Differential RNA editing landscapes in host cell versus the SARS-CoV-2 genome

Summary The SARS-CoV-2 pandemic was defined by the emergence of new variants formed through virus mutation originating from random errors not corrected by viral proofreading and/or the host antiviral response introducing mutations into the viral genome. While sequencing information hints at cellular RNA editing pathways playing a role in viral evolution, here, we use an in vitro human cell infection model to assess RNA mutation types in two SARS-CoV-2 strains representing the original and the alpha variants. The variants showed both different cellular responses and mutation patterns with alpha showing higher mutation frequency with most substitutions observed being C-U, indicating an important role for apolipoprotein B mRNA editing catalytic polypeptide-like editing. Knockdown of select APOBEC3s through RNAi increased virus production in the original virus, but not in alpha. Overall, these data suggest a deaminase-independent anti-viral function of APOBECs in SARS-CoV-2 while the C-U editing itself might function to enhance genetic diversity enabling evolutionary adaptation.


INTRODUCTION
2][3] This virus has moderate genetic variability, as it has a proofreading mechanism to correct the errors caused by the RNA-dependent RNA-polymerase (RdRp). 4,5ut, according to sequencing data of SARS-CoV-2, constant accumulation of new mutations has been shown, leading to the emergence of new variants and subvariants that can be more virulent, more transmissible, or both. 6,7Two sources of SARS-CoV-2 viral mutations/variants can be distinguished: (i) random errors that are not corrected by the built-in proofreading mechanism and (ii) host antiviral responses that cause viral genome mutations/variants. 80][11] ROS could oxidize nucleic acids to cause viral mutations, hypothesized to be related to the G-U and C-A mutations/variants. 8,12DAR enzymes edit adenosine (A) to inosine (I) in double-stranded RNA (dsRNA) to cause A-G mutations/variants, which play important roles in immune response and immune regulation.During viral infection, ADARs act directly through hypermutation of the viral RNA (vRNA), or indirectly by editing of host transcripts that modulate the cellular response.ADAR1 regulates numerous sensors detecting intracellular dsRNA, like MDA5, PKR, RIG-I, or OAS, and is essential in triggering response to viral infection.Data on ADAR1-mediated RNA editing of SARS-CoV-2 are contradictory with some studies showing A-G changes, 9,13,14 while others find no clear evidence of ADAR1 editing in SARS-CoV-2. 15,16The analysis by Picardi et al. shows low levels of editing observed early after infection (4 h post infection [hpi]), when ADAR1 and interferon activation is low, while after 24 hpi, higher levels of A-G editing were observed, accounting for <1% of sites. 14One also has to remember that sequencing and/or polymerase errors can contribute to the detected substitutions.Moreover, according to the RNA antisense purification and mass spectrometry data, ADAR1 and APOBEC1CF were found as proteins interacting with SARS-CoV-2 RNA. 17

