Impact of variants of concern on SARS-CoV-2 viral dynamics in non-human primates

The impact of variants of concern (VoC) on SARS-CoV-2 viral dynamics remains poorly understood and essentially relies on observational studies subject to various sorts of biases. In contrast, experimental models of infection constitute a powerful model to perform controlled comparisons of the viral dynamics observed with VoC and better quantify how VoC escape from the immune response. Here we used molecular and infectious viral load of 78 cynomolgus macaques to characterize in detail the effects of VoC on viral dynamics. We first developed a mathematical model that recapitulate the observed dynamics, and we found that the best model describing the data assumed a rapid antigen-dependent stimulation of the immune response leading to a rapid reduction of viral infectivity. When compared with the historical variant, all VoC except beta were associated with an escape from this immune response, and this effect was particularly sensitive for delta and omicron variant (p<10−6 for both). Interestingly, delta variant was associated with a 1.8-fold increased viral production rate (p = 0.046), while conversely omicron variant was associated with a 14-fold reduction in viral production rate (p<10−6). During a natural infection, our models predict that delta variant is associated with a higher peak viral RNA than omicron variant (7.6 log10 copies/mL 95% CI 6.8–8 for delta; 5.6 log10 copies/mL 95% CI 4.8–6.3 for omicron) while having similar peak infectious titers (3.7 log10 PFU/mL 95% CI 2.4–4.6 for delta; 2.8 log10 PFU/mL 95% CI 1.9–3.8 for omicron). These results provide a detailed picture of the effects of VoC on total and infectious viral load and may help understand some differences observed in the patterns of viral transmission of these viruses.


37
The sever acute respiratory coronavirus 2 (SARS-CoV-2) is the causative agent of the Coronavirus-

51
In that context where human clinical data are difficult to interpret, the non-human primate (NHP) 52 experimental model offers a unique opportunity to describe infection with SARS-CoV-2 in detail in a 53 fully controlled environment. Since 2020, our group has conducted many studies to evaluate the effects 54 of antiviral drugs or vaccines in this model (14,15) , and showed its large predictive value (16). Here, 55 we analysed retrospectively viral load data obtained in 78 animals that were included as control arms of 56 these studies and that were infected with different strains of SARS-CoV-2 (historical, beta, gamma, 57 delta and omicron (BA.1)). In addition, we performed longitudinal measures of viral culture to evaluate 58 a potential effect of VoC on viral infectivity. Using the techniques of mathematical modelling, we 59 characterize the viral kinetics in these animals and we discuss their biological insights. 60 61

62
Variant of concern viral kinetics 63 Several biomarkers were measured, both genomic RNA and subgenomic RNA were quantified at regular 64 interval over all the study period and infectious titers at 2 times points. All macaques developed a rapid 65 infection with genomic viral load peaking between 2-and 3-day post-infection (dpi) for the historical 66 and beta variant, 3.5 dpi for variant delta and 4 dpi for variants gamma and omicron (BA.1). Genomic 67 viral load was cleared at 8 dpi for the historical variant, 10 dpi for the beta variant, at 12 dpi for variants 68 delta and omicron (BA.1) and at 14 dpi for variant gamma (Fig 1 and S1  To account for the quick drop in infectious titers observed in the historical variant, (Fig 1 and S1 Fig)   82 several models incorporating an action of an antigen-mediated immune response were tested (Fig 2).

83
All models, except a model targeting the viral production parameter, provided an improvement of BIC 84 compared to a target cell limited model ( VoC specific effect on viral dynamic parameters 104 Once an effect of the immune response was selected, a covariate search algorithm was used to find the 105 most likely VoC associated effects (see methods) and considered the historical variant as the reference.

106
Several variant-specific covariates were found on viral kinetics parameters that we detail below (Fig 3   107 and S2 Table). First, beta variant was characterized with a reduced infected cells death rate ( ) by a 108 factor of 0.7 (95% CI 0.6 -0.9) compared with the historical variant (p-value < 0.01). This led to an (from 5 to 30) and performed the covariate search on all models. We found that a delay of 3 days yielded 123 the best results (S3 Table)   i.e., 10,000-100,000 times less virus than in the animal model. Using simulations with lower inoculum, considering both uncertainty in the estimation and inter-individual variability (see methods), we are able 139 to derive metrics of interest for each variant.

