COVID-19, the clinical syndrome caused by the SARS-CoV-2 virus, has rapidly spread globally causing hundreds of millions of infections and over two million deaths. The potential animal reservoirs for SARS-CoV-2 are currently unknown, however sequence analysis has provided plausible potential candidate species. SARS-CoV-2 binds to the angiotensin I converting enzyme 2 (ACE2) to enable its entry into host cells and establish infection. We analyzed the binding surface of ACE2 from several important animal species to begin to understand the parameters for the ACE2 recognition by the SARS-CoV-2 spike protein receptor binding domain (RBD). We employed Shannon entropy analysis to determine the variability of ACE2 across its sequence and particularly in its RBD interacting region, and assessed differences between various species’ ACE2 and human ACE2. Recombinant ACE2 from human, hamster, horseshoe bat, cat, ferret, and cow were evaluated for RBD binding. A gradient of binding affinities were seen where human and hamster ACE2 were similarly in the low nanomolar range, followed by cat and cow. Surprisingly, horseshoe bat (Rhinolophus sinicus) and ferret (Mustela putorius) ACE2s had poor binding activity compared with the other species’ ACE2. The residue differences and binding properties between the species’ variants provide a framework for understanding ACE2–RBD binding and virus tropism.
COVID-19, caused by the novel coronavirus SARS-CoV-2, is a zoonotic disease [1–3] that has thus far resulted in over four million deaths worldwide and over 200 million infections . The virus crossed the species boundary, possibly from bats and potentially through an intermediate species, to humans and has spread through the respiratory route across the globe in the past year. SARS-CoV-2 is a member of the betacoronavirus genera, which includes other coronaviruses like SARS-CoV, that caused severe acute respiratory syndrome in a pandemic in 2002–2004 (resulting in ∼800 deaths in 37 countries), and MERS-CoV which caused Middle Eastern Respiratory Syndrome in 2012 (resulting in over 800 deaths in 27 countries). Both of these viruses originated in other species, SARS-CoV in bats, and MERS-CoV in camels, and crossed the species barrier to humans. Other coronaviruses like HCoV-NL63, HCoV-229E, HCoV-HKU1, and HCoV-O43 are known to cause mild respiratory disease in humans and also crossed the species barrier [5–7]. Interestingly, the latter two viruses which cause more clinically mild disease appear to have derived from bovine coronavirus (BCoV) [8–10], which causes respiratory and intestinal disease in cattle [7,11–13]. Betacoronaviruses as a group have a wide host range which includes several agricultural and companion animal species . In this regard, coronavirus disease in humans appears to arise through zoonotic transfer and has resulted in enormous worldwide morbidity and mortality as well as significant economic loss.
Given that coronaviruses appear to have crossed the species barrier several times in human history with devastating consequences, there is a need to understand the interactions between coronaviruses and their receptors in different animal species [14–17]. This is particularly important since humans interact with companion animals as well as many agricultural species, often in close quarters where respiratory spread can easily occur. In addition to the impact of spread from species to species on human health, coronaviruses can cause devastating effects to animals resulting in morbidity, mortality, and major economic losses [5,7,11,18]. Indeed SARS-CoV-2 has been documented to infect dogs (Canis lupus), cats (Felis catus) , mink (Mustela lutrola and Neovison vison) , and white tailed deer (Odocoileus virginianus)  in nature, and ferrets (Mustela putorius) , hamsters (Cricetulus griseus)  and cows (Bos taurus)  have been experimentally infected. Human to tiger transfer occurred at the Bronx Zoo [24,25]. A mink farm in the Netherlands was ravaged by SARS-CoV-2 infection earlier in the pandemic, and farms in at least ten countries have now reported infections [16,20,26]. Additionally, SARS-CoV-2 is studied in vivo in hamsters (Cricetulus griseus) and ferrets (Mustela putorius) as animal models to understand viral pathology and evaluate therapeutics [16,22]. The need to understand animal susceptibility to coronavirus infection is therefore important to public health, the economy, as well as to establish well understood animal models for therapeutic and vaccine development.