ISGs and RNA editing enzymes are upregulated upon infection with SARS-CoV-2 in vitro
APOBECs and ADARs are known antivirals and can be stimulated in a virus-specific manner through interferon signaling and RNA-sensing pathways. 27,28For SARS-CoV-2 variants, it is known that they show different sensitivity to different interferons as we (unpublished data) and others have shown. 29To investigate the expression of interferon-stimulated, APOBEC, and ADAR genes, Caki-1 cells were infected with two different patient isolates representing European original (B.1) and the alpha variant of concern (VOC), respectively (strains EDB-2 and EDB-a-1).The Caki-1 cell line, a human kidney-derived cancer cell line, was chosen for being highly susceptible to SARS-CoV-2, other corona-and respiratory viruses, as well as being interferon competent. 26Cells were inoculated at MOI = 0.1 with either variant.Following infection, cellular RNAs were harvested at 24 or 48 hpi and subjected to 150 bp paired-end RNA Illumina sequencing analysis (150PE RNAseq; Figure 1A).On average, 95.48% (ranging 95.22-95.74%) of reads mapped to human genome in mock infected (control) cells.At 24 hpi, in cells infected with the EDB-2 strain, 29.95% of reads mapped to human genome and 68.37% of reads mapped to the EDB-2 genome.At 48 hpi, 46.12% of reads mapped to human genome and 51.35% reads mapped to EDB-2 genome.In cells infected with the EDB-a-1 strain at 24 hpi, we obtain 46.10% of reads mapped to human genome and 51.85% of reads mapped to the EDBa-1 genome.At 48 hpi, 55.60% reads mapped to human genome and 41.74% reads mapped to the EDB-a-1 genome.
To decipher the nucleotide substitution pattern of SARS-CoV-2 variants, EDB-2 and EDB-a-1 were serially passaged on Caki-1 cells up to passage 4 and vRNA was extracted from the supernatant.The vRNA was subjected to 150PE RNAseq analysis and RNA substitutions were detected using the REDItools package (Figure 1B).In each passage, 99.9% of reads mapped to the corresponding EDB-2 or EDBa-1 genome, respectively.We observed 74 variants with frequency over 1% in EDB-2 P1 (28 variants with frequency over 2%), 57 variants with frequency over 1% in EDB-2 P2 (15 variants with frequency over 2%), 101 variants with frequency over 1% in EDB-2 P3 (30 variants with frequency over 2%), and 73 variants with frequency over 1% in EDB-2 P4 (15 variants with frequency over 2%), while in EDB-a-1 we observed 77 variants with frequency over 1% in P2 (35 variants with frequency over 2%), 135 variants with frequency over 1% in P3 (49 variants with frequency over 2%), and 117 variants with frequency over 1% in P4 (56 variants with frequency over 2%).All variants obtained for examined samples are in Table S1.As already shown in previous studies, 9,14 A-to-I candidates as well as C-to-U candidates displayed very low editing levels, less than 1% in more than 99% of positions.
We analyzed whether there are differences in gene expression between control and infected cells to determine whether virus can impact on basal host cell interferon activation, as previously reported. 14The heatmap in Figure 2A presents the z-scores of the top 100 differentially expressed genes between different time points post infection in Caki-1 cells.The mock infected cells exhibit visibly different gene expression patterns compared to cells infected with EDB-2 or EDB-a-1 at both 24 and 48 hpi.The average Z score at 24 hpi for mock infected cells is À0.38668, rising to 0.257763 for EDB-2 and 0.081738 for EDB-a-1 showing an overall increase in relative gene expression.At 48 hpi with the mock infected control shows an average Z score of À0.36137 rising to 0.194584 for EDB-2 and 0.213967 for EDB-a-1.The greatest difference in expression is demonstrated by NNMT (regulator of epithelial-mesenchymal transition 30 ), AKR1C2, AKR1C3, AKR1C1 (aldo-keto reductase family members 31 ), KCNJ16 (potassium channel 32 ), and LXN (inhibitor of zinc-dependent metallocarboxypeptidases 33 ).Within the 100 upregulated genes gene ontology (GO) pathways such as import into cell (GO:0098657), response to external stimulus (GO:0009605), and response to stress (GO:0006950) were found to be upregulated with FDRs <0.03 through String DB analysis. 34Moreover, from our top 100 differentially expressed genes, three genes (SLC1A3, ZC3HAV1, and DDX58) are also mentioned in the Gencode list of genes involved in the Covid infection with the "updated" status (Gencode: https://www.gencodegenes.org/human/covid19_genes.html, accessed 8/8/2023).
To determine whether the infection with SARS-CoV-2 triggers an immune response cascade in the infected cells related to interferon production and inflammation, we investigated the expression level of interferon-stimulated genes (ISGs), selected on the basis of literature search [35][36][37][38] (Figure 2B).The relative expression heatmap reveals that cells after 24 and 48 hpi of mock infection present a different expression pattern of ISGs than cells infected with EDB-2 and EDB-a-1 variants.The ''classical'' ISGs such as MX1, OAS1/2/3 and more recently identified effectors such as APOBEC3G, TRIM5, ISG15, ADAR, IFITM1/2/3, all show a dramatically increased relative RNA expression at 48 hpi for both EDB-2 and EDB-a-1.Of note, the cells infected with EDB-a-1 at 24 hpi exhibit a different pattern of ISGs expression than all the other infected groups (EDB-a-1 at 48 hpi, and EDB-2 at both 24 and 48 hpi).This is also shown in the classical ISGs listed previously, while their relative RNA expression does increase from the control samples it does not reach the over expression phenotype seen in the other time points.This may be linked to the slower replication of EDB-a-1 compared to EDB-2.At 48 hpi, EDB-a-1 gene expression appears more closely related to the EDB-2 48 hpi expression, which could indicate that EDB-a-1 launches a distinct (but overlapping) pathway of anti-viral defense in cells compared to the EDB-2 strain.
ADARs and the APOBEC family also belong to the ISGs.To assess the detailed expression of these RNA editing enzymes, we focused on their expression during infection.These results are presented on the heatmap in Figure 3. Genes related to editing of ssRNA (APOBEC3B, -3G) and to the editing of dsRNA (ADAR and ADARB1) show higher expression levels in infected cells compared to mock infected controls.The Z score of ADAR's relative RNA expression increases from À1.218 at 24 hpi in the mock infected, to 0.583 in EDB-2 and -0.022 in EDB-a-1, and increases on these values at 48 hpi (mock infected = À1.183,EDB-2 = 0.717, EDB-a-1 = 1.124).APOBEC3B and -3G, display similar patterns to ADAR.However, ADARB1 (encoding the ADAR2 enzyme) expression is not consistent throughout infected and mock infected cells, possibly suggesting higher levels of the proteins encoded by the genes mentioned previously and, consequently, elevated activity of the enzymes.As observed for the ISGs as a whole, upregulation of RNA editing enzymes is lower in EDB-a-1 at 24 hpi but largely aligns with EDB-2 by 48 hpi potentially due to the slower replication of EDB-a-1.There are clear exceptions within the EDB-a-1 expression dataset: AICDA, APO-BEC2, and APOBEC4, which are downregulated and ADARB1, which is strongly upregulated.Overall, the expression of APOBECs and ADARs  is upregulated following infection with both variants of the virus.Therefore, we analyzed the substitutions composition focusing on evidence for RNA editing events.

