Characterization of the dynamics of hepatitis B virus resistance to adefovir by ultra-deep pyrosequencing


  • Christophe Rodriguez,

    1. National Reference Center for Viral Hepatitis B, C and Delta, Department of Virology, Hôpital Henri Mondor, Université Paris-Est, Créteil, France
    2. INSERM U955, Créteil, France
    Search for more papers by this author
  • Stéphane Chevaliez,

    1. National Reference Center for Viral Hepatitis B, C and Delta, Department of Virology, Hôpital Henri Mondor, Université Paris-Est, Créteil, France
    2. INSERM U955, Créteil, France
    Search for more papers by this author
  • Paul Bensadoun,

    1. INSERM U955, Créteil, France
    Search for more papers by this author
  • Jean-Michel Pawlotsky

    Corresponding author
    1. INSERM U955, Créteil, France
    • National Reference Center for Viral Hepatitis B, C and Delta, Department of Virology, Hôpital Henri Mondor, Université Paris-Est, Créteil, France
    Search for more papers by this author

  • Potential conflicts of interest: C.R., S.C., and J.M.P. have received research grants from Gilead Sciences. S.C. and J.M.P. have served as advisors for Gilead Sciences.

  • This work was funded by the French National Agency for Research on AIDS and Viral Hepatitis (ANRS). The authors are particularly grateful to Prof. Jean-François Delfraissy, the Director of ANRS, for his continuous support to our UDPS projects. Christophe Rodriguez is the recipient of a doctoral fellowship from ANRS. The hosting laboratory is funded by the Fondation pour la Recherche Médicale (FRM) as an “FRM Team.“ The authors thank Katyna Borroto-Esoda and Gilead Sciences for providing the samples from patients treated with adefovir used in this study and Prof. Manh-Tong Dao for providing the samples from unselected patients newly observed in the Department of Hepatology of the University Hospital of Caen, France.

Address reprint requests to: Jean-Michel Pawlotsky, M.D., Ph.D., Department of Virology, Hôpital Henri Mondor, 51 avenue du Maréchal de Lattre de Tassigny, 94010 Créteil, France. E-mail:; fax: +33-1-4981-4831.


Hepatitis B virus (HBV) resistance to nucleoside/nucleotide analogs is frequent. Ultra-deep pyrosequencing (UDPS) is a powerful new tool that can detect minor viral variants and characterize complex quasispecies mixtures. We used UDPS to analyze the dynamics of adefovir-resistant HBV variants in patients with chronic HBV infection in whom adefovir resistance occurred during treatment. Amino acid substitutions known to confer resistance to adefovir were detected at baseline in most patients. The dynamics of adefovir-resistant variants were complex and differed among patients as a result of evolving differences in variant fitness. UDPS analysis revealed successive waves of selection of HBV populations with single and multiple amino acid substitutions. Adefovir-resistant variants were partially inhibited by lamivudine, but remained fit in its presence. Conclusion: Substitutions conferring HBV resistance to nucleoside/nucleotide analogs exist before treatment, and that the dynamics of adefovir-resistant populations are much more complex and heterogeneous than previously thought and involve thus far unknown amino acid substitutions. The UDPS-based approach described here is likely to have important implications for the assessment of antiviral drug resistance in research and clinical practice. (Hepatology 2013;53:890–901)


base pairs


chronic hepatitis B


hepatitis B e antigen


hepatitis B surface antigen


hepatitis B virus


multiplex identifier




open reading frames


polymerase chain reaction


standard deviations


ultra-deep pyrosequencing


wild type

Approximately 240 million individuals worldwide are chronically infected with hepatitis B virus (HBV).[1] Chronic HBV infection is the leading cause of chronic liver disease and accounts for nearly 1 million deaths every year.[2-4] Chronic hepatitis B (CHB) can be treated with either pegylated interferon alpha or nucleoside/nucleotide analogs. The latter drugs act by directly inhibiting the enzymatic function of HBV reverse transcriptase, the enzyme responsible for viral replication. Five such drugs have been approved for HBV therapy, namely, three nucleoside analogs (lamivudine, telbivudine, and entecavir) and two nucleotide analogs administered as prodrugs (adefovir dipivoxil and tenofovir disoproxil fumarate). The vast majority of HBV-infected patients have an indication for therapy with nucleoside/nucleotide analogs.

