Co-circulation of SARS-CoV-2 Alpha and Gamma variants in Italy, February and March 2021

Background Several SARS-CoV-2 variants of concern (VOC) have emerged through 2020 and 2021. There is need for tools to estimate the relative transmissibility of emerging variants of SARS-CoV-2 with respect to circulating strains. Aim We aimed to assess the prevalence of co-circulating VOC in Italy and estimate their relative transmissibility. Methods We conducted two genomic surveillance surveys on 18 February and 18 March 2021 across the whole Italian territory covering 3,243 clinical samples and developed a mathematical model that describes the dynamics of co-circulating strains. Results The Alpha variant was already dominant on 18 February in a majority of regions/autonomous provinces (national prevalence: 54%) and almost completely replaced historical lineages by 18 March (dominant across Italy, national prevalence: 86%). We found a substantial proportion of the Gamma variant on 18 February, almost exclusively in central Italy (prevalence: 19%), which remained similar on 18 March. Nationally, the mean relative transmissibility of Alpha ranged at 1.55–1.57 times the level of historical lineages (95% CrI: 1.45–1.66). The relative transmissibility of Gamma varied according to the assumed degree of cross-protection from infection with other lineages and ranged from 1.12 (95% CrI: 1.03–1.23) with complete immune evasion to 1.39 (95% CrI: 1.26–1.56) for complete cross-protection. Conclusion We assessed the relative advantage of competing viral strains, using a mathematical model assuming different degrees of cross-protection. We found substantial co-circulation of Alpha and Gamma in Italy. Gamma was not able to outcompete Alpha, probably because of its lower transmissibility.


Three-strain SARS-CoV-2 model
We adopted a three-strain Susceptible -Infectious -Recovered (SIR) mathematical model to simulate co-circulation of historical lineages of SARS-CoV-2 ("wildtype") and variants of concern Alpha and Gamma. is the degree of cross-protection granted from previous infection with historical or Alpha lineages against re-infection with Gamma ( =1 is complete cross-protection); we assumed that recovering from any infection provides full immunity against re-infection with historical or Alpha lineages for at least the duration of our simulations (3 months).
• is the recovery rate from infection, set equal to the inverse of the average generation time previously estimated for Italy at 6.6 days [1].
The initial fraction of immune individuals was computed as Where T is one of three time periods (1: February 20 -June 30, 2020, 2: July 1 -September 30, 2020; 3: October 1, 2020 -January 15, 2021), CT is the cumulative number of cases notified in period T, + is the infection notification rate in the period (estimated at % =9.4%, The reproduction number of historical lineages on January 15, 2021 was computed as: The number of infectious individuals * from any lineage was initialized by sampling from a Poisson distribution with rate equal to the mean number of daily hospital admissions between January 10 and January 17, multiplied by the average generation time and dividing by the average proportion h of hospitalized infections who require hospitalization, estimated by applying the method reported in [2]. Previous estimates on this parameter reported an approximately 13% hospitalization probability for symptomatic cases admitted between March and September 2020 [2]. When re-applying the same methodology to cases from October 2020 to January 15, we obtained a 7.7% hospitalization probability among symptomatic individuals. Taking into account the overall proportion of infections who become symptomatic [3], we estimate that 2.4% of all infections required hospitalization between October 2020 and January 2021. However, we run sensitivity analyses with alternative values of h of 1.2% and 3.6% (see below). The initial number of infected individuals was distributed in the three compartments $% , $& and "# , according to the initial prevalence of the three lineages, % , & , and (1 − % − & ), respectively.
The reproduction number of historical lineages on January 15, 2021, "# , the relative transmissibility of Alpha and Gamma, % and & , the initial prevalence of Alpha and Gamma, % and & , and the initial number of infectious individuals of any lineage * were free model parameters. The degree of cross-protection was fixed to values between 0 and 1 with a step of 0.1, and for each value of the free model parameters were recalibrated. The modelestimated daily hospital admissions from any lineage were computed as the daily new infections, multiplied by the hospital admission probability and shifted forward by 3 days to allow for delays between the start of infectiousness and the hospital admission. The modelestimated prevalence of lineage x at time t, used for the multinomial likelihood (see main text), was computed as # ' = where q={wt, v1, v2}.
For calibration via MCMC, we used uninformative priors with 50,000 iterations (of which the first 10,000 were discarded to consider only iterations after convergence) and a Manhattan algorithm with recursive normal jumps for accepting or rejecting candidate parameter sets. Because the definition of the likelihood includes only relative prevalence data (see main text), we constrained the MCMC to explored parameter values for which the model-estimated absolute incidence was close to realistic values; to this aim, we assigned L=0 to simulations for which the mean square error between the modeled and observed daily hospital admissions (a reliable proxy of the total incidence) is below 1.5 times the variance of the observations. Convergence was assessed via visual inspection of the MCMC traceplots. The overall code, including the MCMC algorithm was programmed in R version 4.0.3 (2020-10-10). Figure S1 shows posterior means and 95% credible intervals (CrI) for the estimated initial prevalence of Alpha and Gamma, % and & , in Italy and in the three macro-areas where, according to the surveys, Alpha and Gamma are co-circulating. These estimates are insensitive to different values of .  Figure S2 shows the posterior means and 95% credible intervals for estimated reproduction number of historical lineages on January 15, "# . These estimates are insensitive to different values of as well.