C-U substitutions are observed at increased frequency in viral genomic RNA
RNA substitutions in infected cells were profiled at single nucleotide resolution with REDItools applying a stringent filtering scheme (Figure 4) and a list of candidate RNA editing sites was obtained.There were 8,181 sites obtained for 24 h mock control and 84,33 sites for 48 h mock control, while for cells infected with EDB-2 strain, 2,301 sites were obtained after 24 h of infection and 4,214 sites after 48 h of infection.For cells infected with EDB-a-1, 4,000 sites were obtained after 24 h of infection and 5,387 sites after 48 h of infection.All sites obtained for examined samples are in Table S2.By far the most common substitution observed in the host cellular transcriptome (mock and infected with SARS-CoV-2 variants) is A-G, while the second and the third most common substitution types are G-A and U-C, respectively.We observe only a slight increase in the frequency of A-G substitution in infected cells compared to mock infected cells, which may suggest that ADAR1-mediated editing of host RNA is not strongly disturbed by SARS-CoV-2 infection.The slight increase in the frequency of A-G substitution in infected cells is consistent with a predominant activation of ADAR1 by the viral infection.As a control, we analyzed the effects of virus on a transcript of specific host gene, CDK13, which is subjected to a relatively high number of A-I edits (ADAR1-specific) in other cancer cell lines (Figure S1). 39or example, chromosomal location 39950745 exhibits an increase in A-I edit after EDB-2 or EDB-a-1 at 24 hpi (Figure S1B).On the other hand, location 39950928 has a basal level of A-I edit that is not increased by viral infection hour (Figure S1C).These data highlight that, in Caki-1 cells, regions of the CDK13 transcript may be impacted by infection with SARS-CoV-2 through over-expression of ADAR1.
We next performed analysis on substitutions located within the genome of EDB-2 and EDB-a-1 variants.Therefore, we focused the REDItools analysis on the vRNA transcripts in the total RNA (rRNA depleted) extracted from infected cells at 24 and 48 hpi.To allow visualization of regions of the SARS-CoV-2 vRNA genome where mutations are potentially more likely to occur, we overlaid substitutions that occurred with a frequency of R2% with the Wuhan SARS-CoV-2 sequence NC_045512v2 (Figure 5A).During single-round infection, in EDB-2 we see the same substitutions across both timepoints (24 and 48 hpi), with seven extra and one lost substitutions in the genome at 48 hpi.The same pattern is observed in EDB-a-1, with even distribution throughout the RNA genome across both timepoints (24 and 48 hpi) and 17 extra and 5 lost substitutions in the genome in EDB-a-1 at 48 hpi.The difference here is in the number of substitutions: EDB-a-1 presents significantly more substitutions over time (***p value<0.001) that meet the criterium of R2% frequency than EDB-2.This indicates that the EDB-2 variant is more genetically stable than EDB-a-1.Moreover, the most frequent substitution observed in EDB-a-1 was C-U while in EDB-2 there was no obvious pattern observed (Figure 5B).These data suggest that there is a distinct (i.e., non-ADAR1) editing of the vRNA or viral genome.
Since the cellular vRNA will be composed of subgenomic transcripts derived from the viral genome as well as some viral genome replicants, we sequenced the viral genome from RNA extracted from the supernatant of viral culture cells, derived from sequential passages.The first passage of EDB-a-1 was not included as not enough material was obtained from the original clinical isolate.Since Caki-1 cells only exhibit CPE at very late time points, the release of subgenomic RNA from lysed cells is small.This was confirmed in the RNA sequencing analysis where cellular RNA reads represented a larger proportion of reads in cell lysate than in supernatant samples.
Similar to the cellular vRNA, EDB-a-1 exhibits an increase in the number of mutations, which amplify over passages (EDB-a-1-P2, 35 substitutions; EDB-a-1-P3, 49 substitutions; EDB-a-1-P4, 56 substitutions with frequency R2%).The EDB-2 strain remains consistent in the number of substitutions throughout the passages (Figure 5C), showing no clear trend and overall decreasing numbers of substitutions with passaging (EDB-2-P1, 28 substitutions; EDB-2-P2, 15 substitutions; EDB-2-P3, 30 substitutions; EDB-2-P4, 15 substitutions that meet R2% frequency).While no clear prevalent substitution was observed in cellular RNA for EDB-2, the first passage of the supernatant derived virus genomes clearly shows a high rate of C-U substitutions, a pattern observed through passaging for EDB-a-1 (Figure 5C).No clear preference for specific nucleotide substitutions is observed in later passages of EDB-2, this is probably due to the small number of overall mutations throughout the genome.While host RNA appears highly sensitive to the viral activation of ADAR1 and shows high A-G substitutions, the vRNA appears largely unaffected.This is not surprising as the activity of both isoforms of ADAR1 (p110 and p150) can affect the cellular genome, whereas only one of these isoforms can operate outside the nucleus (p150), and therefore have an impact on the cytosolic replication of the virus. 39It is also worth noting that in both strains some of the variants that occurred in the first passage are maintained through the next passages.This could hint at the positive role of the RNA editing enzymes in the virus evolution in vitro (Figure 5A), but some of these could also be RdRp-mediated substitutions.In Figure 5A we furthermore highlight the four coding nucleotide variations occurring in EDB-a-1 passaging.A C-U variance in the orf1a/orf1ab polyprotein encodes for an L550F (L730F in the orf1a/1ab polyprotein) and increases slowly but steadily over four passages.A mutation in S, G72R, caused by a G-A variance gains prevalence over four passages.Lastly, two coding mutations are found in Orf3A, K16T, and S166L, caused by an A-C and a C-U variance, respectively.While K16T decreases from a low level over passaging, S166L becomes predominant in passage 4.
In Figure 5D we analyzed the potential consequences of nucleotide variation on coding.CU substitutions result in a silent mutation in 37.5% (18/48) variations, the highest level on par with U-C variance.Coding changes retain amino acid class in 10/48 variations.In contrast, A-G variance is silent only in 15/48 variations, and only 3/48 variations retain amino acid class.While no new stop codons are introduced in A-G variation (in contrast to 3/48 for C-U), potentially significant changes to protein function may occur in 62.5% of variations (class change STOP/ Start/ATG introduction or removal), whereas this is only the case in 41.7% of variations for C-U.
Together, these data suggest that vRNA within the cell, including the viral genome itself, can be subjected to C-U RNA editing events, which is distinct from the host cell that exhibits largely A-G editing events.