The main issues with nucleoside/nucleotide analogs are the need for long-term (probably life-long) administration and the possible emergence of viral resistance.[5] Resistance is characterized by outgrowth of viral populations bearing amino acid substitutions that confer reduced sensitivity to the drug. This is the result of the quasispecies distribution of HBV in infected individuals, that is, the coexistence of a mixture of genetically distinct, but closely related, viral populations in an unstable equilibrium that depends strongly on their relative fitness (i.e., their ability to propagate efficiently) in a specific replicative environment.[5-7] Resistant variants that emerge during treatment are thought to preexist as minor populations preceding treatment, but this remains to be demonstrated in the case of HBV. The fitness cost of drug resistance can gradually be offset by the accumulation of “compensatory“ amino acid substitutions during replication.[5-7] HBV resistance generally results in virological and biochemical breakthrough, followed by accelerated liver disease progression.[5, 6]

Few techniques are available to study HBV resistance in the clinical setting. Population sequencing (or direct sequencing) is the most widely used, but it can only detect the dominant viral population(s). Reverse hybridization with the line probe assay can only detect variants representing at least 5% of the viral quasispecies and can only identify substitutions already known to confer HBV resistance to a given drug.[8, 9] Sequencing of multiple clones generated after polymerase chain reaction (PCR) amplification is cumbersome and time-consuming.[10-13] In addition, analysis of 20 clones per time point provides only a 95% probability that variants representing 10% or more of the viral quasispecies will be identified, whereas random minor variants with no clinical significance may also be highlighted with this method. Novel technical approaches are therefore needed to study antiviral drug resistance.

Next-generation sequencing techniques are capable of generating vast quantities of data without previous knowledge of a particular gene or sequence of interest. The 454 sequencing technology (454 Life Sciences; Roche Diagnostics Corp., Branford, CT), based on ultra-deep pyrosequencing (UDPS), provides longer reads than most other techniques and is well suited to viral resistance studies.[14-17]

Adefovir dipivoxil is still used as first-line monotherapy or as rescue therapy after lamivudine treatment failure in a very large number of HBV-infected patients in settings or areas of the world where more potent drugs, such as tenofovir or entecavir, are not approved or not affordable by the majority of the population. In this context, we used an original approach based on UDPS to characterize HBV genetic variability at baseline and the dynamics of adefovir-resistant HBV variants in patients receiving this therapy, alone or combined with lamivudine, in the case of adefovir treatment failure.

Patients and Methods


Study patients were selected from a 240-week clinical trial of adefovir monotherapy in nucleoside/nucleotide-naïve hepatitis B e antigen (HBeAg)-negative patients, with a double-blinded phase of 96 weeks followed by an open-label, long-term safety and efficacy study of 144 weeks.[18] Serial HBV DNA level measurements were systematically performed during therapy, as well as population sequencing of the full-length HBV reverse-transcriptase gene at baseline and, in patients with detectable serum HBV DNA, at weeks 48, 96, 144, 192, and 240. Resistance-associated substitutions were selected in 29 patients during the 240-week study. Their viral dynamics classified them into three groups: responders (HBV DNA level reduction of more than 3 Log10 IU/mL) who experienced secondary treatment failure (reincrease in the HBV DNA level of more than 1 Log10 IU/mL above the nadir); suboptimal responders (HBV DNA level reduction of less than 3 Log10 IU/mL) who developed amino acid substitutions in a context of a slow, gradual reincrease in viral replication; and responders who developed the amino acid substitutions without any virological breakthroughs. We selected the first 7 patients displaying these three dynamic patterns to study HBV population dynamics during adefovir exposure.

The 7 patients were 5 men and 2 women (26-59 years of age). Their characteristics, including serial reverse-transcriptase sequence analysis by means of cloning sequencing, have been described elsewhere.[11] Correspondence with the numbering in the previous article is shown in Supporting Table 1. None of them had previously received nucleoside/nucleotide analog treatment, and all received 10 mg/day of adefovir for the full study period. In four cases (patients 1, 2, 5, and 7), lamivudine was added during adefovir therapy because of virological failure. Serial serum samples taken at baseline and during adefovir therapy were analyzed by UDPS. The data were then analyzed and interpreted with a specific software package.