SARS-CoV-2 utilizes its trimeric spike protein to bind to the angiotensin I converting enzyme 2 (ACE2) on target pneumocytes or other host cells [27–33]. This interaction occurs with high affinity, and results in viral membrane fusion to the host cell and initiates the infectious process. SARS-CoV-1 also utilizes ACE2 as a receptor, however their spike proteins bind with lower affinity (31 nM KD) than SARS-CoV-2 (4.2 nM KD) . Crystal structure and electron microscopy analysis of SARS-CoV-2 with ACE2 has revealed the interacting amino acid residues of the spike receptor binding domain (RBD) and the human ACE2 surface[28,30,34,35]. With sequences available for many companion and agriculturally important species, the ability to assess potential spike RBD binding is an important step towards prediction of infection of these alternative hosts. Here we analyze the conservation and diversity of the ACE2 protein in multiple important animal species, with particular emphasis on the region that interacts with SARS-CoV-2 spike RBD. We confirm that ACE2 from humans, cats, hamsters and cows interacts with SARS-CoV-2 spike RBD with high affinity, suggesting that multiple mammalian species may be susceptible to infection with this coronavirus. Somewhat surprisingly, however, horseshoe bat, which was initially thought to be a potential reservoir for SARS-CoV-2, and ferret, a potential animal model for coronavirus infection, both have ACE2 proteins that bind poorly to RBD.
Sequence and structure analysis
ACE2 sequences from 19 species were obtained from the NCBI protein sequence database. A number of multiple sequence alignments were performed using Clustal Omega. Specific multiple sequence alignments for (i) companion animals (Canis lupus familiaris, Felis catus, Mus musculus, Mesocricetus auratus, Oryctolagus cuniculus), (ii) agriculturally important animals (Bos taurus, Capra hircus, Equus asinus, Equus caballus, Gallus gallus, Ovis aries, and Sus scrofa), and (iii) species relevant to animal models or potential reservoirs of interest to vaccine testing research or suspected reservoirs (Macaca mulatta, Macaca fascicularis, Manis javanica, Cricetulus griseus, Mustela putorius, Sus scrofa, Rhinolophus sinicus) were investigated. Neovison vison (American mink) was also included in some analyses. In all sequence alignments the Homo sapiens ACE2 sequence was included as a reference.
Further analysis on the multiple sequence alignments were carried out with a variety of software. Phylogenetic trees and pairwise percent identities were obtained with Clustal Omega  and visualized in UGENE (http://ugene.net). The open-source EMBL Boxshade Server was used to produce the boxshade alignments (https://embnet.vital-it.ch/software/BOX_form.html). The Protein Variability Server (http://imed.med.ucm.es/PVS/) was employed to calculate Shannon Entropy values for each residue position.
For structural analysis, the ACE2 and SARS-CoV-2 Receptor Binding Domain (RBD) complex structure files (PDB) were obtained from the RCSB PDB (PDB ID: 6M17) . Eight contact residues of ACE2 were identified based on the co-crystal structure , and we added 17 more ACE2 residues (nearby residues) that were within at least five angstroms of the RBD region and classified the full list as ‘interaction residues'. Thus, there were a total of 25 interaction residues, with eight known RBD contact residues and 17 nearby residues. Using a de novo python script, these residues were extracted into a separate ‘sequence' and an additional, specific multiple sequence alignment was constructed.
The interaction residue multiple sequence alignments were investigated to give more focused insight into which species would be at risk of infection, with the assumption that residues directly interacting with the SARS-CoV-2 spike protein would be essential for infection. Structural analyses were performed in Visual Molecular Dynamics (VMD, https://www.ks.uiuc.edu/Research/vmd/).
Protein expression and purification
The gene sequences encoding human, bat, cat, hamster, ferret and bovine ACE2 proteins (Genebank accession numbers: Homo sapiens, BAB40370.1; Rhinolophus sinicus, AGZ48803.1; Felis catus, AAX59005.1; Cricetulus griseus, XM_003503235.5; Mustela putorius, BAE53380.1; Bos taurus, XP_024843618.1) were synthesized by Genewiz in a pFuse expression vector (Invivogen). Human, cat, bat, hamster, ferret and bovine ACE2 proteins were produced as fusion proteins to human IgG1 Fc according to our published methods for monoclonal antibody purification [2–4]. Briefly, 30 M HEK293 Freestyle cells were transfected with 293fectin combined with 30 µg of pFuse-based vectors containing the ACE2 constructs. Cells were shaken at 37°C for 4 days with 8% CO2. The supernatant was clarified by centrifugation at 4000 RPM for 5 min followed by filtration through a 0.22 µm filter. The media was concentrated and buffer-exchanged into PBS using Amicon Ultra Centrifugal Filter unit (MWCO = 10,000) (MilliporeSigma) at 4°C. The concentrated media was then loaded onto a protein A-sepharose column (Cytiva) pre-equilibrated with 20 mM sodium phosphate, pH 7.0, followed by washing of the column with 10 column volumes of the same buffer and eluted twice with 1 column volume of 0.1 M glycine-HCl, pH 2.7 into fractions containing 0.1 column volume of 1 M Tris, pH 8. ACE2-Fc proteins were further purified using size exclusion chromatography (SEC) on a Superdex200 Increase 10/300 GL column (Cytiva). The SEC was run on an AKTA Purifier 100 FPLC System (Cytiva) using phosphate buffer (50 mM Sodium Phosphate, 150 mM Sodium Chloride, pH 7.2) with a flow rate of 0.6 ml/min. Fractions containing the dimer peak were collected, concentrated using Amicon Ultra-4 Centrifugal Filters (MWCO = 50 kDa), quantified using 280 nm absorbance on a Nanodrop spectrophotometer (Thermo Fisher Scientific) and resolved by SDS–PAGE and stained with InstantBlue Coomassie Protein Stain (Abcam).
