predictor of sarbecovirus host range. Biologically, the apparent success of ACE2-based models in predicting sarbecovirus host range cannot be interpreted as evidence for a disproportionate mechanistic influence of ACE2 variation in restricting host range. While there is experimental evidence suggesting that variation in ACE2 alone can block infection (e.g., Guo et al., 2020; Thakur et al., 2022), how widely alleles conferring complete resistance occur in nature (particularly in species with no history of sarbecovirus exposure) remains unclear. Indeed, our data show that a broad diversity of ACE2 variation is permissive for infection, so such examples, though biologically verified, may be extremes that do not generalise across taxa. Instead, sarbecovirus host range is likely to be shaped by a range of factors all showing some level of phylogenetic conservation (e.g., receptor distribution, innate immunity, behaviour). These results also mean caution is needed when interpreting the predictive power of specific sites in ACE2 sequences, as phylogenetic correlation means their biological significance cannot be adequately assessed from observational data.

A key drawback of ACE2-based models is that data are only available for a small fraction of potential host species (Figure 5A). This limits their applied use and may also introduce biases since many ACE2 sequences likely stem from existing sarbecovirus research which is geographically biased. One previous effort circumvented the limited availability of ACE2 sequences using host traits believed to follow similar patterns of phylogenetic conservation as ACE2 sequences (Fischhoff et al., 2021). However, although we reiterate that quantitative measures of performance should be considered cautiously, this approach appeared to have reduced performance relative to models directly incorporating host phylogeny (Figure 4). Further, by conditioning a key stage of model development on ACE2 binding, we speculate that this approach may have introduced the same biases plaguing ACE2 sequences. For example, the top 10% of hosts presented as capable of transmitting SARS-CoV-2 by Fischhoff et al., 2021, have a similar geographic distribution to our ACE2-based models (Figure 5C), which in turn mirror overall geographic biases in mammalian ACE2 sequence availability (Figure 5A). Our results show host phylogeny as a viable alternative, which performs comparably to the best ACE2-based models. Anecdotally, this accuracy extends to species only recognised as susceptible after data collection for this study had ended, with 11 of 12 species more recently reported as naturally infected accurately predicted by the phylogeny-based model (Supplementary file 2; Allender et al., 2022; Kuchipudi et al., 2022; Padilla-Blanco et al., 2022; Pereira et al., 2022; United States Department of Agriculture, 2022; United States Department of Agriculture, 2021a; United States Department of Agriculture, 2021b; United States Department of Agriculture, 2021c; Wang et al., 2022). However, beyond data availability, the wide host range of sarbecoviruses represents a more significant practical hurdle, since the scope of surveillance suggested by these models is massive. This issue does not stem from the models, which show high accuracy on current data, but rather are representative of a pattern of widespread susceptibility seen consistently across experimental infections (65.2% of mammals susceptible, CI: 42.7–83.6%), our wider training data (82.2%, CI: 72.7–89.5%), and predictions from both the ACE2-based ensemble and host phylogeny-based models (87.6% and 42.6%, respectively). In essence, the potential host range of sarbecoviruses may be genuinely too broad for accurate predictions to be actionable in the absence of information on the likelihood of cross-species exposures.

Given the broad predicted host range of sarbecoviruses, how should research and surveillance actions proceed? While it may be possible to improve predictive performance further by considering variation in additional host proteins linked to infection (e.g., TMPRSS2; Hoffmann et al., 2020), this is unlikely to be practically useful given expected correlation with phylogeny and the low numbers of species for which sequences of multiple relevant genes will be realistically attainable. Such an approach is also unlikely to significantly reduce the number of susceptible hosts predicted. Alternatively, surveillance might be better informed by targeting hosts in proportion to their predicted risk. However, while conditioning our predictions on the top 30% of species captures the majority of currently known susceptibles (73%), this still means at least 1255 mammals require surveillance while introducing a significant risk of missing key spillovers (Figure 5—figure supplement 1). In the interim, surveillance might therefore target species for which local knowledge indicates high human-animal contact, further refined by our model predictions (e.g., knowledge of likely non-susceptible groups, Figure 5—figure supplement 2, Figure 5—figure supplement 3). Looking further into the future, our data suggest that the majority of susceptible hosts are unlikely to transmit the virus effectively (Figure 1A), and indeed only a limited number of natural spillovers have thus far resulted in onward transmission (Kuchipudi et al., 2022; Oude Munnink et al., 2021). Predicting broader reservoir competence instead of susceptibility to infection may therefore be a path forward to optimise surveillance (Becker et al., 2020). However, the inability of current data to produce models capable of accurately predicting virus shedding (Figure 2—figure supplement 1) indicates a need for greater investment to gather sufficient data on the ability of hosts to transmit sarbecoviruses. Careful interrogation of observed spillover events to evaluate sustained transmission (Kuchipudi et al., 2022; Oude Munnink et al., 2021) and inclusion of naïve hosts in experimental infections to track transmission (Freuling et al., 2020; Shi et al., 2020) would be particularly valuable.