Table 1. Prevalence of Amino Acid Substitutions Known to Confer HBV Resistance to Nucleoside/Nucleotide Analogs at Baseline, as Detected by UDPS in the 7 Patients Who Failed on Adefovir Therapy From This Study, in a Comparator Cohort of Patients Who Succeeded on Adefovir Therapy (HBe Seroconversion and Sustained Undetectable HBV DNA) and in a Comparator Cohort of Unselected, Untreated Treatment-Naïve Subjects With Chronic Hepatitis B
CohortsPatients (Nos.)HBV DNA Level (Log IU/mL)rtV173L (%)rtL180M (%)rtA181V/T (%)rtT184 S/A/I/L (%)rtR202G (%)rtM204 V/I (%)rtN236T (%)
  1. < Frequency of the substitution was below the calculated detection cutoff, which is indicated at the bottom of the table for each position. rtM204V/I confers cross-resistance to lamivudine, telbivudine, and entecavir; rtV173L and rtL180M confer improved fitness to rtM204V/I variants; rtN236T confers resistance to adefovir; rtA181V/T confers cross-resistance to lamivudine, telbivudine, and adefovir; rtT184S/A/I/L and rtR202G confer additional resistance to entecavir and improved fitness to variants bearing rtM204V/I, in the presence of this drug.

  2. a

    The number of sequences obtained with the baseline sample was smaller in this patient than in the others, possibly impacting sensitivity of detection.

  3. b

    The detection cutoff of each substitution was calculated as the mean maximal error rate for the given substitution plus 2 SDs.

Study patients (adefovir treatment failure)16.6<<0.47<<0.310.25
Comparator cohort 1 (adefovir treatment successes)A5.2<<0.47<<<<
Comparator cohort 2 (treatment-naïve patients, no adefovir treatment)AA3.5<<11.290.17<<0.23
Detection cutoff of the substitutionb0.

Two independent groups of patients were used as comparators for the frequency of amino acid substitutions associated with adefovir resistance at baseline. The first group included 5 treatment-naïve, HBeAg-positive patients (3 men and 2 women, 15-18 years of age) treated with 10 mg/day of adefovir (104 weeks in 3 cases, 196 weeks in 2 cases) who responded to therapy and achieved an HBe seroconversion with sustained undetectable HBV DNA after therapy. These patients were randomly selected from a larger group of patients included in a previously reported study reporting on the safety, efficacy, and pharmacokinetics of adefovir dipivoxil in children and adolescents with CHB who responded to this therapy.[19] The second group included 11 treatment-naïve patients with CHB (22-60 years of age) consecutively observed for the first time in the Department of Hepatology of the University Hospital of Caen, France, during the 2008-2011 period. Baseline samples were analyzed by UDPS, and the results were interpreted with the same software package.

First HBV DNA PCR Round

HBV DNA was extracted from 200-500 µL of every available serum sample with the QIAamp DNA blood kit (Qiagen GmbH, Hilden, Germany), according to the manufacturer's instructions. A 630-base-pair (bp) fragment, encompassing domains A-E of HBV reverse transcriptase, was PCR-amplified with primers pol1 and pol2, as previously described.[11, 20]


The first-round PCR amplicon from patient 1's baseline sample was cloned into TOPO TA Cloning 2.1 vector (Invitrogen, Carlsbad, CA), transformed by means of One Shot TOP10 chemically competent Escherichia coli (Invitrogen) and cultured in brain-heart infusion agar Petri dishes with 1 mg/mL of ampicilline. Ten colonies were sequenced with M13 primers, according to the TOPO TA cloning protocol (Invitrogen). Sequences were aligned with ClustalX v2.0.9. One wild-type (WT) colony was selected and amplified into BHI medium containing 1 mg/mL of ampicilline. It was then purified by means of the PureLink HiPure Plasmid Filter Maxiprep Kit (Invitrogen). Plasmid DNA was quantified by means of the Quant-iT dsDNA Assay Kit (Invitrogen) and diluted to achieve final concentrations of 108, 105, and 5 × 103 copies/mL. Final DNA amounts were confirmed by means of a quantitative real-time PCR technique on ABI 7500 software (Applied Biosystems, Carlsbad, CA). Each plasmid dilution was then amplified in triplicate in three independent PCR reactions using primers pol1 and pol2, as described above. The nine controls at three dilutions were used to calculate the error rate of the technique at each amino acid position.