The SARS-CoV-2 RBD gene in plasmid NR-52309 (BEI Resources) was transfected and harvested as described above, but purified using TALON cobalt metal affinity resin (Takara Bio) following the manufacturer's protocol, except that 50 mM, 100 mM, 200 mM and 300 mM imidazole gradient elution fractions (1 column volume of each) were collected. Each elution fraction was resolved by SDS–PAGE, stained, and fractions containing a single RBD band were pooled, buffer-exchanged into PBS and quantified as described above. The monomeric RBD plasmid was created by deleting codons encoding the C-terminal four amino acids (CVNF) which contain an unpaired cysteine from the SARS-CoV-2 RBD plasmid NR-52309. The monomeric RBD was expressed and purified the same way as the original RBD, but was further loaded onto a Superdex75 Increase 10/300 GL column (Cytiva) for size-exclusion purification to remove any remaining dimer fraction. The SEC was run on the AKTA Purifier 100 FPLC System (Cytiva) using phosphate buffer (50 mM Sodium Phosphate, 150 mM Sodium Chloride, pH 7.2) with the flow rate of 0.6 ml/min. Fractions containing the monomer peak were collected, concentrated, quantified and resolved by SDS–PAGE as described above.
Enzyme linked immunosorbent assay
Enzyme linked immunosorbent assay (ELISA) plates (Corning 3690) were coated with 100 ng soluble SARS-CoV-2 RBD at 24°C in 1× PBS (pH 7.4) for one hour. Plates were washed three times with Tris buffered saline pH 7.4 containing 0.1% Tween-20 (TBST) and blocked with 2% milk (Marvel, dried skim milk dissolved in TBST) at room temperature for 1 h. Serially diluted ACE2-Fc were added to the wells in 2% milk/TBST, plates were incubated at room temperature for 1 h, washed four times with TBST and then goat anti-human Fc-HRP (Jackson ImmunoResearch #109-035-098, diluted 1 : 5000 in 2% milk/TBST) was added to the wells. Plates were incubated at room temperature for 30 min and washed five times with TBST, then developed by adding 50 µl TMB substrate solution (Thermo Scientific) per well and incubated at room temperature for 3 min. The HRP-TMB reaction was stopped by adding 50 µl 1.0 N sulfuric acid per well. The optical density at 450 nm was read on a microplate reader (SPECTRAMAX M2, Molecular Devices). Antigen-binding curves and EC50 values were generated and calculated using four parameter logistic regression in GraphPad Prism 8 (GraphPad Software, Inc., San Diego, CA).
Surface plasmon resonance
Surface plasmon resonance of immobilized ACE2 proteins to dimer-containing RBD as the analyte was performed by Biosensor Tools (https://www.biosensortools.com/, Salt Lake City, Utah). ACE2-Fc ligands were captured onto a Protein A surface at four different surface densities. Data were collected in Hepes Buffered Saline (HBS) containing 0.1 mg/ml BSA at 25° C. The RBD analyte (36 kDa) was tested in a 3-fold titration series up to 1 µM and 100 nM for both ACE2 protein surfaces at the four different densities. The data for each data set were fit to a two independent site model. The KD values indicated in Figure 4 are the average of four experiments at different Rmax values for 100 nM analyte.
To assess the potential of SARS-CoV-2 to interact with ACE2 of important companion and agricultural species, we assembled the ACE2 sequences of several organisms (Table 1 and Supplementary Table S1). The human ACE2/RBD cocrystal structure (PDB: 6M17)  was utilized to visualize interacting residues between the SARS-CoV-2 spike RBD and human ACE2 (Figure 1). In ACE2, 8 residues on two helices make direct contact with spike RBD and an additional 17 residues are within 5 angstroms of the RBD. These 25 residues are color coded in Figure 1 as purple (contact) and cyan (nearby), respectively. We focused our evaluation on these interacting residues in the following analyses.
