Prevalence of Single and Multiple Natural NS3, NS5A and NS5B Resistance-Associated Substitutions in Hepatitis C Virus Genotypes 1–4 in Italy

Natural resistance-associated substitutions (RASs) are reported with highly variable prevalence across different HCV genotypes (GTs). Frequency of natural RASs in a large Italian real-life cohort of patients infected with the 4 main HCV-GTs was investigated. NS3, NS5A and NS5B sequences were analysed in 1445 HCV-infected DAA-naïve patients. Sanger-sequencing was performed by home-made protocols on 464 GT1a, 585 GT1b, 92 GT2c, 199 GT3a, 16 GT4a and 99 GT4d samples. Overall, 20.7% (301/1455) of patients showed natural RASs, and the prevalence of multiclass-resistance was 7.3% (29/372 patients analysed). NS3-RASs were particularly common in GT1a and GT1b (45.2-10.8%, respectively), mainly due to 80K presence in GT1a (17%). Almost all GTs showed high prevalence of NS5A-RASs (range: 10.2–45.4%), and especially of 93H (5.1%). NS5A-RASs with fold-change >100x were detected in 6.8% GT1a (30H/R-31M-93C/H), 10.3% GT1b (31V-93H), 28.4% GT2c (28C-31M-93H), 8.5% GT3a (30K-93H), 45.5% GT4a (28M-30R-93H) and 3.8% GT4d (28V-30S-93H). Sofosbuvir RAS 282T was never detected, while the 159F and 316N RASs were found in GT1b (13.4–19.1%, respectively). Natural RASs are common in Italian patients infected with HCV-GTs 1–4. High prevalence of clinically-relevant RASs (such as Y93H) supports the appropriateness of HCV resistance-test to properly guide DAA-based therapy.

In the last few years, the management of chronic Hepatitis C virus (HCV) infection was revolutioned by the introduction of new anti-HCV therapies based on direct-acting antiviral agents (DAAs). Short, highly effective and well tolerated interferon-free combinations of DAAs targeting three distinct non-structural HCV proteins (NS3-4A protease, NS5A, and NS5B polymerase) are now available for clinical use, allowing to reach outstanding cure rates even in patients that once were considered "difficult-to-treat" 1 .
Natural presence of RASs is a unique characteristic of HCV, not shared with other viruses responsible for chronic infections, such as HBV or HIV. It depends on the extremely high intrinsic genetic variability of HCV, consequence of its high mutational rate and replication turn-over, along with lack of proof-reading activity by viral polymerase.
Natural RASs are reported with highly variable prevalence across different HCV genotypes (GTs) 10 but some of them, when present, are able to significantly reduce SVR rates to specific DAAs in complex patients (infection by HCV GT1a or 3, high baseline viremia, previous treatment experience, cirrhosis presence) 1,5,9,11,12 . Even with newer DAAs, for instance, patients with GT3 infection and natural RASs may more frequently experience virological failure 13,14 . Under these conditions, a tailored approach in some (not infrequent, though) real-life situations may help in bringing the cure rate closer to the 98-99% 5,9,11,15 . Thanks to new and highly effective DAAs, HCV treatment is moving towards simpler regimens, the chances to use proficiency very short and ribavirin-free regimens will probably rely on a pre-selection of eligible patients, a selection that could likely take into account also natural resistance. So far, the natural prevalence of RASs was not extensively defined for some populations, such as non GT1 infected patients, patients with cirrhosis, and/or patients with previous IFN experience. In addition, very few reports on circulation of HCV GTs and subtypes in Italy are present in literature to date, as large dataset of HCV sequences were not available. A recent review of HCV GTs distribution in selected West European regions from 2011-2015 estimated, in Italy, a 62% prevalence of GT1 infection (21.8% 1a, 37.4% 1b, 2.8% not specified), 15.4% GT2, 14.9% GT3, 7.6% GT4 and 0.1% GT5 16 .