A second “nested“ PCR amplification was performed with internal primers pol3 and pol411 that were modified to introduce a GS FLX bead adaptor and a specific identity tag (multiplex identifier; MID). A combination of eight different MIDs was used to identify each sample. Amplicons containing the bead adaptor and MID were then purified in Nucleofast 96 PCR plates (Clontech, Moutain View, CA), according to the manufacturer's instructions. Amplicons were then quantified with the Quant-iT PicoGreen dsDNA kit (Invitrogen), fixed to beads, and amplified in a microemulsion with the GS FLX Titanium emPCR kit (454 Life Sciences; Roche Diagnostics). Amplified beads were purified and enriched according to the manufacturer's instructions, counted with a Beckman Coulter Z1 particle counter (Beckman Coulter, Brea, CA), and deposited in a GS FLX Titanium PicoTiterPlate (454 Life Sciences; Roche Diagnostics). The pyrosequencing reaction was performed with the GS FLX Titanium sequencing kit on an FLX Genome Sequencer (454 Life Sciences; Roche Diagnostics).

Computer Analysis of UDPS Data

Data generated with the UDPS method were analyzed with four in-house software programs included in the PyroPack package, including PyroClass, PyroMute, PyroDyn, and PyroLink, designed, respectively, to classify, filter, model, and link viral sequences generated with these methods.

Sequence data analysis with PyroPack is based on the following procedure. Sequences generated by the GS-FLX sequencer are merged into a single file. Each sequence is identified by means of its sample tag and assigned to the file corresponding to the sample of origin by PyroClass. PyroMute then uses a number of quality filters to eliminate unreliable sequences. Sequence portions with a low Phred quality score are removed.[21] Too-short sequences (<50 bp), including those generated by the previous filtering step, are also eliminated. Sequences are then aligned by means of a modified version of Smith-Waterman's algorithm,[22] in which the resolution of alignment matrices is accelerated and the identification of insertions and deletions (so-called indels) and the correction of length errors in homopolymeric sequences are improved. Quality filters subsequently remove sequences with an identity score <80% relative to the consensus sequence of the patient's baseline sample. In the next step, an array of nucleotide substitutions and corresponding Phred quality scores is built and tested by means of a modified statistical test based on the binomial law[23] to eliminate sequences that are too rare and/or of poor quality, likely the result of sequencing errors. Finally, the remaining nucleotide sequences, considered reliable, are converted into amino acid sequences and their respective frequencies are calculated. Each amino acid change is ascribed to its sequence of origin to subsequently analyze linkages between substitutions.

PyroDyn uses data generated by PyroMute to detect and quantify increases or decreases in amino acid substitutions through mathematical modeling of their variations and correlation with an exponential growth model, which provided the best fit with the observed data. Briefly, in every patient, the frequency of each amino acid at each position is established at each time point. Assuming that HBV resistance is governed by exponential outgrowth of selected resistant variants, the best-fit curve is automatically drawn for each substitution in each patient. Combined cut-off values have been established (r2 > 0.8 and growth rate > mean of the growth rates of all substitutions at all positions plus 2 standard deviations [SDs]) to differentiate significant exponential changes from polymorphism fluctuations.

Finally, PyroLink has been designed to analyze genetic linkages between amino acid substitutions that have been selected by PyroDyn and to characterize the dynamics of viral populations bearing one or several amino acid substitutions over time. Briefly, linkages are automatically sought for every substitution identified with PyroDyn as exponentially growing or decreasing over time. Then, sequences that span all of the identified substitutions are extracted from PyroMute data and, when more than 100 such sequences are available, the proportion of sequences bearing no (WT), one, two, three, and so on, substitutions is calculated and used for subsequent analyses.


UDPS Sequence Data

UDPS analysis of the nine controls described in the Materials and Methods section and tested at different concentrations revealed that the mean maximum error rate generated by the technology was 0.120% ± 0.005%. However, this number varied according to the considered amino acid position. Therefore, only substitution frequencies higher than the mean maximal error rate for the corresponding amino acid position plus 2 SDs were taken into account in the analysis. Plasmids were tested in triplicate at different dilutions ranging from 5 × 103 to 108 copies/mL. No effect of the DNA amount on UDPS results was observed (data not shown).

UDPS was then applied to serum samples taken at baseline and frequently during therapy, representing a total of 119 serial samples from 7 patients who developed resistance to adefovir monotherapy (15-24 samples per patient). Approximately 480,000 sequences (111 Mbp) were generated, with 4,010 ± 843 sequences per sample and a mean length of 382 ± 31 nucleotides (nt) after eliminating excessively short sequences (≤50 nt). Overall, 10.2% of the generated sequences were eliminated by the software because of inadequate quality.