ACE2 residues interacting with SARS-CoV-2 spike RBD.
|Species .||% Sequence Identity with Human ACE2 .|
|Full length ACE2 .||Interacting Residues .|
|Cynomolgus monkey||Macaca fascicularis||95.03||100|
|Rhesus monkey||Macaca mulatta||94.78||100|
|Golden hamster||Mesocricetus auratus||84.47||88|
|Chinese hamster||Cricetulus griseus||84.35||88|
|Species .||% Sequence Identity with Human ACE2 .|
|Full length ACE2 .||Interacting Residues .|
|Cynomolgus monkey||Macaca fascicularis||95.03||100|
|Rhesus monkey||Macaca mulatta||94.78||100|
|Golden hamster||Mesocricetus auratus||84.47||88|
|Chinese hamster||Cricetulus griseus||84.35||88|
The percent identity was calculated across all amino acids of ACE2 (Full length ACE2) or only the interacting residues as shown in Figure 1.
To determine the overall differences between ACE2 of different species, we employed protein sequence alignment as well as variability analysis (Supplementary Figures S1 and S2). All vertebrate ACE2 sequences showed significant homology to human ACE2, with cynomologous and rhesus monkeys being 95% and 94% identical with human (Table 1). Other mammals were between 78–87% identical with human. Horseshoe bat (Rhinolophus sinicus), a potential reservoir for SARS-CoV-2 [33,37] was only 81% and 72% identical through the entire sequence and interacting residues, respectively. The percent identity across the entire ACE2 sequence did not fully correlate with percent identity of the interacting residues. For example, cows show lower homology throughout the entire ACE2 sequence at only 79% compared with the other species, but is 84% identical within the identified RBD interacting residues. In contrast, dogs are 83.3% identical across the entire sequence, but only 76% identical in the interacting residues. A similar lower identity in binding site interacting residues is also seen for rabbits (Table 1). The relative contribution of the interacting residues versus residues outside the RBD binding site is currently not known, although it is expected that the interacting residues are far more important to infection relative to the residues outside of the RBD binding site.
For variability analysis, the structural importance of protein regions, and even individual amino acid residues, can be compared across multiple species. Such diversity analyses initially identified the complementary determining regions within antibodies as important interacting domains with antigen by Kabat and Wu [38,39], and more recently Shannon entropy evaluation has been employed to identify conserved and diverse domains of multiple proteins through multiple sequence analysis[40,41]. First, we aligned the ACE2 sequences and determined their percent identities (Table 1 and Supplementary Table S1). Then, we calculated Shannon entropy (SE) across the ACE2 sequences. Of note, ACE2 is remarkably conserved across its sequence, with few residues exhibiting particularly high variability (Figure 2).
Shannon entropy analysis of ACE2 shows low variability.
Within the interacting residues, 21 of 25 are highly conserved, with SE values below 2. Significantly, no residues had values above 3. The more variable residues are colored red and conserved residues blue in Figure 2. Five residues, N330, G352, D355, R357, and R393, are completely conserved, with SE values of zero (Figure 2 and Supplementary Table S1). Four of these are nearby residues with the RBD, with only R357 being a contact residue. The complete conservation of these residues suggests that they play an important role in the protease function or structural integrity of ACE2. Residues with values between 0 and 1 are K353, which is a contact residue, and L45, Y83, T324, which are nearby residues (Supplementary Table S1). Amino acids with SE values between 1–2 are K26, T27, D30, K31, E35, D38, Q42, M82, and G354. The most diverse residues, with SE values over 2, are Q24, H34, L79, and Q81. Of these, Q24 and H34 are contact residues, with H34 found in a central location in the ACE2–RBD interface (Figure 1B), and having by far the highest SE value at 2.88. Others have analyzed the evolution of ACE2 residues and have found positions 24 and 34 to be undergoing positive evolutionary selection pressure , and suggested that these positions could play a role in predicting infectivity by SARS-CoV-2 .