Center
Degree of cross protection Initial fraction of infected individuals (%)   Figure S3 shows the posterior means and 95% credible intervals for the initial number of infectious individuals of any lineage * . These estimates are insensitive to different values of α as well.

Italy
Degree of cross protection

North−East
Degree of cross protection Restrictions in the different tiers since January 14, 2021 A restriction system based on tiers was introduced in Italy on a regional basis since November 6, 2021.

Center
Degree of cross protection Initial number of infectious inidividuals (x1,000)

North−East
Degree of cross protection Initial number of infectious inidividuals (x1,000)

Transmissibility of Alpha in regions without Gamma circulation
We conducted a sensitivity analysis to estimate the transmissibility of Alpha in Lombardy and Veneto, in which Gamma was not circulating and for which an additional data point from an identical survey conducted on cases diagnosed on February 3 and 4 was available (Table S4). The estimate was conducted using a two-strain model that results from a special case of the three-strain model when parameter & is fixed to zero. As shown in Figure S4, the model was able to fit the epidemiological trends on hospital admissions and the prevalence of Alpha estimated in the three surveys.

Increased duration of infection
We run a sensitivity analysis for Lombardy and Veneto regions where instead of considering the generation time equal for historical and Alpha lineages, we leave it as a free model parameter, while we assumed that the transmissibility of the two lineages were the same. As shown in Figure S5, the model was able to fit the epidemiological trends on hospital admissions and the prevalence of Alpha estimated in the three surveys.

Co-circulation of Alpha and Gamma in individual regions of the Central macro-area
We conducted a sensitivity analysis in two individual regions of the Central macro-area with high co-circulation (Tuscany and Lazio) to investigate the potential impact of geographical aggregation on estimates of the relative transmissibility of Alpha and Gamma. As shown in Figures S6 and S7, the model was able to fit the epidemiological trends on hospital admissions and the prevalence of Alpha and Gamma estimated in the two surveys in both regions. As shown in Figure S8, independently on the assumed degree of cross-protections, we found a robust mean estimate for the relative transmissibility of Alpha for Lazio, ranging between 1.80 and 1.85, with confidence intervals ranging between 1.68 and 2.06, and for Tuscany, ranging between 1.54 and 1.69, with confidence intervals ranging between 1.23 and 2.17, compatible with estimates from the main analysis.   Degree of cross-protection as a free parameter We run a sensitivity analysis for Italy where instead of assuming the degree of crossprotection between historical/Alpha lineages and Gamma, we estimate it as a free model parameter. As shown in Figure S9, the estimated degree of cross-protection may take values between 0 and 1 and it has a strong positive correlation with the estimated relative transmissibility of Gamma (correlation = 0.78, p-value < 0.001). The estimated relative transmissibility of Alpha and Gamma at national level are respectively 1.56 (95%CI: 1.48-1.65) and 1.24 (95%CI: 1.01-1.44), consistently with the main analysis.

Alternative assumptions on hospitalization probability
We run sensitivity analyses for Italy where we made different assumptions on the hospitalization probability h. In particular, we assumed i) a 50% lower value (h=1.2%), applied to all lineages; ii) a 50% higher value (h=3.6%), and iii) a 50% increased value in the hospitalization probabilities of Alpha and Gamma lineages only, to account for a possible increased severity [4,5]. In all three analyses, the model was able to fit the epidemiological trends on hospital admissions and the prevalence of Alpha and Gamma estimated in the two surveys in Italy.
As shown in Figure S10, under the three different assumptions on the hospitalization probabilities, we found robust mean estimates for the relative transmissibility of Alpha and Gamma, compatible with estimates from the main analysis. Figure S10. Estimates of the relative transmissibility of Alpha and Gamma. Estimates are provided for different assumed values on the degree of cross protection (0: no cross-protection; 1: complete cross-protection) conferred by previous infection with the wildtype or Alpha against reinfection with Gamma, under the assumption of a 50% lower hospitalization probability (left), a 50% higher hospitalization probability (middle), and a 50% increase in the hospitalization probability for Alpha and Gamma lineages only (right). Points indicate the mean value of the estimated relative transmissibility of Alpha (red) and Gamma (blue) lineages; lines indicate 95%CI.  Table S7. Abbreviations of regions for Figure 2 in the main text.