In December 2019, a novel coronavirus named SARS-CoV-2 emerged in Wuhan, China, in the province of Hubei [1, 2, 3]. SARS-CoV-2 spread rapidly around the world and the number of cases and deaths has increased rapidly (https://ourworldindata.org/coronavirus-data). Since then, finding effective treatments and vaccines has remained a major global challenge. Several drugs with an antiviral effect have been tested in vitro and in vivo, with drug repurposing being one of the strategies applied. Many drugs have shown an inhibition in vitro such as certain antimalarial drugs (chloroquine, hydroxychloroquine), ivermectin, macrolides (azithromycin, spiramycin), several protease inhibitors and some RdRp (RNA-dependent RNA polymerase) inhibitors such as remdesivir and sofosbuvir [4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18]. The combination of azithromycin and hydroxychloroquine showed synergistic effects in vitro [19] at concentrations of 5 $\mu$M for each drug and the combination was widely used in our institute to treat infected patients, as well as in several countries, particularly in Africa (Zambia, Uganda, Egypt, Algeria, Morocco, Tunisia, Senegal, Cameroon) [20, 21]. However, the use and effectiveness of this treatment remains highly controversial [22]. In the literature, and in our experience, some patients, especially immunocompromised patients or those with other comorbidities (such as hypertension and diabetes) have presented a positive viral load with a late clearance [23, 24, 25, 26]. In our institute, we were able to isolate at least one SARS-CoV-2 strain from several patients who had been defined as having a “persistent” infection, with a positive RT-PCR test more than 10 days after admission, despite the fact that they had received a combination of azithromycin and hydroxychloroquine [27]. In those cases, the question of susceptibility to the antiviral drugs of the responsible strain was raised. Indeed, most in vitro studies evaluating susceptibility to antiviral drugs have used a unique strain or clone of SARS-CoV-2, and have considered that this clone is representative of all strains, despite the fact that the variability of antiviral activities on an enlarged panel of strains is unknown. In this study, we decided to use an automated model of Vero E6 to screen a single combination of azithromycin and hydroxychloroquine on multiple strains isolated from patients with persistent and non-persistent infections and who were randomly chosen to detect reduced susceptibility to this combination [28]. After this preliminary screening, a dose-response study to hydroxychloroquine was performed on suspect isolates and controls.

To the best of our knowledge, this is the first time that three SARS-CoV-2 strains have been shown to have a profile of reduced susceptibility to hydroxychloroquine in vitro. Susceptibility to azithromycin was not tested independently, as its effect alone in vitro is limited. In this study, the high-content screening technology, first applied to the high-throughput culture of giant viruses of amoeba then to SARS-CoV-2 [30, 34], was used for the first time to rapidly screen for the susceptibility to drugs of a large panel of viruses. Although other research will be needed to clearly confirm that the procedure can be sufficiently standardised to provide efficient large screening of strains, it has been shown to be efficient at detecting isolates with reduced susceptibility. However, although highly time-consuming and susceptible to many confounding factors, as presented below, when it comes to the fine determination of susceptibility, dose-effect determination using molecular biology remains necessary. Indeed, in vitro sensitivity assays carried out on the same virus can provide divergent results, reflecting great discrepancies due to several key determinants in the experiments. First of all, the cell lines used may harbour different permissivity levels, resulting in differences in viral titres, although those used in these assays needed to be permissive. For SARS-CoV-2, the entry step involves the ACE2 receptor and two independent host protease pathways, TMPRSS2 or the cathepsins B/L that activate the spike viral protein. The virus may not use these two pathways in the same manner, and the expression level of these receptors mediating virus entry are differentially expressed according to the cell lines [35]. For example, VeroE6 engineered to express greater amounts of TMPRSS2 has been used elsewhere, resulting in 100-fold higher titres of SARS-CoV-2 [36]. Inversely, viral titres provided by SARS-CoV-2 infected Calu-3 cells (continuous human lung epithelial cell line) are lower than in Vero cells [36]. It could make sense for the sensitivity assays to use the cells physiologically closest to those of the replication site in vivo. From this perspective, primary cells derived from organ explants were used for sensitivity assays and seemed to present a relevant approach. However, variable effects which are donor-dependent on the sensitivity for some tested drugs should be expected due to differences in viral replication and gene expression [37]. Thus, this approach could be a “false” good idea and testing molecules in a coarse model such as in Vero E6 that has genetic defects in terms of interferon production could help provide evidence of such an effect. Secondly, the multiplicity of infection reported for the drug concentration is not standardised. It seems obvious that the higher the MOI, the lower the relative drug concentration, and the more likely the virus can replicate. This MOI is not even mentioned in some studies. Finally, the time of end point evaluation and the method used for assessing viral replication also varies according to the study from one hour to 120 hours [38, 39]. In addition, the assessment of viral replication by PCR or fluorescent assay or visual inspection to monitor cell viability may not have the same sensitivity. For example, when it comes to visual inspections, some permissive cells such as human intestinal epithelial Caco-2 do not produce cytopathic effects after SARS-CoV-2 infection and thus cannot be evaluated using such a method [40]. As a result, it is risky to draw conclusions on the basis of a single sensitivity test, especially when testing a virus with a high genomic variability.