A clear limitation of our approach is the relatively small number of species for which sarbecovirus susceptibility data are available. Combining diverse data sources and expanding the scope to all sarbecoviruses allowed us to evaluate the rapidly growing number of untested heuristics predicting SARS-CoV-2 host range. However, limited numbers of known non-susceptible species in particular made it difficult to distinguish variation in the specificity of alternative approaches with confidence, a problem exacerbated by low levels of overlap between available data and the species for which predictions were available in individual studies (wide CIs, Figure 4B). Although our strategy to maximise sample size meant including data from cell culture-based experiments, we found no evidence to suggest that this skewed results (Figure 2—figure supplement 2). Nevertheless, once more data are available it may be more appropriate to train separate models for each data type (i.e., separate models predicting receptor binding, susceptibility in experimental infections, etc.), as this would allow training an ensemble model to learn the optimal weighting of predictions based on different data types when predicting real-world susceptibility. We were also unable to incorporate polymorphism in ACE2 sequences or suspected variation in susceptibility below the species level, which has been reported in some rhinolophid bats (Guo et al., 2020), since these fine-scale data were unavailable for most species. SARS-CoV-2 is also continuously evolving and this evolution may ultimately alter host range in ways that cannot be predicted by current models (Thakur et al., 2022), which are necessarily based on infections reported with earlier strains. While previous approaches were not explicitly designed for non-SARS-CoV-2 sarbecoviruses, our results suggest the inclusion of additional sarbecoviruses did not adversely impact model performance (Figure 2—figure supplement 3, Figure 2—figure supplement 4, Figure 4—figure supplement 1). This may be because most species (90.8%) included in our dataset have been either linked to or tested for susceptibility to sarbecoviruses known to use ACE2 as an entry receptor (Khaledian et al., 2022; Starr et al., 2022), including all non-susceptible species (Supplementary file 1). Nine susceptible host species have so far only been reported to be infected by sarbecoviruses which are either not known to use ACE2 (due to a lack of published experiments involving the viruses in question) or are suspected to use an alternative receptor (based on an inability to bind the ACE2 orthologs tested thus far); we detected no differences in accuracy when predicting the susceptibility of these species (Figure 2—figure supplement 3, Figure 2—figure supplement 4, Figure 4—figure supplement 1). This convergence in performance for viruses that putatively use different receptors lends further support to our conclusion that sarbecovirus host range is determined by factors other than ACE2 sequence variation. Finally, available susceptibility data may be affected by the same taxonomic and geographic biases affecting ACE2 sequence availability. For example, directly observed data will be biased to countries with capacity for surveillance, in vivo data will be limited to animal models that can be maintained in captivity, and heterologous ACE2 experiments are limited by requiring prior knowledge of ACE2 sequences. However, while this implies that predictions could be periodically updated as new observations become available, we see no mechanism through which current biases could produce a false equivalence between ACE2 sequences and host phylogeny.

Our results imply major challenges for predicting the host range of generalist viruses such as SARS-CoV-2, even when models are reasonably accurate. Both current data and model predictions suggest a significant ongoing risk of reverse zoonosis to a wide range of species, implying a need for continued surveillance among species in contact with humans. Our results provide quantitatively validated predictions of sarbecovirus susceptibility across all mammals, which together with complementary data on exposure risk could be used to narrow such surveillance. However, they also highlight a data gap which currently precludes models that predict not just susceptibility but competence for onward transmission. Such models will be vital to understand the risk that reverse zoonosis poses for establishing novel reservoirs of SARS-CoV-2 genetic diversity that may compromise global efforts to control the ongoing pandemic.

Data and code required to reproduce all analyses are available at https://doi.org/10.5281/zenodo.6552861.

1. 1. Mollentze N
   2. Keen D
   3. Munkhbayar U
   4. Biek R
   5. Streicker DG

   (2022) Zenodo

   Variation in the ACE2 receptor has limited utility for SARS-CoV-2 host prediction.

   https://doi.org/10.5281/zenodo.6552861

1. 1. Ahmed R
   2. Hasan R
   3. Siddiki A
   4. Islam MS

   (2021) Host range projection of SARS-CoV-2: South Asia perspective

   Infection, Genetics and Evolution 87:104670.

   https://doi.org/10.1016/j.meegid.2020.104670

   • PubMed
   • Google Scholar
2. 1. Alexander MR
   2. Schoeder CT
   3. Brown JA
   4. Smart CD
   5. Moth C
   6. Wikswo JP
   7. Capra JA
   8. Meiler J
   9. Chen W
   10. Madhur MS

   (2020) Predicting susceptibility to SARS-CoV-2 infection based on structural differences in ACE2 across species

   FASEB Journal 34:15946–15960.

   https://doi.org/10.1096/fj.202001808R

   • PubMed
   • Google Scholar
3. 1. Allender MC
   2. Adkesson MJ
   3. Langan JN
   4. Delk KW
   5. Meehan T
   6. Aitken-Palmer C
   7. McEntire MM
   8. Killian ML
   9. Torchetti M
   10. Morales SA
   11. Austin C
   12. Fredrickson R
   13. Olmstead C
   14. Ke R
   15. Smith R
   16. Hostnik ET
   17. Terio K
   18. Wang L

