No evidence for basigin/CD147 as a direct SARS-CoV-2 spike binding receptor

The spike protein of SARS-CoV-2 is known to enable viral invasion into human cells through direct binding to host receptors including ACE2. An alternate entry receptor for the virus was recently proposed to be basigin/CD147. These early studies have already prompted a clinical trial and multiple published hypotheses speculating on the role of this host receptor in viral infection and pathogenesis. Here, we report that we are unable to find evidence supporting the role of basigin as a putative spike binding receptor. Recombinant forms of the SARS-CoV-2 spike do not interact with basigin expressed on the surface of human cells, and by using specialized assays tailored to detect receptor interactions as weak or weaker than the proposed basigin-spike binding, we report no evidence for a direct interaction between the viral spike protein to either of the two common isoforms of basigin. Finally, removing basigin from the surface of human lung epithelial cells by CRISPR/Cas9 results in no change in their susceptibility to SARS-CoV-2 infection. Given the pressing need for clarity on which viral targets may lead to promising therapeutics, we present these findings to allow more informed decisions about the translational relevance of this putative mechanism in the race to understand and treat COVID-19.

www.nature.com/scientificreports/ We sought to validate the interaction between the SARS-CoV-2 spike protein and human basigin after observing that the result had not yet been reproduced despite intense interest in the interaction's proposed consequences. Using a variety of sensitive approaches for detecting binding interactions and validated reagents, we were unable to find any supporting evidence for a direct interaction of the SARS-CoV-2 spike protein with basigin. Consistent with this, we found no functional role for BSG in the infection of a human lung cell line with SARS-CoV-2. Based on our findings, we encourage caution in approaches aimed at addressing the current pandemic caused by SARS-CoV-2 which are rooted in the assumption that basigin acts as a viral recognition receptor without further evidence.

Results
We first investigated whether basigin (BSG) expressed on the surface of human cell lines could bind the spike protein of SARS-CoV-2. The previous reports of this interaction had not performed any binding experiments on full-length basigin displayed on the surface of cells 8 . First, we synthesized constructs to recombinantly express the spike protein of SARS-CoV-2. We emulated the published designs of spike constructs previously determined to be folded and functional 17 . Using a mammalian HEK293 expression system to increase the chances that structurally-critical post-translational modifications would be preserved 18 , we produced both the full extracellular domain of the spike protein, and the S1 domain of the spike that mediates all known receptor binding events (Fig. 1a). When HEK293 cells were transiently transfected with cDNA overexpression plasmids for ACE2, the transfected cells became strongly stained by fluorescent tetramers of spike protein in either S1 or full forms (Fig. 1b). However, no similar gain of binding was observed with HEK293 cells transfected with BSG cDNA (Fig. S1). BSG transfection led to significant upregulation of cell-surface BSG (Welch's t-test of BSG transfected vs mock-transfected cells p = 0.01), yet we also noted that these HEK293 cell lines express BSG at high levels even without cDNA overexpression (Fig. 1c). Despite this, spike protein tetramers had no detectable background staining of our HEK293 cells without ACE2 in either our experiment or similar experiments reported with SARS-CoV-2 and HEK293 cells [19][20][21] .
Next, we sought to leverage the high sensitivity of direct biochemical binding assays to determine if these methods could detect any traces of basigin binding. We have previously expressed the ectodomain of the basigin receptor in a functionally active form and used it to discover pathogen ligands including Plasmodium falciparum RH5 11,22,23 . In a HEK293 human cell line, we expressed recombinant forms of the extracellular domains of both the canonical isoform of basigin (BSG) that contains two Ig-like domains and an alternate isoform which contains an additional Ig-like domain (BSG-long) (Fig. 2a). To confirm our recombinant constructs were folded and biochemically active, we probed the basigin constructs with three different monoclonal antibodies known to bind native basigin at the cell surface 12,24 in enzyme-linked immunosorbent assays (ELISAs). All antibodies specifically bound to both of our recombinant basigin isoforms but not a negative control construct of recombinant rat Cd4 tag (Fig. 2b). The protein epitopes recognized by these antibodies retained their conformation in recombinant basigin but less so in basigin that is denatured by heat and reducing agent, with each antibody having between twofold to > tenfold reduced immunoreactivity after treatment (Fig. 2c). To determine significance, we fit loglogistic dose-response models to each protein and antibody combination that had at least two replicates 25 , with Bonferroni-corrected p-values ranging from 0.02 (BSG and MEM-M6/1) to < 0.0001 (all others) when comparing denatured and control curves by F-tests.
With the functionality of our constructs quality-tested, we performed a plate-based binding assay 26 that uses the avidity gains of multimerized proteins to detect even highly transient protein-protein interactions 27 . The SARS-CoV-2 spike proteins and ACE2 gave clear binding signals in both binding orientations as plate-bound baits and reporter-linked preys, yet no signals were observed for either BSG isoform against either spike construct (Fig. 3a). By contrast the known interaction between BSG and Plasmodium falciparum RH5 was readily detected, as was a control low-affinity interaction between human CD200 and CD200R. Notably, these interactions have similar or even weaker affinity than reported for the BSG-spike interaction 11,28 . Finally, in response to recent reports of a mutation in the SARS-CoV-2 spike that is rapidly displacing the reference sequence 29,30 , we also checked whether the D614G variant of the spike could bind BSG; again, we could not detect any interaction (Fig. 3b). In all configurations the signal from BSG binding spike protein was indistinguishable from the background of non-interacting protein pairs and significantly below the known interaction pairs (Welch's t-test of ACE2-spike interactions vs BSG-spike interactions p = 0.0002) (Fig. 3c).
While these results strongly suggested that the spike protein of SARS-CoV-2 cannot directly bind basigin as an entry receptor, to account for the possibility of basigin acting through less direct routes in vivo, we tested whether basigin has any functional role in viral infection of a commonly used lung epithelial cell line (CaLu-3) 1,31 . We transduced Cas9-expressing CaLu-3 cells with single guide (sg) RNAs targeting ACE2, BSG, and, as a negative control against a similarly abundant receptor, beta-2-microglobulin (B2M) to decrease cell surface MHC class I expression 32 . We achieved a strong decrease in cell surface expression of each protein encoded by these genes (Fig. S2). We then challenged each cell line with an infectious strain of SARS-CoV-2 across a range of doses, measuring which cells the virus entered by staining against the viral nucleocapsid protein (Fig. 4a). Targeting of the known ACE2 receptor by CRISPR-Cas9 led to a near-complete loss of infectivity (two-way ANOVA of ACE2 infection vs B2M control Bonferroni-corrected p-value < 0.001), despite these cells retaining very high expression of BSG (Fig. 4b). In contrast, genetic removal of BSG had no significant effect on infectivity, with similarly high rates of viral entry seen compared to the B2M control (two-way ANOVA of BSG infection vs B2M infection Bonferroni-corrected p-value > 0.90) and to the parental CaLu-3 cells having no genes knocked out.