Since the interacting residues are likely most important for viral interaction with ACE2 on the host cell, we evaluated the residues that differed between the various species’ ACE2 and human ACE2 in this region (Figure 3 and Supplementary Figure S3). As mentioned, horseshoe bat (Rhinolophus sinicus) shares only 18/25 interacting residues (72%) with human ACE2, and only 5/8 contact residues. Specifically, D30E, H34T, Y41H, and M82N (contact residues), and T27M, E35K, and Q81K (nearby residues) are mutated in horseshoe bat ACE2 relative to human ACE2 (Figure 3, upper left), suggesting potentially lower affinity for spike RBD. Pangolin, a possible intermediate host of SARS-CoV-2, has only one contact residue (M82N) and three nearby residues (D38E, L79I, Q81K, and G354H) altered, and both D38E and L79I are conservative changes (Supplementary Figure S3, middle left). Felines, which have had documented natural infection [19,25], have only five active site residues mutated (Q24L, D30E, D38E, M82T, Q81K), and only Q24L and M82T are contact residues (Figure 3, middle left). Dogs, which also have been infected naturally, have similar binding site residues as cat but with the notable exception of H34Y (Supplementary Figure S3, middle right), a residue reported to be important in binding. Cow has four interaction site residues mutated, of which only M82T is a contact residue (Figure 3, middle right), and pig has six mutations where Q24L, D30E, H34L, M82T are contact residues (Supplementary Figure S3 bottom). Ferrets have eight differences, including the contact residues Q24L, D30E, H34Y, M82T and the interacting residues D38E, L79H, and G354R. Mink (Neovison vison) differs from human ACE2 at the same residues as ferret, but G354 is histidine (G354H) in mink but is arginine in ferret (Figure 3, bottom left). Primates appear to have evolved Q at position 81 and M at 82, whereas lower mammals have K at 81 and either N or T at contact residue 82 (Supplementary Figure S3).
Variable residues at the ACE2–RBD interface of individual species.
ACE2 from horseshoe bat and ferret bind poorly to SARS-CoV-2 spike RBD.
Ferret and hamster are animal models, cats have been infected naturally, and bats are potential reservoirs of SARS-CoV-2. Cows serve as a reservoir for bovine coronavirus (BCoV) a respiratory infection of cattle that is a betacoronavirus [11,12] distantly related to SARS-CoV-2 . Of considerable note, two BCoV-related coronaviruses, HCoV-OC43 and HCoV-HKU1, have crossed the species barrier, with OC43 likely from cows to humans to cause ‘common cold' respiratory disease in humans . Whereas BCoV, HCoV-OC43 and HCoV-HKU1 utilize 9-O-acetylated sialoglycans as cellular receptors , and SARS-CoV-2 utilizes ACE2, cows are a potential important species to evaluate for possible SARS-CoV-2 infection as they are a known coronavirus reservoir. If coinfected (e.g. by BCoV and SARS-CoV-2) they could could potentially provide a host for coronavirus recombination, selection, and evolution. From a biochemical standpoint, cows have a somewhat more distantly related ACE2 protein compared with many other vertebrates (78.8% compared with most other species which are over 80%, Supplementary Figure S4), however their binding site residues are more conserved (84%) (Table 1). Therefore, it would be useful to know whether the lower homology across the entire ACE2 sequence prohibits productive ACE2/RBD interaction. We chose cow in addition to cat, bat, hamster and ferret ACE2 to further evaluate for RBD binding due to their (i) importance and (ii) diversity within the RBD binding region compared with human ACE2. The differences in interacting residues for these ACE2 proteins are shown in Figure 3.
To analyze potential differences in binding between ACE2s representing different residues at the RBD interaction surface we expressed human, cat, ferret, hamster, horseshoe bat, and bovine ACE2 as antibody Fc fusion proteins and compared their interaction with SARS-CoV-2 RBD by enzyme linked immunosorbent assay (ELISA). First, we expressed and purified ACE2 (Figure 4A and Supplementary Figure S5) from these six species and tested their binding on highly purified RBD monomer (Figure 4B, Supplementary Figure S6). Notably, we found that commonly used SARS-CoV-2 RBD constructs encode a free cysteine which forms disulfide-linked RBD dimers (Supplementary Figure S6). Such molecules may show aberrantly high affinity to ligands due to avidity effects. Four of the six ACE2 proteins bound RBD monomer in the nanomolar range (Figure 4B) but surprisingly bat and ferret showed little detectible binding at the highest concentrations tested. Bat ACE2 was also unable to bind full-length spike protein (Supplementary Figure S7A) and ferret and bat minimally bound a partially dimeric RBD preparation (Supplementry Figure S6) that could robustly bind the other dimeric ACE2 proteins of the other four species at very low concentrations (Supplementary Figure S7B). By ELISA, the RBD binding in order from highest to lowest affinity was human > hamster > cat > cow > ferret > bat. The EC50s were 0.12 nM for human, 0.26 nM for hamster, 2.88 nM for cat and 10.15 nM for bovine (Figure 4B). Ferret and bat ACE2 binding activities were too low to be determined, however ferret had reproducibly detectible binding over background at the highest concentrations compared with bat which did not. The lower apparent affinity for bovine ACE2 compared with the other species was confirmed by surface plasmon resonance analysis which showed a KD for bovine ACE2 of 36.25 nM versus 7.5 nM for human ACE2 (Figure 4C and Supplementary Figure S8 C). This dissociation constant difference relates primarily to a faster off-rate for bovine ACE2 compared with human ACE2 (Supplementary Figure S8). Of note, despite this lower KD for bovine ACE2, this affinity is very similar to the KD for human ACE2 for the RBD of SARS-CoV-1. For both human and bovine ACE2, the SPR data fit more consistently with a two-site model for interaction, suggesting that the RBD may be multimerizing to produce avidity effects on the chip surface. A potential second site would have 10-fold lower KD values (Supplementary Figure S8), which interestingly, are more in line with the EC50 values of the ELISA (Figure 4B). Thus, our analyses of ACE2–RBD binding shows a gradient of activities with different ACE2 variants derived from different species.