   (2022) Multi-species outbreak of SARS-CoV-2 delta variant in a zoological institution, with the detection in two new families of carnivores

   Transboundary and Emerging Diseases 69:e3060–e3075.

   https://doi.org/10.1111/tbed.14662

   • PubMed
   • Google Scholar
4. 1. Baker FB

   (1974) Stability of two hierarchical grouping techniques case 1: sensitivity to data errors

   Journal of the American Statistical Association 69:440.

   https://doi.org/10.2307/2285675

   • Google Scholar
5. 1. Barrs VR
   2. Peiris M
   3. Tam KWS
   4. Law PYT
   5. Brackman CJ
   6. To EMW
   7. Yu VYT
   8. Chu DKW
   9. Perera R
   10. Sit THC

   (2020) SARS-CoV-2 in quarantined domestic cats from COVID-19 households or close contacts, Hong Kong, China

   Emerging Infectious Diseases 26:3071–3074.

   https://doi.org/10.3201/eid2612.202786

   • PubMed
   • Google Scholar
6. 1. Becker DJ
   2. Seifert SN
   3. Carlson CJ

   (2020) Beyond infection: integrating competence into reservoir host prediction

   Trends in Ecology & Evolution 35:1062–1065.

   https://doi.org/10.1016/j.tree.2020.08.014

   • PubMed
   • Google Scholar
7. 1. Bodenhofer U
   2. Kothmeier A
   3. Hochreiter S

   (2011) APCluster: an R package for affinity propagation clustering

   Bioinformatics 27:2463–2464.

   https://doi.org/10.1093/bioinformatics/btr406

   • PubMed
   • Google Scholar
8. Conference

   1. Chen T
   2. Guestrin C

   (2016) XGBoost: a scalable tree boosting system

   Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining. pp. 785–794.

   https://doi.org/10.1145/2939672.2939785

   • Google Scholar
9. 1. Clopper CJ
   2. Pearson ES

   (1934) The use of confidence or fiducial limits illustrated in the case of the binomial

   Biometrika 26:404–413.

   https://doi.org/10.1093/biomet/26.4.404

   • Google Scholar
10. 1. Damas J
    2. Hughes GM
    3. Keough KC
    4. Painter CA
    5. Persky NS
    6. Corbo M
    7. Hiller M
    8. Koepfli KP
    9. Pfenning AR
    10. Zhao H
    11. Genereux DP
    12. Swofford R
    13. Pollard KS
    14. Ryder OA
    15. Nweeia MT
    16. Lindblad-Toh K
    17. Teeling EC
    18. Karlsson EK
    19. Lewin HA

    (2020) Broad host range of SARS-CoV-2 predicted by comparative and structural analysis of ACE2 in vertebrates

    PNAS 117:22311–22322.

    https://doi.org/10.1073/pnas.2010146117

    • PubMed
    • Google Scholar
11. 1. Dupuy D
    2. Helbert C
    3. Franco J

    (2015) DiceDesign and DiceEval: two R packages for design and analysis of computer experiments

    Journal of Statistical Software 65:1–38.

    https://doi.org/10.18637/jss.v065.i11

    • Google Scholar
12. 1. Fauchère JL
    2. Charton M
    3. Kier LB
    4. Verloop A
    5. Pliska V

    (1988) Amino acid side chain parameters for correlation studies in biology and pharmacology

    International Journal of Peptide and Protein Research 32:269–278.

    https://doi.org/10.1111/j.1399-3011.1988.tb01261.x

    • PubMed
    • Google Scholar
13. 1. Fischhoff IR
    2. Castellanos AA
    3. Rodrigues J
    4. Varsani A
    5. Han BA

    (2021) Predicting the zoonotic capacity of mammals to transmit SARS-CoV-2

    Proceedings. Biological Sciences 288:20211651.

    https://doi.org/10.1098/rspb.2021.1651

    • PubMed
    • Google Scholar
14. Preprint

    1. Frank HK
    2. Enard D
    3. Boyd SD

    (2020) Exceptional Diversity and Selection Pressure on SARS-CoV and SARS-CoV-2 Host Receptor in Bats Compared to Other Mammals

    bioRxiv.

    https://doi.org/10.1101/2020.04.20.051656

    • Google Scholar
15. 1. Freuling CM
    2. Breithaupt A
    3. Müller T
    4. Sehl J
    5. Balkema-Buschmann A
    6. Rissmann M
    7. Klein A
    8. Wylezich C
    9. Höper D
    10. Wernike K
    11. Aebischer A
    12. Hoffmann D
    13. Friedrichs V
    14. Dorhoi A
    15. Groschup MH
    16. Beer M
    17. Mettenleiter TC

    (2020) Susceptibility of raccoon dogs for experimental SARS-CoV-2 infection

    Emerging Infectious Diseases 26:2982–2985.

    https://doi.org/10.3201/eid2612.203733

    • PubMed
    • Google Scholar
16. 1. Frey BJ
    2. Dueck D

    (2007) Clustering by passing messages between data points

    Science 315:972–976.