Discussion
Identifying the host receptors which a virus can recognize is an important step in mechanistically explaining viral infection, and can offer insight in a virus' cellular tropism and factors influencing susceptibility. Despite the importance of determining precisely which entry receptors SARS-CoV-2 uses to infect human cells, there remains considerable uncertainty amid multiple claims of several viral receptors with variable qualities of data to support these assertions 8,[33][34][35][36] . We investigated one of the most prominent claims, that human BSG acts as an alternate receptor for the virus to interact with, which has been the topic of several studies, news and review articles, and a clinical trial 10,14,15,[37][38][39][40][41][42] . Our access to established tools and reagents from previous work studying BSG's role as a host receptor in Plasmodium infection allowed us to rapidly investigate BSG as a SARS-CoV-2 receptor. Despite validating the functionality of all our reagents, we were unable to detect any binding in biochemical or cell-based BSG is highly expressed on many cell types throughout the body, including activated lymphocytes and red blood cells, forming the basis of the Ok blood grouping system 43 . Notably, SARS-CoV-2 has not been found to enter red blood cells 44 . However, the possibility that BSG could act as an accessory binding receptor for the virus has been speculated in several publications to possibly explain in part the link between SARS-CoV-2 infection and hematological symptoms in patients 39,40,45,46 . Our data suggest this hypothesis should be treated cautiously. Similarly, if our negative findings are replicated, it would necessitate a re-interpretation of the clinical trial involving injections of anti-BSG monoclonal antibodies, as any patient benefit would be more likely explained by alternative hypotheses such as immune modulation as opposed to direct blockage of viral invasion through BSG. Hypotheses relying on BSG binding to explain viral tropism may also need closer reconsideration 14 .
A recent genomic study investigating variants in genes reported to be linked to SARS-CoV-2 infection failed to find evidence of BSG variants enriched in COVID-19 patients, despite identifying associations with more established viral entry factors such as TMPRSS2 47 . Combined with our own infection tests in lung cells with BSG genetically ablated, we see no evidence supporting a direct role for BSG in viral infection. While our results all  www.nature.com/scientificreports/ point to the lack of a direct role for BSG, as mentioned above, BSG could still have some biological relevance through indirect routes 48 . For example, through its previously-described role in the development of the immune system, BSG could indirectly influence COVID-19 clinical progression 49 . Although our findings are negative, they nevertheless carry important potential implications to both our understanding of the basic biology of SARS-CoV-2 and efforts to translate knowledge of the virus' host receptors into therapeutics. We encourage greater study in confirming the mechanisms that have been proposed, not just for BSG but also for the multiple other putative viral receptors, so as to resolve the uncertainty around whether SARS-CoV-2 utilizes any receptors beyond ACE2 during infection.

