Distinct Phenotypes of SARS-CoV-2 Isolates Reveal Viral Traits Critical for Replication in Primary Human Respiratory Cells

Since entering the human population, SARS-CoV-2 (the causative agent of COVID-19) has spread across the world, causing >40 million infections and >1 million deaths. While large-scale sequencing efforts have identified numerous genetic mutations in SARS-CoV-2 during its circulation, it remains largely unclear whether these changes impact adaptation, replication or transmission of the virus in its new host. Here, we characterized 14 different low-passage replication-competent human SARS-CoV-2 isolates representing all the major European clades observed during the first pandemic wave in early 2020. By integrating viral sequencing data from patient material, viral stocks and passaging experiments, with kinetic virus replication data from non-human Vero-CCL81 cells and primary differentiated human bronchial epithelial cells (BEpCs), we observed several SARS-CoV-2 sequence features that associate with distinct phenotypes. Notably, naturally-occurring substitutions in Orf3a (Q57H) and nsp2 (T85I) were associated with poor replication in Vero-CCL81 cells but not in BEpCs, while SARS-CoV-2 isolates expressing the Spike D614G substitution generally exhibited enhanced replication abilities in BEpCs. Strikingly, low-passage Vero-derived stock preparation of 3 SARS-CoV-2 isolates selected for substitutions at positions 5/6 of E, and were highly attenuated in BEpCs, revealing a key cell-specific function to this region. Rare isolate-specific deletions were also observed in the Spike furin-cleavage site during Vero-CCL81 passage, but these were rapidly selected against in BEpCs, underscoring the importance of this site for SARS-CoV-2 replication in primary human respiratory cells. Overall, our study uncovers natural sequence features in the SARS-CoV-2 genome that determine efficient virus replication and tropism for the human respiratory epithelium.


INTRODUCTION 44
Severe Acute Respiratory Syndrome Coronavirus 2 (SARS-CoV-2), a novel 45 betacoronavirus belonging to the Coronaviridae family, appears to have first entered 46 humans in late 2019 in the Hubei province of China (1). Since then it has spread 47 Fourteen SARS-CoV-2 primary isolates (IMV1-14) were cultured from samples obtained 115 between 10 th March 2020 and 4 th May 2020 (highlighted on Figure 1A). Together with 116 BavPat1, a SARS-CoV-2 isolate collected on 29 th January 2020 in Munich, Germany 117 (27), all viruses were minimally propagated on Vero-CCL81 cells to obtain working 118 stocks (passage 3 (P3) for IMV1-14; passage 2 (P2) for BavPat1). Original Swiss patient 119 material, as well as both P2 and P3 stocks from all isolates, was subjected to next-120 generation sequencing (NGS) to confirm the absence of additional pathogens, and to 121 provide full-length SARS-CoV-2 genomic sequences (Supplementary Tables S1 & S2). 122 Phylogenetic analysis based on the P2 sequences revealed that the isolated viruses 123 were representative of viruses found across Europe during the first half of 2020 (23) 124 ( Figure 1C). 125

126
As expected, all 14 SARS-CoV-2 isolates harbored several amino acid substitutions 127 when compared to the reference strain SARS-CoV-2/Wuhan-Hu-1, derived from a 128 patient in Wuhan, China in December 2019 (28). We noted that during working stock 129 preparation (i.e. a total of 3 passages from original patient material in Vero-CCL81 cells), 130 some of the virus isolates acquired additional amino acid substitutions or deletions that 131 7 isolates (IMV1: Δ 677-688, 17%; IMV14: Δ 679-685, 81%), and was fully retained in the 139 other 12 isolates. It is likely that the low number of passages that we performed to 140 generate working stocks limited the selection of S furin-cleavage site deletions that have 141 been otherwise readily detected by others (29)(30)(31). We conclude that our set of 14 142 isolates (IMV1-14) is representative of viruses circulating in Europe in early 2020. 143 Together with their lack of consensus-adaptation to Vero-CCL81 cells during stock 144 preparation, these isolates should therefore be suitable to functionally characterize 145 potential features of early SARS-CoV-2 human adaptation and replication. 146 147