    https://doi.org/10.1126/science.1136800

    • PubMed
    • Google Scholar
17. 1. Galili T

    (2015) Dendextend: an R package for visualizing, adjusting and comparing trees of hierarchical clustering

    Bioinformatics 31:3718–3720.

    https://doi.org/10.1093/bioinformatics/btv428

    • PubMed
    • Google Scholar
18. 1. Grantham R

    (1974) Amino acid difference formula to help explain protein evolution

    Science 185:862–864.

    https://doi.org/10.1126/science.185.4154.862

    • PubMed
    • Google Scholar
19. 1. Guénard G
    2. Legendre P
    3. Peres-Neto P
    4. Freckleton R

    (2013) Phylogenetic eigenvector maps: a framework to model and predict species traits

    Methods in Ecology and Evolution 4:1120–1131.

    https://doi.org/10.1111/2041-210X.12111

    • Google Scholar
20. 1. Guo H
    2. Hu BJ
    3. Yang XL
    4. Zeng LP
    5. Li B
    6. Ouyang S
    7. Shi ZL

    (2020) Evolutionary arms race between virus and host drives genetic diversity in bat severe acute respiratory syndrome-related coronavirus spike genes

    Journal of Virology 94:e00902-20.

    https://doi.org/10.1128/JVI.00902-20

    • PubMed
    • Google Scholar
21. 1. Hall JS
    2. Knowles S
    3. Nashold SW
    4. Ip HS
    5. Leon AE
    6. Rocke T
    7. Keller S
    8. Carossino M
    9. Balasuriya U
    10. Hofmeister E

    (2020) Experimental challenge of a North American bat species, big brown bat (Eptesicus fuscus), with SARS-CoV-2

    Transboundary and Emerging Diseases 68:3443–3452.

    https://doi.org/10.1111/tbed.13949

    • PubMed
    • Google Scholar
22. 1. Hammer AS
    2. Quaade ML
    3. Rasmussen TB
    4. Fonager J
    5. Rasmussen M
    6. Mundbjerg K
    7. Lohse L
    8. Strandbygaard B
    9. Jørgensen CS
    10. Alfaro-Núñez A
    11. Rosenstierne MW
    12. Boklund A
    13. Halasa T
    14. Fomsgaard A
    15. Belsham GJ
    16. Bøtner A

    (2021) SARS-CoV-2 transmission between mink (Neovison vison) and humans, Denmark

    Emerging Infectious Diseases 27:547–551.

    https://doi.org/10.3201/eid2702.203794

    • PubMed
    • Google Scholar
23. 1. Hoffmann M
    2. Kleine-Weber H
    3. Schroeder S
    4. Krüger N
    5. Herrler T
    6. Erichsen S
    7. Schiergens TS
    8. Herrler G
    9. Wu NH
    10. Nitsche A
    11. Müller MA
    12. Drosten C
    13. Pöhlmann S

    (2020) SARS-CoV-2 cell entry depends on ACE2 and TMPRSS2 and is blocked by a clinically proven protease inhibitor

    Cell 181:271–280.

    https://doi.org/10.1016/j.cell.2020.02.052

    • PubMed
    • Google Scholar
24. 1. Huang X
    2. Zhang C
    3. Pearce R
    4. Omenn GS
    5. Zhang Y

    (2020) Identifying the zoonotic origin of SARS-CoV-2 by modeling the binding affinity between the spike receptor-binding domain and host ACE2

    Journal of Proteome Research 19:4844–4856.

    https://doi.org/10.1021/acs.jproteome.0c00717

    • PubMed
    • Google Scholar
25. 1. Jones DT
    2. Taylor WR
    3. Thornton JM

    (1992) The rapid generation of mutation data matrices from protein sequences

    Bioinformatics 8:275–282.

    https://doi.org/10.1093/bioinformatics/8.3.275

    • Google Scholar
26. Book

    1. Juretić D
    2. Lučić B
    3. Zucić D
    4. Trinajstić N

    (1998) Protein transmembrane structure: recognition and prediction by using hydrophobicity scales through preference functions

    In: Párkányi C, editors. Theoretical and Computational Chemistry. Elsevier. pp. 405–445.

    https://doi.org/10.1016/S1380-7323(98)80015-0

    • Google Scholar
27. 1. Kalyaanamoorthy S
    2. Minh BQ
    3. Wong TKF
    4. von Haeseler A
    5. Jermiin LS

    (2017) ModelFinder: fast model selection for accurate phylogenetic estimates

    Nature Methods 14:587–589.

    https://doi.org/10.1038/nmeth.4285

    • PubMed
    • Google Scholar
28. 1. Katoh K
    2. Standley DM

    (2013) MAFFT multiple sequence alignment software version 7: improvements in performance and usability

    Molecular Biology and Evolution 30:772–780.

    https://doi.org/10.1093/molbev/mst010

    • PubMed
    • Google Scholar
29. 1. Kaufman L
    2. Rousseeuw PJ

    (1990)

    Finding Groups in Data

    199–252, Agglomerative nesting (program AGNES), Finding Groups in Data, Hoboken, New Jersey, John Wiley & Sons, Ltd, 10.1002/9780470316801.ch5.