Knockdown of APOBECs but not ADAR increases viral replication in a variant-dependent manner
Our results thus far suggest that APOBEC-mediated editing (i.e., C-U) is occurring in both EDB-2 and EDB-a-1 viral genomes as observed through the frequency of C-U substitutions.Since vRNA editing through cellular proteins may significantly impact the mutation rate of coronaviruses, we aimed to assess the impact of APOBECs and ADAR1 activity on viral production in different VOCs using RNAi-mediated knockdown in Caki-1 cells.Knocking down gene expression allows for unbiased cell populations and manipulates the endogenous levels of the protein in contrast to protein overexpression.We selected the cytosolic APOBEC3D and the nuclear APOBEC3B and primarily nuclear ADAR (under the caveat that one isoform, ADAR1 p150 , can operate in the cytosol as well as the nucleus 40 ).Unfortunately, pre-designed and validated siRNAs for 3F and 3G had previously been deemed non-specific, wherefore were not included in this analysis.Two different pre-designed siRNAs for each of the selected APOBECs and ADAR were reverse transfected into Caki-1 cells.At 72 h post transfection, cells were inoculated at MOI = 0.5 with EDB-2 (original), EDB-a-1 (alpha), EDB-d-1 (delta), or EDB-o-10 (omicron, BA.1).Virus production was measured by direct lysis RT-qPCR at 48 hpi and visualized as a percentage relative to a mock transfected control (Figure 6A).The knockdown efficiency of the siRNA used to against APOBEC3B, APOBEC3D, and ADAR was assessed for a reduction in the mRNA of all three proteins to account for any non-specific effects (Figure 6B).Knockdown of APOBEC3B with siRNA 2 led to a significant reduction in the mRNA of APOBEC3D in addition to the reduction in the target mRNA of APOBEC3B.All other siRNA resulted in a significant reduction in mRNA for only the target gene.RNA editing events in cellular RNA from mock infected cells (Ctrl) or cells infected with EDB-2 or EDB-a-1 at MOI = 0.1 at 24 and 48 hpi were analyzed using the REDItools package. 56The frequency of each type of substitution is depicted as a percentage of the total number of mutations across all locations.The most frequent substitution types observed were A-G, G-A, and U-C respectively.Knockdown of all APOBECs, independent of predominant cellular expression localization, showed a significant increase in production of EDB-2.This suggests that the C-U RNA editing machinery is not playing a positive role in viral propagation, but rather a negative regulatory role.However, despite higher mutation rates, none of the APOBEC knockdowns led to an increased production of EDB-a-1.No increase in virus production could consistently be observed for any of the other VOCs for any of the APOBECs across both siRNAs tested.ADAR knockdown had no effect on any of the viruses.These data would indicate that, although APOBEC family members are likely driving the C-U RNA editing events in EDB-a-1 as evidence by the $40% C-U mutation detection in passages 2-4 of the virions (Figure 5C), there is no correlation between these high levels of C-U editing and virus titer (Figure 6A).Thus, APOBECs are unlikely therapeutic targets, although it is possible that manipulation of the C-U mutation rate could impact on evolutionary fitness.