Characterization of SARS-CoV-2 Isolates in Vero-CCL81 Cells and Primary Human 148
Bronchial Epithelial Cells. Using Vero-CCL81 cells, we assessed the plaque 149 phenotypes and growth kinetics of all 14 SARS-CoV-2 isolates, as well as BavPat1. All 150 viruses plaqued well on this substrate, although there was some heterogeneity in plaque 151 size between isolates: for example, as compared to BavPat1, IMV4 and IMV11 152 appeared to produce slightly smaller plaques (Figure 2A). Greater differences between 153 isolates were observed during multi-cycle growth analysis. While most viruses grew to 154 titers around 10 6 or 10 7 PFU/mL over 72h, some viruses were clearly attenuated: final 155 titers of IMV2, IMV8, IMV9, and IMV14 were 10-100-fold lower than the other isolates, 156 and kinetic analysis showed that IMV1, IMV3 and IMV13 grew slower than other isolates human bronchial epithelial cell (BEpC) model ( Figure 3A, (32)). BEpCs were grown at 161 air-liquid interface (ALI) for a minimum of four weeks, and their differentiation into a 162 pseudostratified respiratory epithelium was validated by measuring increased trans-163 epithelial electrical resistance (TEER) and epithelium-specific cell and tight junction 164 markers, such as β -tubulin and zona occludens protein 1 (ZO-1) (Figures 3B & C). 165 BEpCs were infected from the apical side with each SARS-CoV-2 isolate and viral titers 166 in both the apical and basolateral compartments were monitored over 72h. Strikingly, 167 there were clear differences in viral replication kinetics between the individual isolates in 168 the apical washes of the infected epithelium (Figures 3D & F). BavPat1, as well as 5 169 other isolates (IMV2, IMV7, IMV9, IMV10, and IMV12) grew to high titers (>10 7 PFU/mL) 170 within 72 hours. In stark contrast, 2 isolates (IMV6 and IMV4) were unable to replicate in 171 BEpCs at all, while IMV14 was strongly attenuated in this tissue substrate, only reaching 172 titers below 10 4 PFU/mL. IMV11 also exhibited a notable attenuated growth phenotype, 173 yielding 100-fold lower titers compared to BavPat1, while IMV1, IMV3, IMV5, IMV8, and 174  in Vero-CCL81 cells and BEpCs ( Figure 4A). While some virus isolates replicated very 185 well in both cell types (BavPat1, IMV10, IMV12), other isolates displayed striking host-186 specific phenotypes. Firstly, IMV2, IMV8 and IMV9 replicated relatively well in BEpCs, 187 but were highly attenuated in Vero-CCL81 cells ( Figure 4A). These 3 isolates were 188 unique in this phenotype, and each harbored combined amino acid substitutions in 189 Orf3a (Q57H) and nsp2 (T85I) that were not identified in any other isolate 190 Table S2), suggestive of an association between these specific 191 changes and the observed tropism. In contrast, IMV4, IMV6 and IMV11 exhibited a high 192 replication capacity in Vero-CCL81 cells, but were strongly attenuated in BEpCs, with 193 IMV4 and IMV6 being particularly attenuated ( Figure 4A). Notably, these 3 isolates each 194 harbored amino acid substitutions at positions 5 (V5G/A) or 6 (S6W) of E 195 Table S2). We also noted that IMV14 was highly attenuated in BEpCs 196 and moderately attenuated in Vero-CCL81 cells, which could be due to >80% of the 197 virus population containing a deletion of the furin-cleavage site in S (Δ679-685). As 198 observed by others (21), and apparent following exclusion of the SARS-CoV-2 isolates 199 noted above that exhibited extreme host-cell specific phenotypes, isolates such as 200 BavPat1, IMV5, IMV7, IMV10, IMV12 and IMV13 expressed S G614, and replicated 201 more efficiently in BEpCs than IMV1 and IMV3 harboring S D614 ( Figure 4A & 202 Supplementary Table S2). These comparative replication data across 14 different 203 SARS-CoV-2 isolates associate specific amino acid residues with distinct substrate 204 phenotypes, and reveal that the functional viral traits underlying these residues must 205 have important roles for virus replication in the human respiratory epithelium. 206 Laboratory Stocks. We investigated prevalence of the phenotype-associated SARS-209

(Supplementary
CoV-2 isolate variants in Orf3a, nsp2, E, and S in our different Vero-CCL81 passages, 210 original patient material, and worldwide. Orf3a (Q57H) and nsp2 (T85I) variants were 211 found in 20% of our patient samples, a similar prevalence to that found worldwide for the 212 individual variants ( Figure 4B). Notably, this proportion did not change upon passaging 213 of isolates in Vero-CCL81 cells during stock preparation. In contrast, E protein variants 214 at positions 5/6 and furin-cleavage site deletions in S were all undetectable in worldwide 215 and patient samples, but clearly became more prevalent among our SARS-CoV-2 216 isolates upon passaging in Vero-CCL81 cells (Figures 4C & D). This passaging effect 217 was further exemplified when re-analyzing mNGS data from each individual isolate and 218 its parental patient material: the frequency of E S6W, V5G or V5A increased 219 substantially during stock preparation of IMV4, IMV6 and IMV11, respectively (Figure 220 substrates (29,30). To experimentally define the importance of the SARS-CoV-2 S furin-cleavage site in primary human respiratory cells, we took advantage of two isogenic 233 IMV14 isolate stocks that exhibited differing frequencies of the S furin-cleavage site 234 deletion (P2, 23.5%; and P3, 81.1%, Figure 5A). Comparative passaging for 4 235 independent replicates in Vero-CCL81 cells and BEpCs revealed that IMV14 P2 (23.5% 236 S furin-cleavage site deletion) replicated in both cell systems (Figures 5B & 5D). 237 Furthermore, sequencing of each passage supernatant revealed that, in Vero-CCL81 238 cells, IMV14 P2 generally increased its frequency of S furin-cleavage site deletion from 239 ~20% to 50-80% (3 out of 4 replicates; Figure 5C). In contrast, there was a rapid loss of 240 this deletion variant during passage in BEpCs, with 100% of recovered sequences for all 241 4 replicates harboring an intact S furin-cleavage site ( Figure 5E). Broadly similar results 242 were obtained when using IMV14 P3 (81.1% S furin-cleavage site deletion): the virus 243 replicated well in Vero-CCL81 cells and retained the S furin-cleavage site deletion at 244 ~80% frequency (Figures 5F and 5G). However, IMV14 P3 was highly attenuated in 245 BEpCs, and virus was only recovered in sufficient quantities from 1 of 4 replicates 246 ( Figure 5H). Strikingly, sequencing of this BEpC-recovered virus revealed a rapid loss of 247 the S furin-cleavage site deletion ( Figure 5I). Together, these data provide strong 248 experimental evidence that the S furin-cleavage site is an essential viral trait required, 249 and selected, for efficient SARS-CoV-2 replication in primary human respiratory cells. Herein, our functional characterization of a spectrum of first wave SARS-CoV-2 isolates 254 has revealed several viral traits that are critical, and potentially adaptive, for efficient viral 255 replication in the human respiratory system:

257
Firstly, using fully replication-competent SARS-CoV-2 isolates in primary human 258 bronchial epithelial cells, we observe that viruses expressing the naturally-occurring 259 G614 substitution in S generally replicate more efficiently than otherwise similar viruses 260 expressing the parental D614 version of S. These observations are in-line with previous 261 work using virus-like particle or vesicular stomatitis virus (VSV) pseudotype systems, 262 where S G614 pseudotypes were found to be more infectious than S D614 pseudotypes 263 CoV-2 Orf3a pore, however, the Q57H substitution does not appear to influence Orf3a 281 channel activities (35). Nevertheless, the Orf3a Q57H variant was acquired shortly after 282 introduction into humans, and is not found in early SARS-CoV-2 sequences or in related 283 bat coronaviruses (35), potentially suggesting positive-selection of this residue similar to 284 S D614G. Notably, the mutation leading to the Orf3a Q57H variant also changes the 285 sequence of two other putative (and poorly characterized) overlapping reading frame 286 products, Orf3c and Orf3d (36,37). Further mechanistic studies, including reverse 287 genetics experiments, will be required to determine whether these variants are 288 functionally linked to one another, and how each variant may mechanistically determine 289 this tropism phenotype. While Vero-CCL81 cells are clearly not a physiological substrate 290 for SARS-CoV-2, it may be that the cell-specific tropism that we observe with isolates 291 expressing Orf3a (Q57H) and nsp2 (T85I) variants also occurs in other, non-respiratory, 292 cell-types of the human body, potentially impacting disease pathogenesis. 293 294 Thirdly, we observed that SARS-CoV-2 stock preparation in Vero-CCL81 cells led many 295 of the isolates (6 out of 14) to acquire passage-derived amino acid substitutions in the E 296 protein, which were not observed in patient material (either from our samples or 297 worldwide). Strikingly, substitutions at E positions 5/6 (V5G/A or S6W) were highly 298 attenuated in primary human bronchial epithelial cells but replicated well, and were 299 potentially selected for, in Vero-CCL81 cells. Previous work using SARS-CoV-1 has 300 shown that E deletion viruses are only mildly attenuated in Vero-E6 cells as compared to 301 their attenuation in human cell-lines or in in vivo models (14), and such viruses have 302 been considered as the basis for live-attenuated vaccine concepts (38-41). These data, 303 together with the observations presented here, suggest that SARS-CoV-2 E has important specific functions in human respiratory cells, as well as in vivo, and further 305 indicate that residues at positions 5 and 6 play key roles in E activities in human cells. 306 307 Finally, as also observed by others (29-31), we found that in-frame deletions of the furin-308 cleavage site in S could be selected for and enriched in SARS-CoV-2 isolates during 309 passage in Vero-CCL81 cells. This was apparent even for very low passage stocks as 310 supplemented with 10% FCS, 100 U/mL penicillin, 100 μ g/mL streptomycin, and 2.5 350 µg/mL amphotericin B (#15290018; Gibco). Cells were seeded in 24-well plates and 351 incubated at 37°C for 4-5 days until cytopathic effect was apparent. Supernatants were centrifuged at 1500 rpm for 5 minutes, and 250 μ L of this cleared supernatant (termed 353 passage 1; P1) was used to inoculate a 25 cm 2 flask of freshly-seeded Vero-CCL81 354 cells, which were then cultured in the same way for a further 4-5 days at 37°C. Cell 355 supernatants were harvested, clarified by centrifugation at 1500 rpm for 5 minutes, and 356 aliquoted before freezing at -80°C (termed P2). P2 stocks were verified by our in-house 357 diagnostics service to be PCR-positive for SARS-CoV-2. Following titer determination by 358 plaque assay (32), a P3 working stock was generated by infecting Vero-CCL81 cells at Weaver SC, Shi P-Y. 2020. Spike mutation D614G alters SARS-CoV-2 fitness 520 and neutralization susceptibility. bioRxiv 521 doi: 10.1101/2020.09.01.278689:2020.2009.2001.278689. 522