Presence of Amino Acid Substitutions Known to Confer HBV Resistance to Nucleoside/Nucleotide Analogs in Baseline Samples

Table 1 shows the prevalence of amino acid substitutions known to confer HBV resistance to nucleoside/nucleotide analogs detected by UDPS at baseline in the 7 patients. Five samples were found to harbor rtA181V/T substitutions, one harbored the rtN236T substitution, and two harbored rtM204V/I substitutions. No substitutions known to improve the fitness of rtM204V/I variants in the presence of lamivudine, telbivudine, or entecavir (rtV173L and rtL180M) were found in baseline samples. One sample harbored the rtT184S/A/I/L substitution and none harbored the rtR202G substitution; both these substitutions are known to confer full resistance to entecavir when associated with rtM204V/I and rtL180M.

Table 1 also shows the prevalence of amino acid substitutions known to confer HBV resistance to nucleoside/nucleotide analogs detected by UDPS in the two comparator groups described in the Patients and Methods section. Among the 5 HBeAg-positive patients who seroconverted and maintained undetectable HBV DNA levels after adefovir therapy, 2 were found to harbor rtA181V/T substitutions at baseline, whereas none harbored the rtN236T substitution; 2 harbored rtM204V/I substitutions, including 1 in whom it represented nearly 10% of the viral quasispecies and was associated with an rtL180M substitution, suggesting earlier exposure to lamivudine; rtT184S/A/I/L and rtR202G substitutions were found in 1 and 1 patient, respectively. Among the 11 unselected treatment-naïve patients with CHB consecutively observed for the first time in the Department of Hepatology of the University Hospital of Caen, 3 were found to harbor rtA181V/T substitutions (representing more than 11% of the viral quasispecies in one case), and 6 harbored the rtN236T substitution (representing more than 3% of the viral quasispecies in one case); rtM204V/I substitutions were present in 5 patients and rtT184S/A/I/L in 5, whereas none harbored rtV173L, rtL180M, or rtR202G substitutions at baseline.

Characterization of the Dynamics of HBV Resistance to Adefovir in Patient 1

Patient 1 responded suboptimally to adefovir, and the HBV DNA level started to increase gradually after a nadir at month 6, until the end of follow-up at month 24. Figure 1A shows the time course of the HBV DNA level, together with the dynamics of HBV viral populations during adefovir therapy in this patient, as assessed by UDPS. Results are presented as the absolute amount of each viral variant (in Log10 IU/mL) at each time point, taking into consideration the reverse-transcriptase sequence only. The findings in Fig. 1A can be summarized as follows. (1) Immediately after treatment initiation, we observed the persistence of minor variants with the single amino acid substitutions, rtN138K, rtR139K, and rtR212T, that were present at baseline and remained quantitatively unchanged during adefovir administration, whereas the WT virus was profoundly inhibited. (2) Immediately after the HBV DNA nadir was reached at month 2, the absolute amount of WT virus started to increase again, whereas the minor variants gradually lost their relative fitness and became nearly undetectable when outgrowth of adefovir-resistant variants started to be observed. (3) The first wave of resistant variant outgrowth was detected at month 17 and peaked at months 21-22, when the WT virus became undetectable. This wave was composed of viral variants bearing single amino acid substitutions known to confer adefovir resistance, including a majority of rtN236T and a minority of rtA181T. (4) A second wave of outgrowth of adefovir-resistant variants then gradually replaced the first wave, probably subsequent to fitness acquisition by resistant variants bearing single and double amino acid substitutions, including, by order of frequency, rtN236T+rtA181T, rtY245H, rtN236T plus rtY245H, and rtN236T plus rtD238N.

Figure 1.

Dynamics of viral variants on adefovir-based antiviral therapy in patient 1. (A) Dynamics of HBV viral variants as assessed by analysis of UDPS data on the reverse-transcriptase domain. (B) Combined analysis of UDPS data on both the reverse-transcriptase and HBsAg domains, including amino acid substitutions at positions 172/173 of the s region associated with rtA181V/T substitutions. sW172* indicates conversion of a tryptophan codon into a stop codon at position 172. Results are expressed as the absolute amount of each HBV variant (in Log10 IU/mL) at each time point. Dynamics of HBV DNA levels are shown at the back in blue.