DISCUSSION
Gene expression patterns of the top 100 expressed genes in Caki-1 cells change in a similar fashion for both EDB-2 and EDB-a-1 infection.However, at 24 hpi, upregulated genes in EDB-a-1 infections appear less strongly induced.A similar pattern is visible when focusing on ISGs.EDB-2 appears to induce an early and strong induction at 24 hpi, which appears to slightly decrease at 48 hpi.In contrast, EDB-a-1 does not induce as strong an upregulation of ISGs at 24 hpi, but similar overall levels are observed at 48 hpi.The underlying cause could be an adaptation of EDB-a-1 to the human host, improving mechanisms by which to avoid host recognition and stimulation of an interferon response. 41lternatively, early levels of EDB-a-1 production are lagging behind EDB-2 at 24 hpi and response induction may be delayed. 26However, EDB-a-1 48 hpi stimulation levels are still below ISG expression levels observed in EDB-2 at 24 hpi.
Among the upregulated ISGs are also nucleotide-modifying enzymes, including ADAR1 and members of the APOBEC3 family.Particularly APOBEC3B, C, D, F, and G are upregulated in their expression.Previous investigations show that APOBECs appear to be upregulated in a pathogen-specific pattern as shown for hCoV-NL63 and IAV in human airway epithelium cultures. 21Whereas hCoV-NL63 was shown to upregulate expression of APOBEC3A, C, D, and F, and IAV only upregulated G and H, it has been shown that C, F, and H possess antiviral activity.Here, we see a clear difference between EDB-2 and EDB-a-1 infection-mediated APOBEC3 upregulation with alpha showing a downregulation of F and G at 24 hpi before upregulation at 48 hpi, a pattern also observed for EDB-2 in APOBEC3D.In contrast, APOBEC3 F and G are strongly upregulated in EDB-2 at 24 hpi.
ADAR expression patterns in both EDB-2 and EDB-a-1 show an upregulation of this ISG following infection.While host RNA then also shows the highest mutation rates of A-G, this pattern appears to be infection-independent as similar frequencies are observed in mock infected samples.A-G variants are not observed at an increased frequency in vRNA.This is not further surprising since ADAR1 is primarily active in the nucleus 42 while SARS-CoV-2 replication occurs in the cytosol.Although the interferon-induced p150 isoform of ADAR1 can shuttle between the nucleus and the cytoplasm and so could potentially have a minor effect on the vRNA. 39None of the other nucleotide changes appear to increase in frequency in the host RNA following infection, indicating that upregulation of both ADAR and APOBECs has little impact on host RNA in general but may affect specific transcripts such as CDK13. 43imilar to previous observations by Kim et al., 25 we found a very high C-U mutation rate in SARS-CoV-2.However, this pattern was much more pronounced in EDB-a-1 (intracellularly and in released virus), whereas EDB-2 only showed increased C-U mutation in early passages of released virus.Due to the low mutation rate in later passages, conclusive preferences of hypermutation are difficult to make for EDB-2.These results, however, are in agreement with previously observed patterns of C-U hypermutation. 11Previous observation of broadly prevalent C-U hypermutation also corroborates observations made in this study showing a lack of SARS-CoV-2 strain specificity.
We found three coding mutations increasing in frequency over the course of passaging of EDB-a-1.None of them were linked to previously observed cell culture adaptations linked to the furin cleavage motif. 44The mutation in S at G72R has been previously reported to occur in the inactivated, adjuvanted Valneva vaccine as well as in field strains; 45 however, no further investigations on the function of this mutation are reported.Similarly, the S166L mutation, while observed in some isolates, for example of the B.1.609lineage, there is no information on the functionality of this mutation.However, the fast positive selection of these two mutations in vitro indicates an advantage over other variants in in vitro culture.