The aim of this study was to extensively characterize the presence of natural RASs in the NS3, NS5A and NS5B regions of HCV, in a large clinical multicentre database from real-life clinical practice in DAA naïve patients, able to represent the main HCV GTs and subtypes circulating in Italy.
Prevalence of NS5A RASs did not significantly differ according to presence of cirrhosis, nor to treatment experience, (see Supplementary Fig. S3).

Discussion
Thanks to a nation-wide collaboration (Vironet C), the present study collects a large number of sequences from DAA-naïve patients, able to represent the circulation of HCV resistant strains in Italy. Among the 1455 patients included in the analysis, 20.7% were infected with viral strains harboring at least one RASs in one of the 3 DAA-targeted regions.
Notably, when the analysis was limited to amino acid substitutions associated with a clinically relevant level or resistance (defined as "likely resistant"), their prevalence dropped to 6.8% in GT1a, 10.3% in GT1b and 8.5% in GT3a. In addition, "likely resistant" substitutions were found with high prevalence in GT2c patients (28.4%, mainly due to F28C presence), and GT4a (45.5% for L30R and V28M presence). To our knowledge, F28C essentially affects the efficacy of daclatasvir in GT2a replicons in vitro (FC = 400) 20 , but no data are available for F28C in GT2c patients treated with daclatasvir or velpatasvir. However, it is interesting to note that in Italy the prevalence of F28C in DAA naïve population (25.4%) is much lower than that found in NS5A-failing GT2c patients (66.7%, 4/6; p = 0.05) 21 . Since GT2 patients are generally treated with short, RBV-free regimens, the role of F28C in treatment failure will probably deserve further attention.
Much better defined is the role of other "likely resistant" viral variants, such as those harboring substitutions at NS5A positions 93 and 30. In our population, the Y93H was far the most common natural NS5A-RASs detected, with an overall prevalence of 5.1% (56/1090), but reaching 10.1% prevalence in GT1b and 9.1% in GT4a. Recently, this RASs has been reported with very high prevalence also in other "rare" GT4 subtypes, such as GT4b (50%) and GT4r (13%) 10 .
The Y93H was associated with a significant drop in DAA efficacy in some HCV GTs, such as GT1a and GT3 and clinical conditions, as presence of cirrhosis 11,22,23 . For instance, even though tested on a very small number of patients, the Y93H presence in GT3 IFN-experienced, DAA naïve patients reduced by 50% the efficacy of 12-weeks glecaprevir/pibrentasvir treatment 22 . In the present study, Y93H was rarely detected in both GT1a (0.3%) and GT3a (4.5%). This result reflects that of a recent global analysis of natural HCV RASs in 7893 subjects, where the Y93H was found with a prevalence of 11% in GT-1b and 6% in GT3a, while it was never detected in GT1a 10 .
In clinical trials, the presence of multiple natural NS5A RASs was associated with a further reduction in treatment efficacy as compared to single RASs 11 . The detection of multiple NS5A RASs in NS5A-naïve patients is quite uncommon. In our population, only 2.8% of patients (31/1090) had more than one natural NS5A RAS, and 0.6% (7/1090) had "likely resistant" RASs. Multiple RASs detection was more common in GT1b (4.8%), a GT considered easy-to-treat and thus not usually investigated for natural RASs and preferentially treated with short and RBV-free regimens. Nevertheless, no data are available on a possible impact of multiple NS5A RASs in   treatment efficacy for GT1b infected patients, and results obtained in GT1 up to now comes from a limited number of patients, and could be confounded by the presence of RASs with major impact even in the absence of other NS5A polymorphisms 11 . In addition, as full-length sequencing was not performed in our study, we cannot drawn the conclusion that individual amplicons of each region actually belong to the same virus 24 . NS3 RASs were also common (20.4%, 211/1032), particularly due to their high prevalence in GT1a (45.2%, 152/336), concordant with previous literature data 17,25 , and consequence of the frequent detection of Q80K RAS (17.3% in GT1a, 6.3% overall). Much less frequent were natural RASs at position 168, always found with a prevalence <5%.