Figure 1B shows combined analysis of UDPS data on both the reverse-transcriptase and hepatitis B surface antigen (HBsAg) domains. As expected, the rtA181V substitution was systematically associated with an sL173F substitution in the HBsAg sequence, resulting from the overlapping nature of the open reading frames (ORFs) coding for both viral proteins. The rtA181T substitution was associated with changes at position sW172; their distribution remained stable over time in this patient, with approximately 80%-90% of sW172* (stop codon) and 10%-20% of sW172L. In addition, HBsAg substitutions not encoded by the nucleotide changes responsible for substitutions in the reverse-transcriptase region were linked to reverse-transcriptase substitutions selected by adefovir (sS143T with rtA181T and sM197T with rtN236T). Variants bearing the s143T and sM197T substitutions were present at a very low level at baseline. The double sM197T+rtN236T variant emerged and outgrew at the time of virological breakthrough (Fig. 1B). In contrast, outgrowth of the triple sS143T plus rtA181T(sW172L/*) plus rtN236T variant followed outgrowth of the single sS143T variant by 4 months, both variants being present in equivalent amounts at the end of follow-up (data not shown). Both s143T and sM197T HBsAg substitutions appeared to contribute to the fitness gain of the second wave of adefovir-resistant variants, with preferential outgrowth of sM197T plus rtN236T and sS143T plus rtA181T(sW172L/*) plus rtN236T variants over single or other multiple adefovir-resistant mutants at this late stage of follow-up (Fig. 1B).

Characterization of the Dynamics of HBV Resistance to Adefovir in the Other 6 Patients

The results of UDPS analyses in the other 6 patients are shown in Fig. 2 and summarized below. Contrary to patient 1, amino acid substitutions were not selected in the HBsAg region in the other patients, except those at positions sW172 and sL173 associated with rtA181V/T, when present.

Figure 2.

Dynamics of viral variants on adefovir-based antiviral therapy in patients 2-7. Results are expressed as the absolute amount of each HBV variant (in Log10 IU/mL) at each time point in patients 2 (A), 3 (B), 4 (C), 5 (D), 6 (E), and 7 (F). Dynamics of HBV DNA levels are shown at the back in blue.

Patient 2 was a suboptimal responder who experienced a slow, gradual reincrease in viral replication. In this patient (Fig. 2A), WT HBV declined gradually during adefovir administration, but reincreased when treatment was stopped after approximately 1 year. When adefovir was reintroduced a few weeks later, WT HBV declined again slowly and plateaued at approximately 104 IU/mL. The emergence of resistance was characterized by simultaneous selection of variants with the single rtN236T and rtA181V(sL173F) substitutions at week 27. Subsequently, the rtA181V(sL173F) variant became predominant and was responsible for the virological breakthrough. This variant was partially inhibited, but remained dominant, when lamivudine was added to adefovir after 43 months of therapy.

Patient 3 was a responder who experienced a virological breakthrough. In this patient (Fig. 2B), resistance occurred at month 29 and was characterized by initial outgrowth of HBV variants with single or double amino acid substitutions at positions rtA181 and rtN236. In contrast to patient 2, a variant with the single rtN236T substitution took over and was responsible for the virological breakthrough. As in patient 2, this variant was partially inhibited by lamivudine, but remained predominant on combination therapy.

Patient 4 exhibited a mixed virological response pattern and a more complex resistance pattern (Fig. 2C). This patient had a suboptimal response to adefovir. During the plateau phase, which lasted approximately 20 months, with mild fluctuations, WT HBV was gradually replaced by a mixture of variants with single (rtY124H or rtN236T), double (rtY124H plus rtN236T), and triple (rtY124H plus rtN236T plus rtN238T) amino acid substitutions that replicated at low levels. WT virus returned when adefovir treatment was interrupted. Adefovir was reintroduced approximately 2 months later, and resistance then developed, along with a typical virological breakthrough resulting from outgrowth of a viral population bearing the single rtN236T substitution. This variant was partially inhibited by lamivudine. Subsequently, on adefovir-lamivudine combination therapy, the same mixture of single, double, and triple variants as observed earlier during therapy emerged and replicated at a low level.

Patient 5 (Fig. 2D) exhibited a simple resistance pattern, characterized by a suboptimal response to adefovir and virological breakthrough at month 37, resulting from outgrowth of a resistant variant bearing two amino acid substitutions (rtA181V(sL173F) plus rtN236T).

Patient 6 (Fig. 2E) responded to adefovir, but low-level viremia persisted, with transient selection of a variant bearing an rtN238E substitution. Subsequently, a variant bearing the double rtN236T plus rtN238E substitution took over and persisted at a low level (approximately 102 IU/mL).