Following the differential ISG expression experiments, ideally all upregulated APOBEC3 variants would have been tested but due to availability of specific siRNAs we had to focus on 3B and D only alongside ADAR1.Both APOBEC3B, and D knockdowns yielded increased virus production in EDB-2 infected Caki-1 cells but no effect was observed in newer variants including alpha, delta, and omicron.The effect of the APOBEC3B knockdown was slightly unexpected since this nuclear APOBEC protein would not be expected to have a direct effect on SARS-CoV-2.However, it has been identified as an antiviral host factor, and as previously highlighted, deaminase function is not the only route APOBECs function through.They are also involved in the general interferon-stimulated antiviral response ( 46 , and reviewed in 47 ).
The decreased sensitivity of SARS-CoV-2 variants to APOBEC attenuation is in juxtaposition to the increased frequency of C-U mutations observed in EDB-a-1.Previous experiments have shown that overexpression of APOBECs did not appear to have a negative effect on SARS-CoV-2 production.In contrast, overexpression of APOBEC3 in selected overexpression cell lines increased replication of an unidentified Continued substitution observed in EDB-a-1 strain through passages was C-U.The sequences analyzed here were obtained from the cell supernatant.Significance between the EDB-2 and EDB-a-1 substitutions was calculated using a Student's t test achieving a p value of 0.000183 and a t-value of À5.52954.(D) Consequences on nucleotide variation on amino acid coding.Changes of nucleotides in a codon have been analyzed and categorized as silent (same aa or STOP coding), same amino acid class (non-polar, polar, +charge, -charge), different amino acid class, adding or removing a STOP codon, adding or removing an ATG/Start codon.SARS-CoV-2 isolate. 25We cannot align the results from this study with our knockdown findings but selection of overexpressing CaCo-2 cells, a cell line known for heterogeneity, may have led to selection of particularly susceptible cells.Previous findings in other hCoVs however show that overexpression of APOBEC3C, F, and H decreased virus production, particularly in hCoV-NL63. 21The hCoV-NL63 nucleocapsid was furthermore found to bind the APOBEC3s and through this interaction impacted the level of virus production.This is similar to previous findings where HIV VLPs containing SARS-CoV or hCoV-229E nucleocapsid (fragments) were found to bind APOBEC3G and package the protein into VLPs. 30verall, our results indicate that APOBECs play a role as antiviral ISGs in early variant infections of SARS-CoV-2.Surprisingly, the antiviral effect of APOBECs disappears with the emergence of VOCs and no effect of APOBEC knockdown is observed anymore for alpha, delta, and omicron variants.Future experiments should assess the mutation rate in new VOCs to assess whether higher adaptation to the human host continues to favor C-U mutation or whether eventually an equilibrium is reached.In hCoV-NL63 it was previously observed that APOBECs showed the ability to bind to the nucleocapsid protein. 21It should be further investigated whether this interaction leads to an integration of APOBEC into virions as observed for HIV. 48Binding efficiency to the SARS-CoV-2 nucleocapsid protein of APOBECs may change with new variants which may be linked to the reduction in antiviral effect of the APOBECs on new VOCs shown in the experiments presented here.
Interestingly, despite APOBEC3B and D showing an antiviral impact on EDB-2, this impact disappeared in VOCs, while C-U mutation rate increased in the alpha variant.This could be a beneficial strategy of CoVs to increase genetic variability of this generally very stable RNA virus. 492][53] It is likely that, similar to seasonal coronaviruses, this increased mutation rate eventually leads to an increase in the frequency of U in SARS-CoV-2.In turn, overall population diversity will probably decrease, after an initial increase following host adaptation, into a more settled state driven by the lack of targets for APOBEC editing and the coronavirus exoribonuclease activity. 11,47,51,54mitations of the study Like any cell-based system to study infections, monitoring the effects of cellular responses and the modification thereof on virus replication brings with it limitations.Here, as a model we have used the clear-cell renal carcinoma-derived, Caucasian, male, human cell line Caki-1.Previously, we could show this cell line to be highly susceptible to be infectable with several respiratory viruses. 55However, as a cell line model there is no complex interplay between different cell types, which may be observed in an air-liquid interface respiratory epithelial culture, nor is