    • Google Scholar
30. 1. Kawashima S
    2. Ogata H
    3. Kanehisa M

    (1999) AAindex: amino acid index database

    Nucleic Acids Research 27:368–369.

    https://doi.org/10.1093/nar/27.1.368

    • PubMed
    • Google Scholar
31. 1. Khaledian E
    2. Ulusan S
    3. Erickson J
    4. Fawcett S
    5. Letko MC
    6. Broschat SL

    (2022) Sequence determinants of human-cell entry identified in ACE2-independent bat sarbecoviruses: a combined laboratory and computational network science approach

    EBioMedicine 79:103990.

    https://doi.org/10.1016/j.ebiom.2022.103990

    • PubMed
    • Google Scholar
32. 1. Klein P
    2. Kanehisa M
    3. DeLisi C

    (1984) Prediction of protein function from sequence properties: discriminant analysis of a data base

    Biochimica et Biophysica Acta 787:221–226.

    https://doi.org/10.1016/0167-4838(84)90312-1

    • PubMed
    • Google Scholar
33. 1. Kuchipudi SV
    2. Surendran-Nair M
    3. Ruden RM
    4. Yon M
    5. Nissly RH
    6. Vandegrift KJ
    7. Nelli RK
    8. Li L
    9. Jayarao BM
    10. Maranas CD
    11. Levine N
    12. Willgert K
    13. Conlan AJK
    14. Olsen RJ
    15. Davis JJ
    16. Musser JM
    17. Hudson PJ
    18. Kapur V

    (2022) Multiple spillovers from humans and onward transmission of SARS-CoV-2 in white-tailed deer

    PNAS 119:e2121644119.

    https://doi.org/10.1073/pnas.2121644119

    • PubMed
    • Google Scholar
34. Software

    1. Kuhn M
    2. Wickham H

    (2020) Tidymodels: a collection of packages for modeling and machine learning using tidyverse principles, version 0.1.3

    Tidymodels.

    https://www.tidymodels.org
35. 1. Kumar S
    2. Stecher G
    3. Suleski M
    4. Hedges SB

    (2017) TimeTree: a resource for timelines, timetrees, and divergence times

    Molecular Biology and Evolution 34:1812–1819.

    https://doi.org/10.1093/molbev/msx116

    • PubMed
    • Google Scholar
36. 1. Kumar A
    2. Pandey SN
    3. Pareek V
    4. Narayan RK
    5. Faiq MA
    6. Kumari C

    (2021) Predicting susceptibility for SARS-CoV-2 infection in domestic and wildlife animals using ACE2 protein sequence homology

    Zoo Biology 40:79–85.

    https://doi.org/10.1002/zoo.21576

    • PubMed
    • Google Scholar
37. 1. Lam SD
    2. Bordin N
    3. Waman VP
    4. Scholes HM
    5. Ashford P
    6. Sen N
    7. van Dorp L
    8. Rauer C
    9. Dawson NL
    10. Pang CSM
    11. Abbasian M
    12. Sillitoe I
    13. Edwards SJL
    14. Fraternali F
    15. Lees JG
    16. Santini JM
    17. Orengo CA

    (2020) SARS-CoV-2 spike protein predicted to form complexes with host receptor protein orthologues from a broad range of mammals

    Scientific Reports 10:16471.

    https://doi.org/10.1038/s41598-020-71936-5

    • PubMed
    • Google Scholar
38. Preprint

    1. Liu Y
    2. Hu G
    3. Wang Y
    4. Zhao X
    5. Ji F
    6. Ren W
    7. Gong M
    8. Ju X
    9. Zhu Y
    10. Cai X
    11. Wu J
    12. Lan X
    13. Xie Y
    14. Wang X
    15. Yuan Z
    16. Zhang R
    17. Ding Q

    (2021) Functional and Genetic Analysis of Viral Receptor ACE2 Orthologs Reveals a Broad Potential Host Range of SARS-CoV-2

    bioRxiv.

    https://doi.org/10.1101/2020.04.22.046565

    • Google Scholar
39. 1. Luan J
    2. Lu Y
    3. Jin X
    4. Zhang L

    (2020) Spike protein recognition of mammalian ACE2 predicts the host range and an optimized ACE2 for SARS-CoV-2 infection

    Biochemical and Biophysical Research Communications 526:165–169.

    https://doi.org/10.1016/j.bbrc.2020.03.047

    • PubMed
    • Google Scholar
40. Book

    1. Lundberg SM
    2. Lee SI

    (2017)

    A unified approach to interpreting model predictions

    In: Guyon I, Luxburg UV, Bengio S, Wallach H, Fergus R, Vishwanathan S, Garnett R, editors. Proceedings of the 31st International Conference on Neural Information Processing Systems. Curran Associates Inc. pp. 4765–4774.