144
The reduced infected cell clearance rate of the beta variant resulted in a longer period of viral load 145 shedding. The duration of the acute infection stage was consequently increased from 10.9 days (95% CI 146 9.5 -13.1) for the historical variant to 13.4 days (95% CI 11.1 -15.7) for the beta variant.

147
All variants except beta have shown an effect on the antigen-mediated response, greatly reducing its 148 impact on viral kinetics. As the effect of the antigen-mediated response was reduced, the infectious ratio 149 was increased leading to more infectious particles produced over longer periods of time. This led to the 150 increase of the infectious titers clearance stage duration from 1.5 days for the historical variant (95% CI 151 0.6 -1.9) to 6 days (95% CI 4.4 -7.5), 3.8 days (95% CI 3.1 -4.6) and 3.7 days (95% CI 2.8 -4.5) for 152 the gamma, delta and omicron variants respectively (Fig 5). This is in line with numbers of studies   153 showing the immune escape capabilities of those variants (20-22).

154
An effect increasing the viral production parameter ( ), as observed for the delta variant, results in 155 largely higher peak viral load of 7.6 log 10 copies/mL (95% CI 6.8 -8.2) and peak infectious titers of 3.7 156 PFU/mL (95% CI 2.4 -4.6). Conversely, an effect reducing the viral production parameter, as observed 157 for the omicron variant, results in lower peak viral load compared to the historical variant of 5.6 log 10 158 copies/mL (4.8 -6.3) but very similar peak infectious titers at 2.8 PFU/mL (95% CI 1.9 -3.8). This is 159 due to an effect of omicron on the infectious ratio, increasing the proportion of infectious virus produced.

163
Discussion Here, we used mechanistic models to characterize in detail the viral dynamics of the main variants of 165 concern in an experimental model of non-human primates. We evaluated the impact of an antigen-166 mediated immune response on the viral dynamics and found that an effect reducing the infectious ratio 167 best described our data. Some of the variants of concern, gamma, delta and omicron (BA.1) showed a 168 strong ability to escape this response greatly increasing the number of infectious viruses produced over 169 the course of the infection compared to the historical variant. Interestingly, the delta variant was 170 associated with an increased viral production rate, whereas the omicron variant was associated with a 171 lower viral production rate but a higher infectious ratio.

172
Using simulations in a natural infection scenario, we found that omicron infections, relative to delta 173 infections, are associated with lower peak viral RNA and reduced duration of viral RNA clearance while 174 having similar peak infectious titers and duration of infectious titers clearance.

175
These results suggest that omicron's infectiousness cannot be attributed to an increased viral RNA 176 production but maybe due to an immune escape coupled with an increased infectious ratio, greatly 177 increasing the number of infectious particles produced.

178
Although many other factors are at play to explain the increased transmissibility of certain variants of 179 concern, differences in viral dynamics can provides insights into the biology of those variants. As such,

234
The results mainly focus on the genomic viral load as the subgenomic is a directly proportional to the 235 latter.
236 Basic viral dynamic model 237 We

284
In the following only the compartment 20 will serve as the effector for the action of the immune system.

285
The transfer rate parameter is then written as and fixed to 6.67 d

Selection of variant-specific effect on the viral dynamic parameters
328 Using the best model selected at the previous step, we sought to identify VoC-specific effect on the 329 parameters of the model ( , , , and ). We first performed a backward selection of the random effects 330 removing non-significant ones (i.e. relative standard error > 50%) if the BIC wasn't degraded by more 331 than 2 points. We then used the Conditional Sampling use for Stepwise Approach on Correlation tests 332 (COSSAC) to identify variant specific effect (31). Then a backward procedure was used to remove any 333 non-significant covariate effect with a Wald test (i.e. the covariate was removed if its coefficient effect 334 relative standard error was > 50%). This procedure was repeated until all nonsignificant covariate effects 335 had been eliminated. Additionally, we performed a sensitivity analysis on our best structural model. We