The COVID-19 pandemic has spread rapidly across the globe through human populations, but additionally has also infected several animal species. While zoonotic in origin, it is still unclear which species provided the reservoir for transfer to humans. Coronaviruses as a group have a very wide range of host species, and have jumped the species barrier multiple times to humans [44,45]. While bats appear to be a host species for coronaviruses related to SARS-CoV and SARS-CoV-2, it is possible that an as yet unidentified species serves as the reservoir for this virus. Additionally, it is clear that SARS-CoV-2 can naturally infect other species, such as cats, dogs, deer and mink .
To understand the host range of SARS-CoV-2 as well as identify potential reservoirs for the virus it is critically important to understand (i) details about the identity and binding properties between the virus and its host cell receptor and possible co-receptors, (ii) other biological requirements needed for the virus to replicate and transmit, for example host cell enzymes needed for viral processing, replication, and assembly. For SARS-CoV-2, ACE2 appears to be the major receptor required for cell entry, so understanding its interaction with ACE2 from humans as well as other species is important to enable predictive methods for viral host range. Here we analyze ACE2 diversity across several species, and specifically evaluate binding of SARS-CoV-2 RBD to human, cat, hamster, ferret, horseshoe bat, and bovine ACE2, finding key interacting residues to be highly conserved. Surprisingly, horseshoe bat ACE2, a potential host species for SARS-CoV-2, showed very poor interaction with the RBD. Additionally, ferret, which is an animal model for coronavirus pulmonary infection, had an ACE2 which also had poor interaction with RBD. These results may suggest that horseshoe bats are poor candidates for a zoonotic reservoir for SARS-CoV-2. Indeed, Liu et.al. showed poor binding of bat ACE2 to cells by flow cytometry, and quantified the biochemical affinity at 0.4 µM. Alternatively, many polymorphisms of Rhinolophus sinicus ACE2 have recently been reported, with at least eight variants that have key differences in RBD binding site residues, and may bind with varying affinities to endogenous coronaviruses . Thus, it remains possible that alternative ACE2 polymorphisms in horseshoe bat may allow productive binding to RBD and enable this species to host SARS-CoV-2. Ferret ACE2 also bound very poorly in our assays. Ferret has been experimentally infected as an animal model, however infection efficiency and ACE2 binding ability has been debated . A pseudovirus system could not detect entry mediated by either recombinant ferret or horseshoe bat ACE2 compared with human and several other mammalian species, including cat and cow . Importantly, here we used the RBD of the Wuhan-Hu-1 strain; it is possible that undiscovered precursor SARS-CoV-2 variants could potentially have provided better binding to bat or ferret ACE2 prior to a jump to humans and further mutation events.
Across the six species’ ACE2 proteins, we evaluated differences in eleven of the interacting positions where five are contact residues (Q24, D30, H34, Y41, and M82) and six are nearby residues (T27, E35, D38, L79, Q81, G354). With these eleven positions, however, a total of 16 different residues were sampled. The increase in total residues is due to four positions (24, 34, 79 and 82) having more than one amino acid change in different ACE2 variants. Thus, position Q24 was changed to E (bat) or L (cat and ferret); H34 was changed to T (bat), Q (hamster), or Y (ferret); L79 to M (cow) or H (ferret); M82 to N (bat, hamster) or T (cat, cow, and ferret) (Figure 3). Notably, position H34 had the highest diversity with a Shannon entropy score of 2.86 and L79 and Q24 were the second and third most diverse positions (SE of 2.4 and 2.2, respectively). The remaining positions had SE values between 1–2 except Y41 at 0.57. Thus, a range of diverse amino acid positions, content, and combinations were tested for binding SARS-CoV-2 RBD.
