key: cord-0856306-23c0c8oj
authors: Elie, Baptiste; Roquebert, Bénédicte; Sofonea, Mircea T.; Trombert‐Paolantoni, Sabine; Foulongne, Vincent; Guedj, Jérémie; Haim‐Boukobza, Stpéhanie; Alizon, Samuel
title: Variant‐specific SARS‐CoV‐2 within‐host kinetics
date: 2022-05-05
journal: J Med Virol
DOI: 10.1002/jmv.27757
sha: 931f66fbe6ab9649c2ad8791576e48d2c1f77c01
doc_id: 856306
cord_uid: 23c0c8oj

Since early 2021, SARS‐CoV‐2 variants of concern (VOCs) have been causing epidemic rebounds in many countries. Their properties are well characterized at the epidemiological level but the potential underlying within‐host determinants remain poorly understood. We analyze a longitudinal cohort of 6944 individuals with 14 304 cycle threshold (Ct) values of reverse‐transcription quantitative polymerase chain reaction (RT‐qPCR) VOC screening tests performed in the general population and hospitals in France between February 6 and August 21, 2021. To convert Ct values into numbers of virus copies, we performed an additional analysis using droplet digital PCR (ddPCR). We find that the number of viral genome copies reaches a higher peak value and has a slower decay rate in infections caused by Alpha variant compared to that caused by historical lineages. Following the evidence that viral genome copies in upper respiratory tract swabs are informative on contagiousness, we show that the kinetics of the Alpha variant translate into significantly higher transmission potentials, especially in older populations. Finally, comparing infections caused by the Alpha and Delta variants, we find no significant difference in the peak viral copy number. These results highlight that some of the differences between variants may be detected in virus load variations.

| 1 longitudinal data from hospitalized patients, 17 health workers, 18 and experimental infections in nonhuman primates. 19 More recently, studies focusing on the effect of the vaccine showed an overall decrease in viral loads among vaccinated infected patients, 20 or a faster viral load decrease, 21, 22 although this remains unclear for symptomatic infections. 23 These studies illustrate that viral kinetics reflect some properties of the infection and can explain some of the variations in the detection probability.

Here, we analyze SARS-CoV-2 kinetics in a large number of individuals using linear mixed models, which allows us to explore how virus load dynamics may vary depending on the setting (general population or hospitals) or the variant causing the infection.

This study was approved by the CHU of Montpellier's Institutional Review Board and is registered at ClinicalTrials.gov (no NCT04653844). The data originates from variant-specific RT-qPCR tests performed in France between February 06 and August 21, 2021 on SARS-CoV-2 positive samples, 13, 15 from four different kits, following the manufacturer's instructions. In particular, we used synthetic DNA as an internal control.

The multiplex assays use probes targeting variant-specific mutations (see Table 1 ), as well as a region in the N virus gene for control purposes. We use the Ct of the later probe in our analyses. Therefore, we consider the infecting virus as "historical lineage" if it does not possess any of the key mutation associated with VOCs. The variant-specific qPCRs results were validated internally using nextgeneration sequencing.

Given the specificity of each assay used, different formatting steps were used. In particular, the initial IDS1 test designed to distinguish between the historical lineages and the Alpha VOC may increase the proportion of the latter VOC for high Ct values. Further details are shown in the flow diagram in Figure 1 and can be found in earlier studies. 8, 13, 15 2.2 | Digital droplet PCR Using Ct measures as a proxy for virus load has several limitations, especially in the case of coronaviruses. 11 Here, we do not attempt to equate the two but rather assume that temporal variations in Ct values are associated in changes in infectiousness.

To further investigate the biological significance of Ct values, we analyzed samples from infections by a known virus lineage, that is, historical or VOC, using both a variant-specific PCR and a digital droplet PCR (ddPCR). More precisely, we used the SARS-CoV-2 droplet digital PCR (ddPCR) Kit (BioRad), which has two couples of primers targeting regions N1 and N2 in the virus, and the cellular human RNAse P for internal control.