RT-qPCR of intracellular RNA
Total RNA was extracted from knockdown cells in 12-well plates using the RNeasy Mini kit according to the manufacturer's instruction.RNA levels of APOBEC3B, APOBEC3D, and ADAR were analyzed by RT-qPCR using the GoTaq 1-Step RT-qPCR kit (Promega) according to the manufacturer's instructions using primers at a final concentration of 250 nM and 30 ng of RNA.Annealing temperature was set at 60 C and reactions were carried out on an Agilent Mx3005.This experiment was performed with 3 biological replicates and 3 technical replicates (N = 3*2).

QUANTIFICATION AND STATISTICAL ANALYSIS
Details on quantification of genome reads and dataset analysis may be found in the respective experimental sections above.
To assess the knockdown efficiency of siRNAs, Dunnett's multiple comparisons test was used to compare the mock transfected populations to each siRNA knockdown.Statistical analysis and graphs were produced using GraphPad prism v9.1.0on 3 replicates and 2 technical replicates (N = 3*2).

Figure 1 .
Figure 1.Experimental overview of the methodology performed during this study (A and B) Caki-1 cells were infected using SARS-CoV-2 variants EDB-2 (B.1), or EDB-a-1 (B.1.1.7)at an MOI = 0.1.(A) Cellular RNA was extracted at 24 and 48 hpi or (B) supernatant was harvested at 24 hpi and serially passaged on Caki-1 cells a maximum of four times (P1-4).For each passage of virus, vRNA was extracted from the supernatant.All RNA was amplified prior to Illumina Sequencing.Created using biorender.com.

Figure 2 .
Figure 2. Differential gene expression in SARS-CoV-2 infected cells (A) Top 100 differentially expressed genes in control cells compared to EDB-2 and EDB-a-1 SARS-CoV-2 infected cells The heatmap shows the top 100 differentially expressed genes in mock infected cells (Ctrl) and in cells infected with EDB-2 and EDB-a-1 variants at 24 and 48 hpi.The expression values are normalized and log2 scaled, presented as Z score values.(B) A heatmap of selected antiviral interferon-stimulated genes (ISGs) expression in control cells compared to EDB-2 and EDB-a-1 SARS-CoV-2 infected cells.The heatmap shows selected ISG expression levels in mock infected cells (Ctrl) and in cells infected with EDB-2 and EDB-a-1 variants at 24 and 48 hpi.The expression values are normalized and log2 scaled, presented as Z score values.