    • Google Scholar
41. 1. McAloose D
    2. Laverack M
    3. Wang L
    4. Killian ML
    5. Caserta LC
    6. Yuan F
    7. Mitchell PK
    8. Queen K
    9. Mauldin MR
    10. Cronk BD
    11. Bartlett SL
    12. Sykes JM
    13. Zec S
    14. Stokol T
    15. Ingerman K
    16. Delaney MA
    17. Fredrickson R
    18. Ivančić M
    19. Jenkins-Moore M
    20. Mozingo K
    21. Franzen K
    22. Bergeson NH
    23. Goodman L
    24. Wang H
    25. Fang Y
    26. Olmstead C
    27. McCann C
    28. Thomas P
    29. Goodrich E
    30. Elvinger F
    31. Smith DC
    32. Tong S
    33. Slavinski S
    34. Calle PP
    35. Terio K
    36. Torchetti MK
    37. Diel DG

    (2020) From people to Panthera: natural SARS-CoV-2 infection in tigers and lions at the Bronx zoo

    MBio 11:e02220-20.

    https://doi.org/10.1128/mBio.02220-20

    • PubMed
    • Google Scholar
42. 1. Melin AD
    2. Janiak MC
    3. Marrone F
    4. Arora PS
    5. Higham JP

    (2020) Comparative ACE2 variation and primate COVID-19 risk

    Communications Biology 3:641.

    https://doi.org/10.1038/s42003-020-01370-w

    • PubMed
    • Google Scholar
43. 1. Minh BQ
    2. Schmidt HA
    3. Chernomor O
    4. Schrempf D
    5. Woodhams MD
    6. von Haeseler A
    7. Lanfear R

    (2020) IQ-TREE 2: new models and efficient methods for phylogenetic inference in the genomic era

    Molecular Biology and Evolution 37:1530–1534.

    https://doi.org/10.1093/molbev/msaa015

    • PubMed
    • Google Scholar
44. 1. Munster VJ
    2. Feldmann F
    3. Williamson BN
    4. van Doremalen N
    5. Pérez-Pérez L
    6. Schulz J
    7. Meade-White K
    8. Okumura A
    9. Callison J
    10. Brumbaugh B
    11. Avanzato VA
    12. Rosenke R
    13. Hanley PW
    14. Saturday G
    15. Scott D
    16. Fischer ER
    17. de Wit E

    (2020) Respiratory disease in rhesus macaques inoculated with SARS-CoV-2

    Nature 585:268–272.

    https://doi.org/10.1038/s41586-020-2324-7

    • PubMed
    • Google Scholar
45. 1. Mykytyn AZ
    2. Lamers MM
    3. Okba NMA
    4. Breugem TI
    5. Schipper D
    6. van den Doel PB
    7. van Run P
    8. van Amerongen G
    9. de Waal L
    10. Koopmans MPG
    11. Stittelaar KJ
    12. van den Brand JMA
    13. Haagmans BL

    (2021) Susceptibility of rabbits to SARS-CoV-2

    Emerging Microbes & Infections 10:1–7.

    https://doi.org/10.1080/22221751.2020.1868951

    • PubMed
    • Google Scholar
46. 1. Oreshkova N
    2. Molenaar RJ
    3. Vreman S
    4. Harders F
    5. Oude Munnink BB
    6. Hakze-van der Honing RW
    7. Gerhards N
    8. Tolsma P
    9. Bouwstra R
    10. Sikkema RS
    11. Tacken MG
    12. de Rooij MM
    13. Weesendorp E
    14. Engelsma MY
    15. Bruschke CJ
    16. Smit LA
    17. Koopmans M
    18. van der Poel WH
    19. Stegeman A

    (2020) SARS-CoV-2 infection in farmed minks, the Netherlands, April and May 2020

    Euro Surveillance 25:2001005.

    https://doi.org/10.2807/1560-7917.ES.2020.25.23.2001005

    • PubMed
    • Google Scholar
47. 1. Oude Munnink BB
    2. Sikkema RS
    3. Nieuwenhuijse DF
    4. Molenaar RJ
    5. Munger E
    6. Molenkamp R
    7. van der Spek A
    8. Tolsma P
    9. Rietveld A
    10. Brouwer M
    11. Bouwmeester-Vincken N
    12. Harders F
    13. Hakze-van der Honing R
    14. Wegdam-Blans MCA
    15. Bouwstra RJ
    16. GeurtsvanKessel C
    17. van der Eijk AA
    18. Velkers FC
    19. Smit LAM
    20. Stegeman A
    21. van der Poel WHM
    22. Koopmans MPG

    (2021) Transmission of SARS-CoV-2 on mink farms between humans and mink and back to humans

    Science 371:172–177.

    https://doi.org/10.1126/science.abe5901

    • PubMed
    • Google Scholar
48. 1. Padilla-Blanco M
    2. Aguiló-Gisbert J
    3. Rubio V
    4. Lizana V
    5. Chillida-Martínez E
    6. Cardells J
    7. Maiques E
    8. Rubio-Guerri C

    (2022) The finding of the severe acute respiratory syndrome coronavirus (SARS-CoV-2) in a wild eurasian river otter (Lutra lutra) highlights the need for viral surveillance in wild mustelids

    Frontiers in Veterinary Science 9:826991.

    https://doi.org/10.3389/fvets.2022.826991

    • PubMed
    • Google Scholar
49. 1. Pagel M

    (1999) Inferring the historical patterns of biological evolution

    Nature 401:877–884.