Finally, in patient 7, who initially responded and subsequently broke through (Fig. 2F), the virological breakthrough was related to the selection of a major viral population bearing the rtA181V(sL173F) substitution and a minor population bearing the double rtA181V(sL173F) plus rtF221Y substitution. Both variants were partially inhibited, but remained dominant when lamivudine was added to adefovir.


Drug resistance is the principal cause of antiviral treatment failure, which may result in clinical disease progression. Next-generation sequencing technologies, such as UDPS, have the capacity to generate thousands of sequences from complex genomic mixtures, including sequences from dominant, intermediate, and minor viral populations.[24, 25] UDPS-based GS FLX technology provides sequence reads of sufficient length to span the region of interest when studying HBV resistance to nucleoside/nucleotide analogs. To interpret the data generated with this method, we developed an original package of four complementary software programs capable of analyzing large numbers of sequences (nearly 500,000 in this study) in the specific context of viral resistance and used it in this study.

The HBV DNA level in the starting sample may theoretically have an effect on the sensitivity of detection of minor quasispecies variants by UDPS. This did not influence our description of baseline distributions of HBV variants because all patients had high viral levels in the absence of therapy. On adefovir or adefovir-lamivudine treatment, we were able to generate sequence information by UDPS down to HBV DNA levels of the order of 2-3 Log10 IU/mL. To avoid a bias linked to differences in the HBV DNA levels in different blood samples, we chose to express the results as absolute amounts of viral variants in each sample (in IU/mL), rather than proportions of the full quasispecies. In addition, the use of PyroDyn and PyroLink software allowed us to avoid a bias related to the HBV DNA amount in the sample, because only viral populations that were exponentially growing or decreasing over time in serial blood samples were detected and described, regardless of HBV DNA content.

Preexistence of resistant HBV variants in patients never exposed to nucleoside/nucleotide analogs is an accepted concept. However, classical techniques, such as population sequencing, reverse hybridization, or even cloning of PCR products followed by sequencing of the clones, are generally not sensitive enough to detect variants present in very small proportions because of their poor relative fitness, compared to WT virus. UDPS has been used to assess the preexistence of HBV variants resistant to nucleoside/nucleotide analogs in a few studies. However, these works suffered from important methodological flaws, including lack of sensitivity, no consideration of the error rate of the method to establish reliable cutoffs and ensure specificity, too-short genomic region analyzed, and/or no linkage studies.[17, 26, 27]

Using UDPS, we found that variants with amino acid substitutions at positions rtA181 and rtN236 were already present as minor populations at baseline in most of the treatment-naïve patients who subsequently developed adefovir resistance, with a sensitivity ≤0.22%. These substitutions were also detected during therapy in the remaining patients, suggesting that they may also have been present at baseline, but in amounts too small to be detected by UDPS. Frequency of adefovir resistance substitutions at baseline may have been overestimated, compared with the general population, because the patients we studied were selected because adefovir resistance occurred during treatment. To address this question, we tested at baseline two additional groups of patients, including a cohort of HBeAg-positive patients who seroconverted to anti-HBe and remained HBV DNA undetectable after successful adefovir therapy and a group of unselected treatment-naïve patients newly observed in a tertiary referral center in France. In the latter group, which was comparable in age, gender, and HBeAg status to our 7 patients who failed on adefovir therapy, a similar prevalence of rtA181V/T and rtN236T substitutions was found at baseline, ruling out an overestimation of the frequency of adefovir resistance substitutions in our study cohort. In the HBe seroconverter group, 2 patients harbored rtA181V/T substitutions, but none of them harbored the rtN236T substitution. This could suggest a lower frequency of UDPS-detectable amino acid substitutions in these young adults than in older patients at baseline, possibly resulting from the shorter duration of infection. However, interpretation should be extremely careful, given the small number of patients studied in each group that did not allow for reliable statistical comparison. Substitutions at position rtM204, which confer cross-resistance to lamivudine, telbivudine, and entecavir, were also found as minor populations at baseline in several patients from the three groups, whereas amino acid substitutions that confer resistance to entecavir when associated with rtM204 substitutions were more rarely found. These results were in keeping with the recent report, based on UDPS analysis, of preexisting amino acid substitutions known to confer resistance to nucleoside/nucleotide analogs in 3 patients who subsequently responded to this therapy and 1 patient who developed multidrug resistance.[27] Larger scale UDPS studies are now needed to assess the prevalence of primary resistance to nucleoside/nucleotide analogs in the general HBV-infected population.