Along with HCV sequence data, we were also able to collect clinical information, such as treatment history and liver status, that allowed a more in-depth characterization of the phenomenon of natural HCV drug resistance. Overall, we found no significant associations among presence or absence of cirrhosis and prevalence of natural RASs. Similarly previous treatment with IFN-based regimens was not significantly correlated with higher RASs prevalence in any of the HCV-GT analyzed. However, we noticed that both L159F and C316N were more frequently detected in interferon/ribavirin-experienced (7% and 11% respectively) than in naïve patients (3% and 4% respectively). The reason of this different prevalence, though not statistically significant, is unclear. Ribavirin is a guanosine analogue that, among the proposed mechanism of action, may directly inhibit NS5B polymerase 32 and/or promote mutagenesis 33 , thus potentially contribute to explain the more frequent generation and then selection of these substitutions in particular conditions (such as in GT1b). However, no experimental model of viral relapse with ribavirin are available, and the lack of clear data on ribavirin resistance makes this hypothesis difficult to further support. This is one of the few studies, to our knowledge, able to assess the prevalence of patients showing contemporaneous presence of multiple RASs on more than 1 DAA target. Indeed, by analyzing within the 4 main HCV-GT circulating in Italy, we found multiclass resistance in 7.3% of patients, most frequently in GT-1b, represented by a combination of RASs in NS3 and in NS5A. Bartels et al. already analyzed the combinations of natural resistant variants across NS3, NS5A, and NS5B in GT1, and in their study they were rare, and most combinations were observed in a single patient 34 .
While the role of single-class NS3 or NS5A RASs has been discussed quite extensively, the role of multiclass resistance is still to be defined. Among DAAs combinations currently approved for clinical use, or soon to be, several involve a combination of NS3 and NS5A-inhibitors 1,[35][36][37][38] . The excellent efficacy of these combinations in DAA-naïve patients makes difficult to interpret the role of multiclass resistance, even though the analysis of the few patients who failed to achieve HCV cure, especially with shorter regimens, may suggest a potential impact of double NS3 and NS5A class resistance. For instance, natural NS3 + NS5A RASs were recently observed in treatment-experienced GT3 patients who failed glecaprevir/pibrentasvir 12-weeks regimen 22 , and the presence of natural RASs specifically relevant for this genotype (i.e. A30K and Y93H in NS5A, ± NS3 polymorphisms) in this category of patients would probably require a prolongation of DAAs treatment duration to 16 weeks 22 . Similar results were highlighted in NS5A-experienced patients, in whom the presence of double-class resistance for glecaprevir/pibrentasvir is undoubtedly much relevant, and the impact of retreatment efficacy is higher 39,40 .
In conclusion, our study contributed to a better definition of the presence and circulation of resistant viral strains in a large Italian population of patients infected with HCV-GTs 1-4. We were able to uncover natural RASs presence across all HCV-GT analyzed, highlighting multiple NS5A-RASs presence in 2.8% of patients, and multiclass resistance in 7.3%. Clinical application of these findings remains challenging, as genotypic resistance testing is not available everywhere, and prescription of DAA is often limited by costs and reimbursement policies. Nevertheless, especially in countries with high prevalence of HCV infection and where RAS testing is available and validated, the knowledge of natural RASs burden in specific HCV GTs and subtypes may help focusing HCV diagnostic interventions in clinical conditions and settings that would maximize DAA efficacy rate, potentially avoiding wrong and expensive therapies.

Methods
Patients. This is an Italian multicentre, observational study involving DAA-naïve patients with chronic HCVinfection, recruited from 23 Italian clinical centres between 2011 and 2016. A detailed description of Italian regions involved is included in Supplementary Fig. S1.