    https://doi.org/10.1038/44766

    • PubMed
    • Google Scholar
50. 1. Paradis E
    2. Claude J
    3. Strimmer K

    (2004) APE: analyses of phylogenetics and evolution in R language

    Bioinformatics 20:289–290.

    https://doi.org/10.1093/bioinformatics/btg412

    • PubMed
    • Google Scholar
51. 1. Pereira AHB
    2. Vasconcelos AL
    3. Silva VLB
    4. Nogueira BS
    5. Silva ACP
    6. Pacheco RC
    7. Souza MA
    8. Colodel EM
    9. Ubiali DG
    10. Biondo AW
    11. Nakazato L
    12. Dutra V

    (2022) Natural SARS-CoV-2 infection in a free-ranging black-tailed marmoset (Mico melanurus) from an urban area in mid-west Brazil

    Journal of Comparative Pathology 194:22–27.

    https://doi.org/10.1016/j.jcpa.2022.03.005

    • PubMed
    • Google Scholar
52. 1. Qiu Y
    2. Zhao YB
    3. Wang Q
    4. Li JY
    5. Zhou ZJ
    6. Liao CH
    7. Ge XY

    (2020) Predicting the angiotensin converting enzyme 2 (ACE2) utilizing capability as the receptor of SARS-CoV-2

    Microbes and Infection 22:221–225.

    https://doi.org/10.1016/j.micinf.2020.03.003

    • PubMed
    • Google Scholar
53. Software

    1. R Development Core Team

    (2021) R: A language and environment for statistical computing

    R Foundation for Statistical Computing, Vienna, Austria.

    https://www.r-project.org
54. 1. Revell LJ

    (2012) Phytools: an R package for phylogenetic comparative biology (and other things)

    Methods in Ecology and Evolution 3:217–223.

    https://doi.org/10.1111/j.2041-210X.2011.00169.x

    • Google Scholar
55. 1. Rodrigues J
    2. Barrera-Vilarmau S
    3. M C Teixeira J
    4. Sorokina M
    5. Seckel E
    6. Kastritis PL
    7. Levitt M

    (2020) Insights on cross-species transmission of SARS-CoV-2 from structural modeling

    PLOS Computational Biology 16:e1008449.

    https://doi.org/10.1371/journal.pcbi.1008449

    • PubMed
    • Google Scholar
56. 1. Schlottau K
    2. Rissmann M
    3. Graaf A
    4. Schön J
    5. Sehl J
    6. Wylezich C
    7. Höper D
    8. Mettenleiter TC
    9. Balkema-Buschmann A
    10. Harder T
    11. Grund C
    12. Hoffmann D
    13. Breithaupt A
    14. Beer M

    (2020) SARS-CoV-2 in fruit bats, ferrets, pigs, and chickens: an experimental transmission study

    Lancet Microbe 1:e218–e225.

    https://doi.org/10.1016/S2666-5247(20)30089-6

    • PubMed
    • Google Scholar
57. 1. Shi J
    2. Wen Z
    3. Zhong G
    4. Yang H
    5. Wang C
    6. Huang B
    7. Liu R
    8. He X
    9. Shuai L
    10. Sun Z
    11. Zhao Y
    12. Liu P
    13. Liang L
    14. Cui P
    15. Wang J
    16. Zhang X
    17. Guan Y
    18. Tan W
    19. Wu G
    20. Chen H
    21. Bu Z

    (2020) Susceptibility of ferrets, cats, dogs, and other domesticated animals to SARS-coronavirus 2

    Science 368:1016–1020.

    https://doi.org/10.1126/science.abb7015

    • PubMed
    • Google Scholar
58. 1. Sit THC
    2. Brackman CJ
    3. Ip SM
    4. Tam KWS
    5. Law PYT
    6. To EMW
    7. Yu VYT
    8. Sims LD
    9. Tsang DNC
    10. Chu DKW
    11. Perera R
    12. Poon LLM
    13. Peiris M

    (2020) Infection of dogs with SARS-CoV-2

    Nature 586:776–778.

    https://doi.org/10.1038/s41586-020-2334-5

    • PubMed
    • Google Scholar
59. 1. Soubrier J
    2. Steel M
    3. Lee MSY
    4. Der Sarkissian C
    5. Guindon S
    6. Ho SYW
    7. Cooper A

    (2012) The influence of rate heterogeneity among sites on the time dependence of molecular rates

    Molecular Biology and Evolution 29:3345–3358.

    https://doi.org/10.1093/molbev/mss140

    • PubMed
    • Google Scholar
60. 1. Starr TN
    2. Zepeda SK
    3. Walls AC
    4. Greaney AJ
    5. Alkhovsky S
    6. Veesler D
    7. Bloom JD

    (2022) ACE2 binding is an ancestral and evolvable trait of sarbecoviruses

    Nature 603:913–918.

    https://doi.org/10.1038/s41586-022-04464-z

    • PubMed
    • Google Scholar
61. 1. Thakur N
    2. Gallo G
    3. Newman J
    4. Peacock TP
    5. Biasetti L
    6. Hall CN
    7. Wright E
    8. Barclay W
    9. Bailey D