We used a linear model to study the potential link between the Ct value of the screening test and the number of virus copies obtained using ddPCR, using virus lineage and test assay as a cofactor. The results of the statistical model were used to convert Ct values from all assays into a number of virus copies.

We analyzed the longitudinal data of number of virus copies (or Ct values for IDS1) with linear mixed models and used the R package lme4 24 to fit the restricted maximum likelihood parameters to the data. The response variable was the number of virus copies (or the Ct value for IDS1), and we included two random effects on the individual and on the region of sampling.

To select which additional effects to include in the linear mixed model, we compared models with all possible parameters combinations listed in Table 2 . The best model was chosen based on the Akaike information criterion (AIC). 26 We verified that models with ΔAIC < 2 were qualitatively identical to the model with the best AIC. 26 We also verified that the censored data points (i.e., the Ct values above the limit of detection, which was set at 37, or the viral copies number below 10 000 copies/ml) did not have a significant impact on the results by computing all the linear mixed models selected by a ΔAIC < 2 with a censored effect using the lmec R package.

Differences in viral copies between populations from the linear model outputs were statistically assessed using the Tukey adjustment and the emmeans R package.

We first calculated the global epidemic reproduction number in each French region using hospital admission data. 27 We then adjusted this number by using the estimated relative proportion of each variant F I G U R E 1 Data set formatting steps. n indicates the number of tests, that is, Ct values, analyzed. For the IDS1 assay, the analyses were performed directly on the Ct values because the provider did not have any remaining tests to use with the ddPCR calibration. Furthermore, the assessment of the virus lineage (i.e., wild type or variants of concern) was only performed for tests with Ct values lower or equal than 30. For results obtained using the IDS2 and Perkin assays, Ct values were converted into viral genome copies before subsequent analysis and the assessment of the virus lineage was only performed for tests with more than 5.4log 10 Interaction between day and each effect This represents the impact of each effect on the viral copy number decay after peak

Interaction between the age and virus strain This represents the differential impact of the variant on the viral copy number peak in function of the age among the daily infections caused, 5, 8, 15 and the mean transmission advantage computed by an independent study. 28 

We used the infectiousness profile estimated from the data of He et al. 29 This analysis being restricted to one test (IDS1), we studied the correlation between the Ct value and the instantaneous infectiousness, that is, the infectiousness profile density.

We then used this mapping to infer a transmission potential, which can be seen as a proxy for the basic reproduction number. 31 Using the outputs of our model capturing within-patient dynamics, we integrated the infectiousness obtained from the Ct values from Day 0 to Day 15.

For a given RT-qPCR assay, we found a log-linear relationship between the Ct value and the absolute number of viral copies measured by ddPCR ( Figure 2 ). Compared to the reference (i.e. IDS2), IDS3 yielded a Ct value 1.96 lower (Student t-test, Tukey adjustment:

p < 10 −4 ), and Perkin a Ct value 6.9 higher (p < 10 −4 ). For the Perkin assay, we also detected a significant effect of the variant on the Ct 

After formatting the database for the IDS1 assay (Figure 1 ), we identified 12,536 suitable samples ( Figure S1 ).

To further investigate the implications of these results at the population level, we performed a mapping between the daily infectivity and the daily Ct value after its peak for the wild-type We used this mapping between Ct values and infectivity to study transmission potential differences between lineages. We found that the advantage of the Alpha VOC over the historical strain was more pronounced in countries with older populations ( Figure 4B ).

We then analyzed the data which involved Ct values from three other assays that we were able to convert into virus copy numbers (see Section 2 and Figure 2 ). In the following, we use a log base 2 relationship with time since peak of viral genome copies to compare the orders of magnitude with the previous results.

Infections consistent with viruses from Beta/Gamma/Eta lineages were too rare to be analyzed and removed them from the data set. Overall, we were able to compare infections caused by the Alpha or the Delta variant by analyzing the remaining 2515 samples (Table 3 ). Compared to the previous data set, patients were slightly younger (median age 35 vs. 41) and the follow-up duration was shorter (median 6 vs. 8 days). This latter difference was corrected by the variable taking into account the delay between first and second test.