The ACE2 species variants we tested for binding RBD differ in 3 (hamster), 4 (cow), 5 (cat), 7 (bat), or 8 (ferret) binding site residues compared with human. Of the four ACE2–RBD interactions that we could quantify, cow was ∼50-100-fold worse than human and hamster (which were approximately equal) by ELISA. Cat was intermediate between cow and human/hamster and neither horseshoe bat nor ferret could be easily quantified. The M82N and H34Q mutations in hamster did not abrogate binding. Notably, hamster and bat share the M82N change, which produces a new N-linked glycosylation motif (N-X-S/T) in both species which had previously been speculated to play a role in bat's decreased infectivity through ACE2 . While it is possible that different glycosylation patterns may occur between hamster and bat cells, glycosylation itself likely does not inhibit binding since our recombinant proteins were produced in the same HEK293 cell line. Thus, since the same N-linked glycosylation mutation occurs in both hamster and bat, glycosylation seems less likely to play a role in the lower binding of bat ACE2 to RBD. The other amino acid change between human and hamster, at position 34 from histidine in human to glutamine in hamster, also appears to have little effect on binding. The five changes in cat (M82T, Q24L, D30E, D38E, Q81K) have modest impact on affinity, however not enough to prevent human to cat infectivity. Cat and cow share the D30E and M82T changes, however the additional mutation at L79M in cow causes an 3–5-fold decrease in binding activity. Alternatively, the additional Q24L and D38E mutations in cat may abrogate the negative effect of D30E and M82T mutations in cow ACE2. Site-specific mutations of these residues would need to be evaluated to understand the relative contributions of each amino acid change. All of the non-human ACE2 proteins have Q81K, which is found in all non-primate mammals we evaluated. This change does not appear important as hamster and human have approximately equal binding activities. Interestingly mink and ferret show the same altered positions, but have the single residue difference at G354R in ferret and G354H in mink (Figure 3, bottom). SARS-CoV-2 is clearly highly transmissible in mink, but not in ferret . This arginine to histidine change may be important in interacting with the RBD, and would require further mutation and evaluation for binding and infectivity to understand it's role.
Several studies have used sequence homology and/or structural modeling to attempt to predict SARS-CoV-2 RBD binding to various species’ ACE2 in order to predict the virus host range [19,42,44,45,49–51]. In an effort to predict species permissive to infection, Damas et.al. developed a five-tiered scoring scheme based on percent identity of 410 vertebrate species, as well as specific structural features of SARS-CoV-1 or SARS-CoV-2 interactions with ACE2 . They also focused on 25 amino acid residues at the RBD binding interface, however their 25 residues differed from ours in that they included S16, N53, N90, and N322, with the asparagines included as potential glycosylation sites that may impact RBD binding. However, our approach was agnostic in choosing residues that were either (i) known contact residues with the RBD, or (ii) within 5 Å of contact residues. Residues included in our analysis which were not included in Damas et al. were K26, Q81, and N352. Damas et al. also note that the host range of SARS-CoV-2 might be quite broad and suggest new species that should be evaluated for animal models of virus infection, which notably include cows, which scored in their ‘medium' category for predictive binding to RBD. Like our assessment of homology and experimental results, they find that bats score very low in predicted ACE2 binding. Interestingly, using sequence evolution and selection analysis, they identify Q24 and H34 as positions undergoing positive selection and evolution. We identified these as amongst the most variable residues by Shannon entropy analysis, and these also appear to be important residues at the ACE2/RBD interface. As in our analysis, pangolins scored low in potential RBD binding based on interacting residue homology, suggesting that SARS-CoV-2 may bind other pangolin receptors, or have other mechanisms to interact with ACE2. As pangolins are thought to be a possible intermediate host for SARS-CoV-2, much more biochemical and infectivity data with this controversial species should be obtained.
In a different approach, Lam et al.  used structural modeling to predict the change in free energy, ΔΔG, for 215 ACE2 sequences derived from different species. They correlated the ΔΔG with published infectability information to provide a framework to predict which species may be susceptible to SARS-CoV-2 infection. Like other studies, their work suggests a broad range of mammal susceptibility, with the exception of non-placental mammals. They similarly find that horseshoe bats have higher ΔΔG values (i.e. lower affinity), calling into question their susceptibility and potential as a reservoir for SARS-CoV-2. Their predictions are in line with our direct demonstration of lower binding activity for horseshoe bat ACE2.