Eligible patients were DAA-naive who had at least one HCV genotypic resistance testing (GRT) performed by population sequencing on any target region (NS3, NS5A and/or NS5B) for routine clinical purposes or for research. Approval by ethics committee was deemed unnecessary for all patients evaluated for diagnostic purpose according to Italian law, because this was not an hypothesis of clinical trial on medicinal products for clinical use (art. 6 and art. 9, leg. decree 211/2003). In the cases evaluated only for research purpose, approval by the local Ethics Committees and patient written informed consent were obtained. All samples used for HCV sequencing, either for clinical or only research purpose, were all anonymously collected and analyzed according to Italian law (leg. decree 196/2003). This study was conducted in accordance with the principles of the Declaration of Helsinki. All information, including virological, clinical, and therapy data, were recorded in an anonymous database.
HCV Sanger sequencing. Sanger sequencing of the NS3-protease (aa 1-181), NS5A domain I (aa 1-213) and NS5B (aa 1-591) proteins was performed in plasma samples by using home-made protocols specific for each HCV genotype/subtype in 4 Italian laboratories. Overall, 93% of sequences were obtained in Tor Vergata SCiEntifiC RePoRTS | (2018) 8:8988 | DOI:10.1038/s41598-018-26862-y laboratory, and all laboratories who contributed with HCV sequences to the present study, recently participated to a nationwide validation program for HCV GRT, in which concordance in mutation detection and report was > 85% in all 3 genes 41 , thus making the possible discordance of sequencing methods marginally relevant to the present study. Detailed sequencing procedure is reported elsewhere 21,42-45 . HCV resistance mutation analysis. Wild-type amino acids were defined according to references sequences from Geno2Pheno tool 46 .
The list of amino acid substitutions reported to confer reduced susceptibility (>2-fold increased EC 50 ) to currently and soon-to-be available DAAs was defined according to published data, that also allowed the interpretation and classification of clinical impact of RASs 47 . Details are as follows.
NS5A RASs were classified both according to the in vitro fold-change reduction, and to their potential association with resistance in vivo. For 1st generation NS5A-inhibitors, RASs with fold-change >100× are defined as resistance likely; RASs with fold-change 20-100 or only in vivo RAS (no fold-change) are defined as resistance possible; RASs with fold-change 2.5-20× are defined as likely susceptible. For 2nd generation NS5A-inhibitors elbasvir, pibrentasvir and velpatasvir, RASs with fold-change >10× are defined as resistance likely; RASs with fold-change 2.5-9 or only in vivo RAS (no fold-change), are defined as resistance possible; RASs with fold-change <2.5× are defined as likely susceptible. If RASs were found in vivo at virological failure, the level of resistance was increased by 1 level.
The criteria to be included in the present analysis as in vivo RASs were: a) to be found as de novo developed variant in failing patients in literature; or b) to have a demonstrated impact on virological response even if found as natural RAS in literature.
Phylogenetic analysis of NS3, NS5A and NS5B sequences. Phylogenetic analysis was performed to check the possibility of cross-contamination or sample mix-up during laboratory procedures, and also to evaluate concordance with previous subtype assignment by commercial genotyping assays.

Statistical analysis.
Results are expressed as median values and interquartile range (IQR) for continuous data and number (percentage) for categorical data. Categorical variables were compared using the Chi-squared test and fisher's exact test when was appropriate. All the analyses were performed using the SPSS software package (version 23.0) for Windows (SPSS Inc., Chicago, IL).

Mutation covariation and cluster analysis.
Binomial-correlation coefficient (phi) was calculated to assess covariation among RASs, either on the same or on different genomic regions. Statistically significant pairs of RASs were identified by Fisher's exact test, and then corrected for multiple-testing by Benjamini-Hochberg method (FDR = 0.05).
In order to identify and summarize higher-order interactions of RASs, we transformed the pairwise phi correlation coefficients into dissimilarity values. A dendrogram was then computed by hierarchical clustering, and its stability was assessed from 100 bootstrap replicates.
All analyses were performed in R software. The details of this explorative data analysis procedure have been described elsewhere 49 .