The best model according to the AIC criterion is detailed in Table 5 and the dynamics of the number of viral genome copies is shown in Figure 3B . On 15 out of 17 models with a ΔAIC < 2, we found no significant difference between the kinetics of the infections caused by the Alpha and the Delta VOC on the peak viral copy number. However, we found a significant difference on the decay rate on 13 models.

To make sure that this result was not due to confounding factors, we included several covariates in our analysis which revealed a strong effect of the sampling date, which could interfere with the variant effect (since the Delta VOC rapidly replaced the Alpha VOC). The vaccine coverage in the population, which was used a proxy for the probability that the individual was vaccinated, as well as the delay between first and second test, measuring behavioral differences, were retained as a cofactor in the best linear model. Furthermore, the variant-specific reproduction number R(t), estimated at the regional level in France was also retained as a significant covariate, which is consistent with previous observations, 13, 14 although the effect was not statistically significant.

Finally, we also ran a survival analysis as a second approach, and observed a similar trend, with individuals infected by the Alpha variant having a median infectious period 1.75 days longer than the ones infected by the Delta variant ( Figure S2 ).

Understanding the within-host kinetics of SARS-CoV-2 infections yielded original insights on infection virulence, 17 or efficiency of screening strategies. 18 To further investigate the consequences of these variations in within-host kinetics at the epidemiological level, we assumed a linear correlation between Ct value and daily infectiousness, which is consistent with an earlier study 33 and confirmed by a modeling approach. 34 Translating the estimated Ct kinetics into transmission potential profiles revealed that the high viral copy number observed in Alpha VOC infections were consistent with a 25% increase in transmission potential compared to historical strains.

Kinetics of samples collected in hospitals exhibited higher peak viral genome copies (i.e., lower Ct values), which is consistent with earlier studies. 17 The results were unaffected by the removal of hospital data from the analysis.

We were unable to convert the Ct values from the first assays into number of virus genomes, which precluded us from comparing the Delta variant to the ancestral lineages. Therefore, we limited our comparison of the Delta variant to the infections caused by the Alpha variant. Our models show no difference in the peak of viral copy number, but they detect a faster decrease in infections caused by the Delta variant than in that caused by the Alpha variant. This may appear as counter-intuitive given the latter has a clear transmission advantage over the former. 5, 28 This could be explained by the fact that ddPCR itself is limited because it does not count the number of infectious virions. 11 Indeed, experiments suggest that for a given Ct value, samples from Delta VOC infections have a higher number of infectious particle than that from Alpha VOC infections. 35 Finally, this result is also consistent with others that find little differences in Ct values between infections caused by the Alpha and the Delta VOC. 22 A limitation of this analysis is that we do not have any indication regarding the date of the infection or of the symptom onset. This uncertainty prevented us from analyzing more mechanistic models with nonlinear mixed-effect models. 17 However, since the nature of the virus causing the infection is unlikely to affect the number of days between infection and screening, we do not expect our assumption that the lowest Ct value corresponds to the peak viral genome copies to introduce biases.

Our comparison between the Alpha and Delta VOCs is potentially subject to additional biases. This part of the analysis covers a large period of sampling, with an important increase in the vaccine coverage, and potential behavioral changes. We attempted to correct for those biases by including covariates such as vaccine coverage in the general population, variant reproduction number, or the delay between the first two tests of an infected individual.

The use of age-structured vaccine coverage is a proxy that makes the simplifying assumption that the vaccination does not impact the probability of being infected nor being tested, and therefore the vaccine coverage in the general population would be the same as the one in the samples we studied. However, even if this is a strong assumption, this seems to capture at least part of the effect, and it was found to be significantly impacting the peak viral genome copies, with a lower peak with increasing vaccine coverage.

Moreover, the use of antigenic tests as a first line of screening is likely to have delayed the delay between infection and RT-qPCR testing. Their use increased from 20% to 50% of the total Covid-19

screening tests between May and August in France. 36 We observe that the median delay between the first and the second test (which is usually done 7 days after the symptoms) decreased from 7 to 5 days during the analysis. We included this variable in the model, and a shorter delay was significantly associated with a smaller peak viral genome copy number. Indeed, we interpret a shorter delay as a first test done later on the infection.