There have been several studies that have measured the KD between SARS-CoV-2 spike (or RBD) and ACE2, with values ranging from 1.2 to 44 nM [28,30,34,35]. In these studies, differences in the experimental conditions, such as whether ACE2 or RBD was immobilized, and technique used such as biolayer interferometry versus surface plasmon resonance, could account for differences in kinetic values. In our study, we immobilized ACE2 and measured RBD interaction as the analyte using surface plasmon resonance, which is similar to Lan et al. who reported a value of KD = 4.6 nM, which is close to our value of 7.5 nM. However, Lan et al. as well as all of the other studies utilized a 1 : 1 model for binding, whereas we found evidence for a two site complexation on the surface, and applied a two site model that gave KD1 and KD2 values of 7.5 nM and 0.4 nM, respectively. Of considerable note, Forssen et al.  reanalyzed the binding data from Lan et.al. and Tian et al. and found evidence of multimeric interactions, similar to our study. We initially utilized a commonly used RBD expression construct which we found contains a free cysteine, and demonstrated that the expressed protein has ∼40% covalently linked dimer, which may explain the multimeric interactions in all of these studies. When this cysteine was removed, pure monomeric RBD could be produced. Since the RBD exists in close proximity to two other RBD subunits in the spike protein trimer, it is plausible that cooperative interactions between RBDs exist during ACE2 interaction events, and thus the dimeric protein may be a useful biochemical surrogate for these interactions. In ELISA analysis, coating a plate with high concentrations of antigen enables dimeric test molecules, like ACE2-Fc, to bind bivalently. In vivo, where receptors on either the virus or cell surface may be clustered, such events could considerably enhance the interactions providing avidity to enable infection. In this regard, ACE2 variants from species with mutations in the RBD interface may still be permissive to infection because of the cooperative and enhanced avidity of binding events. The lower EC50s in our ELISAs, relative to SPR, are likely due to these avidity effects. In the case of horseshoe bat ACE2, however, we could not detect binding to monomeric or dimeric RBD, suggesting that it may not produce effective complexes for infection at all, at least with this particular horseshoe bat ACE2 isotype.
Avidity interactions would be useful to explore to understand the details of interaction between coronavirus spike RBD and ACE2 on the cell surface. These interactions would be important to inhibit by therapeutic agents, for example by monoclonal antibodies (which also can bind bivalently) targeting the virus. The avidity interaction may decrease the KD (increase the affinity) of bovine ACE2 with spike RBD from 36 nM to 2.5 nM, and 7.5 nM to 0.4 nM for human ACE2, a substantial enhancement of the interaction. Regardless of the mechanisms of interaction, it is clear that bovine ACE2 still has high affinity towards SARS-CoV-2 RBD, albeit with worse binding than human ACE2, but yet can still mediate infection of bovine cells in experimental conditions .
One of the major challenges in developing predictive methods for viral infectivity is the limited amount of both infection and RBD-ACE2 biochemical interaction data for most species. For example, relatively few species have had documented infection in the real world, or even in well-controlled experimental systems [19,42,45,52]. In this regard, detailed biochemical analysis of multiple species’ receptors would provide valuable information that could be used in predictive modeling studies. With the decreased costs of synthetic DNA and high-throughput screening approaches, such analyses could potentially be accomplished rapidly. As a first step in this process, here we evaluate ACE2 from several species for binding SARS-CoV-2 RBD. Cows have the lowest affinity, however it has documented experimental infection by SARS-CoV-2 and is a species of significant agricultural importance that serves as a reservoir for other betacoronaviruses, including at least one, HCoV-O43, that has jumped the species barrier to become an endemic respiratory virus in humans. With the recent finding of naturally infected white-tailed deer , another ruminant, understanding ACE2 residues important for binding and infection is important in these related mammals. Further study into various species’ viral receptors, including the recently discovered neuropilin-2 co-receptor for SARS-CoV-2 , should enable much more accurate and rapid development of predictive algorithms based on viral receptor sequence and structural modeling.
The authors declare that there are no competing interests associated with the manuscript.
This project was supported in part by National Institutes of Health grants R01 HD088400 and GM105826 to V.S.
CRediT Author Contribution
Vaughn V. Smider: Conceptualization, Resources, Supervision, Funding acquisition, Writing — original draft, Writing — review and editing. Jessie L. Gan: Conceptualization, Software, Investigation, Methodology, Writing — original draft. Ruiqi Huang: Supervision, Investigation, Methodology. Abigail Kelley: Investigation, Project administration. Gabrielle Warner: Formal analysis, Investigation. Duncan McGregor: Conceptualization, Supervision, Investigation.
The following reagent was produced under HHSN272201400008C and obtained through BEI Resources, NIAID, NIH: Vector pCAGGS Containing the SARS-Related Coronavirus 2, Wuhan-Hu-1 Spike Glycoprotein Receptor Binding Domain (RBD), NR-52309. We thank David Myszka of Biosensor Tools for surface plasmon resonance data analysis.
These authors contributed equally to this work.