One of the most interesting perspectives should also be to test multiple viral strains to check the concordance of the results. Currently, in vitro assays essentially use one or two SARS-CoV-2 strains. Our work suggests it is risky to draw conclusions on a single sensitivity test when testing a virus with a high genomic variability. Indeed we observed heterogeneity in the hydroxychloroquine antiviral activity screening of 30 strains and were able to detect three strains with a lower susceptibility profile. For isolates from patients during the first wave of the epidemic, persistence was clearly not associated with a lowered susceptibility profile to hydroxychloroquine in vitro. This confirmed the observation that persistence and severity are rather associated with host factors, as suggested by recent genetic research on COVID-19 severity-associated factors [41, 42] or immunocompromised status [43, 44]. Moreover, genomic analyses did not reveal any modification in these isolates that could explain persistence, neither in the sequence of the strain isolated upon admission nor in that of the strain isolated during the course of the disease. The less susceptible hydroxychloroquine strain, IHUMI-2123, which belongs to Marseille 1 genotype, was isolated in early summer 2020 at the beginning of the second wave, from a patient returning from Tunisia [33], a country in which hydroxychloroquine was massively used [21]. We evidenced a close phylogenetic proximity between all strains of the Marseille 1 clade (IHUMI-2123, IHUMI-2122, IHUMI-2178, and IHUMI-2177) with strains isolated in Senegal and Gambia, two countries which use hydroxychloroquine to treat patients with COVID-19 [45, 46]. We believe that it is possible that the widespread use of hydroxychloroquine in these countries led to strains being selected with reduced, which were later transmitted to the Marseille population. This observation is worthy of analysis on a statistically larger number of strains of this genotype. Paradoxically, the patients tested in Marseille hospitals during the early summer of 2020 and infected by isolates of this genotype presented milder infections and lower mortality than that observed during the first wave of the epidemic, despite the fact that the viral loads in their respiratory secretions were higher. This observation raises several questions that will be difficult to resolve, such as “does the lowered susceptibility to hydroxychloroquine reduce the severity of infection?” or “is it useful to use hydroxychloroquine in such cases or only in patients with severity markers or risk factors such as anticoagulant lupus, which is treated with hydroxychloroquine and the likely efficiency of which is therefore not due to an antiviral effect?” [47, 48, 49]. Finally, the hydroxychloroquine concentration to achieve 50% of viral inhibition was around 3.125 $\mu$M for the two strains with high hydroxychloroquine susceptibility and $>$12.5 $\mu$M for the three strains with reduced susceptibility, which require at least four times more hydroxychloroquine for the same effect (Fig. 2). Moreover, a 90% viral inhibition required around 12.5 $\mu$M for susceptible strains and $>$25 $\mu$M for less susceptible strains. However, these concentrations remain consistent with concentrations observed in human plasma and lungs. An oral intake of 400 mg of hydroxychloroquine led to a maximum blood concentration (Cmax) of 1.22 $\mu$M [50]. However, hydroxychloroquine accumulated 30 times more in the lungs than in the blood [51], allowing a potential efficiency of hydroxychloroquine even against strains with reduced susceptibility. An oral intake of 400 mg of hydroxychloroquine would still be effective in vivo in humans infected with the current strains with in vitro reduced susceptibility to hydroxychloroquine.

However, these genotypic and phenotypic variations could be frequent in the viral populations in the future and could apply to more drugs and need to be considered in the global repurposing strategy. Following the description by Korber et al. [32], we know that the spike population evolved between February and April 2020 and constituted a fast replacing situation by the G614. Recently, a major situation was reported in Denmark, where minks were infected with a strain presenting a few mutations, notably in the spike protein and associated with a selection pressure in a potential zoonotic transfer. Those aspects need to be carefully considered, in terms of testing and using antiviral compounds, but also in terms of epidemiology and vaccination strategy.

To summarise, we observed that the persistence of SARS-CoV-2 in some patients was unrelated to any decrease in the in vitro sensitivity of their strains to hydroxychloroquine. In contrast, we unexpectedly found, in some control strains, a decrease in sensitivity. This characteristic is linked to particular genotypes observed during the second wave of the epidemic in our region and the emergence of the first variants, which were certainly imported from Africa. Finally, it should be recalled that these simple cell models on Vero cells, even if they certainly reflect specific characteristics of the strains tested, are not necessarily relevant for use in humans. In vitro work on other cells such as Calu-3 lung cells [52] or using animal models such as ferrets [53] show a clear divergence of the effect of the drugs tested compared to the Vero model.

Conceptualization—PC, AL and BLS; software—JD, SA and AL; Investigation—CB, MLB, JA, PJ and MG; formal analysis—CB, MLB, JA; writing—original draft preparation—CB, MLB, JA, MG; writing—review and editing—PC, BP, AL and BLS.

The collection of strains was conducted according to the guidelines of the Declaration of Helsinki, and approved by the ethical committee of the University Hospital Institute Méditerranée Infection (No: 2020-029). Patient consent was waived due to the fact that diagnosis by culture was part of routine diagnosis procedure and patients’ data anonymized.

Not applicable.

This research was funded by a grant from the French State managed by the National Research Agency under the “Investissements d’avenir” (Investments for the Future) program with the ANR-10-IAHU-03 (Méditerranée Infection) reference.

The authors declare no conflict of interest.