Finally, our statistical analysis does not take into account the fact that Ct values beyond 40 are censored, that is, all we know in that case is that the individual cleared the viral genome before the sample date. The statistical tools for linear mixed models with censored effects are less efficient, but we compared the best models obtained without taking into account the censoring effect to the same models with a censoring effect, and most of the models (12/12 for historical strain vs. alpha, 15/17 for alpha vs. delta) yielded similar results both in term of viral copy number peak difference and decay rate.

Furthermore, we ran a survival analysis using a Cox model, which also 

Increased mortality in community-tested cases of SARS-CoV-2 lineage B.1.1.7

Assessing transmissibility of SARS-CoV-2 lineage B.1.1.7 in England

Detection of a SARS-CoV-2 variant of concern in South Africa

Genomics and epidemiology of the P.1 SARS-CoV-2 lineage in Manaus

Rapid spread of the SARS-CoV-2 Delta variant in some French regions

Risk of mortality in patients infected with SARS-CoV-2 variant of concern 202012/1: matched cohort study

Evaluation of the relative virulence of novel SARS-CoV-2 variants: a retrospective cohort study in Ontario

Detection of rapid SARS-CoV-2 variant spread

Epidemiological and clinical insights from SARS-CoV-2 RT-PCR crossing threshold values

Increased infections, but not viral burden, with a new SARS-CoV-2 variant

SARS-CoV-2 viral RNA levels are not'viral load

Ct threshold values, a proxy for viral load in community SARS-CoV-2 cases, demonstrate wide variation across populations and over time

Superspreading genomes

Estimating epidemiologic dynamics from cross-sectional viral load distributions

SARS-CoV-2 variants of concern are associated with lower RT-PCR amplification cycles between

Viral kinetic modeling: state of the art

Modeling SARS-CoV-2 viral kinetics and association with mortality in hospitalized patients from the French COVID cohort

Estimating the effectiveness of routine asymptomatic PCR testing at different frequencies for the detection of SARS-CoV-2 infections

SARS-CoV-2 viral dynamics in non-human primates

Longitudinal analysis of SARS-CoV-2 vaccine breakthrough infections reveal limited infectious virus shedding and restricted tissue distribution

Community transmission and viral load kinetics of the SARS-CoV2 delta (B.1.617.2) variant in vaccinated and unvaccinated individuals in the UK: a prospective, longitudinal, cohort study

Viral dynamics of SARS-CoV-2 variants in vaccinated and unvaccinated persons

Characterisation of vaccine breakthrough infections of SARS-CoV-2 delta and alpha variants and within-host viral load dynamics in the community

Generalized linear mixed models: a practical guide for ecology and evolution

Infectious diseases of humans: Dynamics and controls>

Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach

Rt2: computing and visualising COVID-19 epidemics temporal reproduction number

Increased transmissibility and global spread of SARS-CoV-2 variants of concern as at

Temporal dynamics in viral shedding and transmissibility of COVID-19

COVID-19 infectivity profile correction

Variation in HIV-1 set-point viral load: epidemiological analysis and an evolutionary hypothesis

SARS-CoV-2 viral dynamics in infections with variants of concern in the French community

Transmission of COVID-19 in 282 clusters in Catalonia, Spain: a cohort study

Quantifying the relationship between SARS-CoV-2 viral load and infectiousness

Quantitative measurement of infectious virus in SARS-CoV-2 Alpha, Delta and Epsilon variants reveals higher infectivity (viral titer:RNA ratio) in clinical samples containing the Delta and Epsilon variants. medRxiv. 2021 36. (2021) COVID: Exécutions et Ventes de Tests Antigéniques Rapides en Pharmacies. Data.gouv

The authors acknowledge the ISO 9001 certified IRD i-Trop HPC (South Green Platform) at IRD montpellier for providing HPC resources that have contributed to the research results reported within this study. We thank all the ETE modeling team for thoughtful discussions.

The authors declare no conflicts of interest.

The data that support the findings of this study are available from the corresponding author upon reasonable request. Data