During treatment, UDPS data analysis from a large number of serial samples obtained over several years allowed us to thoroughly characterize the complex dynamics of HBV populations on adefovir therapy, revealing successive waves of selection of HBV viral populations (Figs. 1 and 2). Our findings can be summarized as follows:

  1. The dynamics of viral variants differed among the patients, despite the fact that they were receiving the same treatment, emphasizing the importance of the HBV quasispecies composition at the start of therapy and of individual viral variant fitness, both of which are patient specific.
  2. In patients 1 and 4, the WT virus rapidly declined at the beginning of treatment, whereas a few preexisting variants did not appear to be affected, suggesting primary resistance. These variants were soon replaced by variants with single amino acid substitutions (including at position rtA181 or rtN236), which were, in turn, replaced by more complex variants with multiple amino acid substitutions.
  3. Other patients exhibited a different pattern of resistance, with initial selection of variants with multiple substitutions, that were subsequently replaced by simpler variants.
  4. Other substitutions, such as those at position rt238 in patients 1, 4, and 6 (including reversion to N in patient 1) and at rtY245 in patient 1, appeared to be associated with a fitness gain of rtN236T variants. Interestingly, rtN238T has already been reported in 2 patients who developed resistance to adefovir.[12, 28]
  5. Variants with substitutions at both rtA181 and rtN236 were present during follow-up in several patients and finally took over in two cases (patients 1 and 5).
  6. The addition of lamivudine always reduced HBV DNA levels, but did not alter the relative fitness of adefovir-resistant variants, which remained dominant during combination therapy. Interestingly, variants with rtA181V/T substitutions partially responded to lamivudine.
  7. As expected, the rtA181V substitution was systematically associated with the sL173F substitution as a result of overlapping ORFs coding for HBV reverse transcriptase and HBsAg. The rtA181T substitution was present in large amounts in patient 1 only and was associated with an sW172L substitution in 10%-20% of variants and a stop codon in the remaining 80%-90% of variants during therapy, suggesting transcomplementation of defective variants by sW172L variants. Positions sW172 and sL173 of HBsAg are located on the internal side of the membrane and are thus unlikely to play a role in the immune control of infection that may have influenced the fitness of the corresponding variants.
  8. As shown in patient 1, substitutions in other genomic regions (such as HBsAg) not encoded by the nucleotide changes responsible for substitutions in the reverse-transcriptase region selected by adefovir can be associated with the latter and play a role in viral fitness in vivo. This hypothesis is strengthened by the fact that the two selected substitutions, sS143T and sM197T, are located at the external side of HBsAg, within known immunodominant epitopes. However, this was noted in only 1 patient in this study.

Adefovir dipivoxil is still widely used worldwide, alone or in combination with lamivudine. As shown in this study, variants with amino acid substitutions known to confer resistance to various nucleoside/nucleotide analogs, including adefovir, can be detected in a substantial proportion of treatment-naïve patients with CHB. Larger scale studies are now required to determine whether baseline testing with UDPS will be useful to orientate HBV treatment strategies. In addition, next-generation sequencing methods, such as UDPS, could be of considerable interest for early diagnosis of viral resistance during antiviral therapy. Indeed, with our approach, resistance could be diagnosed at approximately the same time as with cloning and sequencing (thus considerably earlier than with population sequencing), in a user-friendly and rapid way, compatible with clinical practice (data not shown). This will be facilitated in the future when the costs are reduced and the technology becomes easily available through specialized platforms.

In conclusion, using an original software package for analyzing viral sequences generated by UDPS and other next-generation sequencing methods in the context of antiviral resistance, (1) we showed that substitutions conferring HBV resistance to nucleoside/nucleotide analogs preexist in patients who have never been exposed to these drugs, (2) we characterized the complex and heterogeneous dynamics of adefovir-resistant viral populations in a group of HBV-infected patients in whom resistance emerged during long-term adefovir therapy, and (3) we identified thus far unknown amino acid substitutions that appeared to play an important role in HBV resistance to adefovir. These findings will also be helpful for understanding resistance to tenofovir, which shows cross-resistance with adefovir in vitro. Our findings imply that next-generation sequencing data analysis will have a number of applications in viral resistance assessment, as we recently reported with hepatitis C virus and human immunodeficiency virus.[29, 30]


The authors thank Thierry Ravard for his help with mathematical modeling and Françoise Roudot-Thoraval for her help with statistical tests, as well as Katyna Borroto-Esoda and Manh-Tong Dao.