Figure 3 .
Figure 3. APOBECs and ADARs gene expression in control cells compared to EDB-2 and EDB-a-1 SARS-CoV-2 infected cells The heatmap presents the expression of APOBECs and ADARs in mock infected cells (Ctrl) and in cells infected with EDB-2 and EDB-a-1 variants at 24 and 48 hpi.Normalized and log2 scaled, presented as Z score values.

Figure 4 .
Figure 4. Frequency of host substitutions observed in control cells compared to EDB-2 and EDB-a-1 SARS-CoV-2 infected cellsRNA editing events in cellular RNA from mock infected cells (Ctrl) or cells infected with EDB-2 or EDB-a-1 at MOI = 0.1 at 24 and 48 hpi were analyzed using the REDItools package.56The frequency of each type of substitution is depicted as a percentage of the total number of mutations across all locations.The most frequent substitution types observed were A-G, G-A, and U-C respectively.

Figure 5 .
Figure 5.Nucleotide substitutions in SARS-CoV-2 (A) Visualization of identified substitutions location in the SARS-CoV-2 genome from intracellular and culture supernatant-purified RNA.Depiction of the locations of substitutions within the SARS-CoV-2 genome from vRNA amplified and sequenced for EDB-2 (passages 1-4) and EDB-a-1 (passages 2-4) variants during single round infection experiments at 24 and 48 hpi from cellular RNA, or from supernatant over a number of viral passages.Substitutions with the frequency R2% were taken under consideration and are presented on the diagram.Highlighted below are the non-silent mutations observed in EDB-a-1 over the course of cell passaging decreasing or increasing in frequency.Absolute numbers may be found in Table S1.(B and C) Frequency of substitutions observed in SARS-CoV-2 variants.Substitutions with frequency of R2% were counted.Numbers of substitutions fulfilling those criteria are highlighted above each column.(B) Frequency of substitutions observed in SARS-CoV-2 variants during the time course of infection.Bar colors correspond to the legend in (C).The most frequent substitution observed in EDB-a-1 strain was C-U.The sequences analyzed here were obtained from intracellular RNA.(C) Frequency of substitutions observed in SARS-CoV-2 virus genome as a function of passaging on Caki-1 cells.The most frequent

Figure 5 .
Figure 5. Continued substitution observed in EDB-a-1 strain through passages was C-U.The sequences analyzed here were obtained from the cell supernatant.Significance between the EDB-2 and EDB-a-1 substitutions was calculated using a Student's t test achieving a p value of 0.000183 and a t-value of À5.52954.(D) Consequences on nucleotide variation on amino acid coding.Changes of nucleotides in a codon have been analyzed and categorized as silent (same aa or STOP coding), same amino acid class (non-polar, polar, +charge, -charge), different amino acid class, adding or removing a STOP codon, adding or removing an ATG/Start codon.
mutation has the highest frequency of silent mutation as a consequence of a base change in a codon.No stop codons or start codons are removed and only in 3 of the possible 48 variations is a stop codon introduced.At the same time C-U mutation decreases the CpG frequency in a viral genome.RNA sequences enriched for CpG dinucleotides are recognized by the zinc finger antiviral protein (ZAP)

Figure 6 .
Figure 6.Impact of knockdown of selected APOBECs and ADAR on virus replication of VOCs on Caki-1 cells Selected APOBECs and ADAR were knocked down by transfection of Caki-1 cells with RNAi.(A) 72 h post transfection cells were inoculated with MOI = 0.5 of SARS-CoV-2 original, alpha, delta, or omicron VOCs (strains highlighted in the legend).Produced virus was measured at 48 hpi by measuring vRNA in the supernatant through direct lysis RT-qPCR as previously described. 55(B) 72 h post transfection RNA was extracted from cells and knock down efficiency was assessed by RT-qPCR.mRNA fold-change was calculated relative to a mock transfected control.Error bars represent +/À the SEM, dashed line indicates no change in replication (100%).Significance was calculated by two-way ANOVA relative to a mock transfected control whereby * denotes p % 0.05, ** denotes p % 0.01, and **** denotes p % 0.0001.N = 3*3.
or RESOURCE SOURCE IDENTIFIER