    (2022) SARS-CoV-2 variants of concern alpha, beta, gamma and delta have extended ACE2 receptor host ranges

    The Journal of General Virology 103:001735.

    https://doi.org/10.1099/jgv.0.001735

    • PubMed
    • Google Scholar
62. Website

    1. United States Department of Agriculture

    (2021a) Confirmation of COVID-19 in Otters at an Aquarium in Georgia

    Accessed October 6, 2022.

    https://www.aphis.usda.gov/aphis/newsroom/stakeholder-info/sa_by_date/sa-2021/sa-04/covid-georgia-otters
63. Website

    1. United States Department of Agriculture

    (2021b) Confirmation of COVID-19 in a Canada Lynx at a Pennsylvania Zoo

    Accessed October 6, 2022.

    https://www.aphis.usda.gov/aphis/newsroom/stakeholder-info/sa_by_date/sa-2021/sa-12/covid-lynx-pa
64. Website

    1. United States Department of Agriculture

    (2021c) Confirmation of COVID-19 in Hyenas at a Colorado Zoo

    Accessed October 6, 2022.

    https://www.aphis.usda.gov/aphis/newsroom/stakeholder-info/sa_by_date/sa-2021/sa-11/covid-hyenas
65. Website

    1. United States Department of Agriculture

    (2022) Confirmed cases of SARS-CoV-2 in animals in the United States

    Accessed October 6, 2022.

    https://www.aphis.usda.gov/aphis/dashboards/tableau/sars-dashboard
66. 1. Wang L
    2. Gyimesi ZS
    3. Killian ML
    4. Torchetti M
    5. Olmstead C
    6. Fredrickson R
    7. Terio KA

    (2022) Detection of SARS-CoV-2 clade B.1.2 in three snow leopards

    Transboundary and Emerging Diseases 69:e3346–e3351.

    https://doi.org/10.1111/tbed.14625

    • PubMed
    • Google Scholar
67. 1. Wilkinson SP
    2. Davy SK

    (2018) Phylogram: an R package for phylogenetic analysis with nested lists

    Journal of Open Source Software 3:790.

    https://doi.org/10.21105/joss.00790

    • Google Scholar
68. 1. Zhao Y
    2. Wang J
    3. Kuang D
    4. Xu J
    5. Yang M
    6. Ma C
    7. Zhao S
    8. Li J
    9. Long H
    10. Ding K
    11. Gao J
    12. Liu J
    13. Wang H
    14. Li H
    15. Yang Y
    16. Yu W
    17. Yang J
    18. Zheng Y
    19. Wu D
    20. Lu S
    21. Liu H
    22. Peng X

    (2020) Susceptibility of tree shrew to SARS-CoV-2 infection

    Scientific Reports 10:16007.

    https://doi.org/10.1038/s41598-020-72563-w

    • PubMed
    • Google Scholar
69. 1. Zimmerman JM
    2. Eliezer N
    3. Simha R

    (1968) The characterization of amino acid sequences in proteins by statistical methods

    Journal of Theoretical Biology 21:170–201.

    https://doi.org/10.1016/0022-5193(68)90069-6

    • PubMed
    • Google Scholar

Author details

1. Nardus Mollentze

   1. School of Biodiversity, One Health & Veterinary Medicine, College of Medical, Veterinary, and Life Sciences, University of Glasgow, Glasgow, United Kingdom
   2. Medical Research Council – University of Glasgow Centre for Virus Research, Glasgow, United Kingdom

   Contribution

   Conceptualization, Data curation, Formal analysis, Investigation, Visualization, Writing - original draft, Writing - review and editing

   For correspondence

   nardus.mollentze@glasgow.ac.uk

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-2452-6416

2. Deborah Keen

   School of Biodiversity, One Health & Veterinary Medicine, College of Medical, Veterinary, and Life Sciences, University of Glasgow, Glasgow, United Kingdom

   Contribution

   Data curation, Investigation

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0002-6735-6330

3. Uuriintuya Munkhbayar

   School of Biodiversity, One Health & Veterinary Medicine, College of Medical, Veterinary, and Life Sciences, University of Glasgow, Glasgow, United Kingdom

   Contribution

   Data curation

   Competing interests

   No competing interests declared

4. Roman Biek

   School of Biodiversity, One Health & Veterinary Medicine, College of Medical, Veterinary, and Life Sciences, University of Glasgow, Glasgow, United Kingdom

   Contribution

   Conceptualization, Supervision, Investigation, Writing - review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-3471-5357

5. Daniel G Streicker

   1. School of Biodiversity, One Health & Veterinary Medicine, College of Medical, Veterinary, and Life Sciences, University of Glasgow, Glasgow, United Kingdom
   2. Medical Research Council – University of Glasgow Centre for Virus Research, Glasgow, United Kingdom

   Contribution

   Conceptualization, Supervision, Funding acquisition, Investigation, Writing - review and editing

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-7475-2705

• 796

  views

• 122

  downloads

• 4

  citations

Views, downloads and citations are aggregated across all versions of this paper published by eLife.