Materials and methods
Expression construct design. cDNA expression constructs were taken from a previously-assembled library of full-length human cDNAs in human expression vectors 50 . The BSG construct was cloned from a copy (Origene #RG203894) of the canonical 2-domain isoform of BSG (NM_198589.1), while the ACE2 cDNA (NM_021804.2) was expressed from a similar expression vector utilizing a CMV promoter (Geneocopia #EX-U1285-M02). A small amount of MCT4 (NM_001042422.1), which is known to facilitate BSG surface expression 51 , was co-transfected along with BSG at a 1:4 ratio to achieve significant BSG overexpression. The recombinant human BSG ectodomain constructs have been previously described 11,23 and span from M1-L206 www.nature.com/scientificreports/ (BSG) or M1-L322 (BSG-long). The extracellular domain truncations for the SARS-CoV-2 spike proteins have also been previously published 17 . The S1 domain was defined as spanning Q14-Y647, while the full spike ectomain spanned Q14-K1211. The endogenous viral signal peptide was replaced with an efficient mouse antibody signal peptide 52 . As previously described, the full spike ectodomain was mutated at its polybasic protease cleavage site (682-685 RRAR to SGAG), had a proline stabilizing mutation introduced (986-987 kV to PP), and to mimic the natural trimerized structure of the spike had a foldon trimerization domain introduced at its C-terminus. The ACE2 ectodomain spanned M1-S740, retaining its endogenous signal peptide.

Recombinant protein expression and purification. Human embryonic kidney (HEK)-293E cells were
transiently transfected with polyethylenimine as previously described 53,54 . Per 100 mL of cells, 50 µg of plasmid was transfected along with 1 µg of a plasmid encoding the biotin-ligase BirA to direct biotinylation of the recombinant proteins 55  Monoclonal antibody ELISAs. Streptavidin-coated 96-well plates (Nunc #436014) were pre-washed in 175 μL hepes-buffered saline (HBS) with 0.1% tween-20 (HBS-T), then blocked in 2% (m/v) bovine serum albumin (BSA, Sigma #A9647) in HBS for 1 h at room temperature. In a separate 96-well plate, a 1:4 dilution series of biotinylated BSG or control protein was prepared in 2% BSA HBS, then 100 μL of the protein dilution transferred to the blocked streptavidin-coated 96-well plate. In the experiments to determine the protein's sensitivity to heat and reduction treatment, one half of the protein sample was denatured by heating at 80 °C for 10 min in the presence of 5% beta-mercaptoethanol. After capturing protein for 1 h at room temperature, plates were washed three times with 150 μL HSB-T. Anti-human basigin monoclonal antibodies were added at the following concentrations: 1.7 μg/mL Ab-1 (Zenonos et al., formerly known as ch6D9, recognizing the IgC2 domain) 12 , 2.2 μg/mL MEM-M6/1 (Abcam #ab666), and 1.3 μg/mL MEM-M6/6 (Abcam #ab119114) 24 . A control mouse anti-rat Cd4 domain 3 + 4 monoclonal antibody (OX68) against the tags of our recombinant proteins was used at a 1.6 μg/mL concentration. After 1 h of incubation with the primary antibody and three HBS-T washes, secondary antibody was added as 1:7000 donkey anti-human IgG (Abcam #ab102407) for ch6D9 and for all other antibodies as 1:3500 goat anti-mouse IgG (Sigma #A9316). Both secondary antibodies were conjugated to alkaline phosphatase. After 45 min of incubation with the secondary antibody, the plates were washed again three times with HBS-T. A substrate of 60 μL 2 mg/mL para-Nitrophenylphosphate (Sigma #P4744) in diethanolamine buffer was added to each well to develop signal over 30 min. Absorbance was measured on a Tecan plate reader at 405 nm.

SARS-CoV-2 infection assays.
The SARS-CoV-2 virus used in this study is the clinical isolate named "SARS-CoV-2/human/Liverpool/REMRQ0001/2020" 31,57 . In total, the stock used was passaged three times in VeroE6 cells, once in CaCo-2 cells and once in CaLu-3 cells. Viral titre was determined by 50% tissue culture infectious dose (TCID 50  Data processing. For flow cytometry experiments, measurement events for analysis were gated on live singlet cells (based on DAPI and forward and side scatter profiles) using FlowJo version 9. No compensation was done because only a single fluorochrome was used for cell staining. Cytometry data was visualized using the CytoML package 58 in R version 3.6.1. For plate-based experiments, raw absorbance values had background subtracted. Background for ELISAs was defined as the minimum absorbance of any well on the measured plate, and background for binding assays was defined as the median absorbance of each respective tetrameric prey. For better comparisons across replicates, these corrected absorbances were rescaled by min-max normalization so that the maximum absorbance on that replicate's entire plate is defined as 1. Statistics on binding assay data were calculated using the t.test function in the R base stats package (version 3.6.1). Two-way ANOVAs on viral infection data were likewise done in R with the aov function for an additive model of viral nucleocapsid-positive cells from viral dose and knockout condition. Statistics on ELISA data were calculated by performing an F-test comparing a two-parameter log-logistic dose response model fitted to the ELISA data to a null model where both the denatured and control protein conditions were assumed to be identical. Model fitting and statistical procedures were done using the drc package in R as previously described 25 . Data is available upon request.