Recurrent and novel evolutionary pathways drive in vitro HIV-1 lenacapavir resistance with diverse phenotypic consequences

Abstract

The small-molecule inhibitor lenacapavir (LEN) has firmly established the HIV-1 capsid as a promising clinical target. However, early studies have shown that HIV-1 can develop resistance to LEN by acquiring capsid mutations. These mutations are restricted to the drug-binding pocket, but the breadth of the virus’s capacity to evolve under LEN-directed pressure remains incompletely understood. Here, we employ an accelerated evolution approach to probe HIV-1’s capacity for LEN resistance. Despite strong sequence conservation across thousands of clinical strains, two positions within the binding pocket exhibit high mutational tolerance and diverse resistance profiles. These low-level resistance mutations recurrently select the high-level resistance mutation M66I and together with further mutations, synergistically enhance viral fitness by restoring nuclear entry. Subsequently, multiple charge-reversing substitutions distal to the binding pocket emerge. Located at inter-hexamer interfaces, these substitutions modulate capsid stability and mitigate LEN’s capsid-disrupting effects. A highly resistant variant carrying the E213K substitution becomes drug dependent; it requires LEN to generate infectious particles. As the novel pathways observed here may represent natural extensions of recurrent ones, this work supports surveillance and forecasting efforts while guiding further drug development to sustain the capsid as a viable antiviral target.

Introduction

Lenacapavir (LEN) is a small-molecule compound developed by Gilead that targets the viral capsid, blocking HIV-1 infection by interfering with both early and late stages of the viral replication cycle¹. Its mechanism of action involves direct binding to the interface of two monomers of the CA hexamer of HIV-1, the basic building block of the capsid lattice^1,2. Clinical testing of LEN yielded encouraging safety and efficacy outcomes, and led to its approval as the first-in-class capsid inhibitor for highly treatment-experienced individuals living with multidrug-resistant HIV-1 infection³. LEN, a long-acting agent, has been shown to be potent in preventing HIV-1 infection^4,5, thus garnering attention as a transformative tool in global HIV prevention efforts.

Despite its picomolar potency, LEN—like all other antiretroviral agents—faces the challenge of viral resistance. The relative sequence conservation of CA, attributed to its stringent functional and structural requirements^6,7, is often cited as the basis for LEN’s high genetic barrier to resistance. Indeed, amino acid residues near the LEN binding site in CA are highly conserved, with sequence identity ranging from 94 to 100% across major HIV-1 subtypes^6,7,8. However, HIV-1 does develop LEN resistance in both in vitro and in vivo settings^{1,9,10,11,12,13}. Most amino acid substitutions associated with drug resistance mutations (DRMs) are located near the LEN binding site and confer resistance levels ranging from a few-fold to over 1000-fold. High-level DRMs, such as the well-known M66I substitution, often come at a significant cost to viral replicative fitness; however, they still emerge in individuals treated with LEN¹². Although clinical trials have highlighted LEN’s strong potential as a pre-exposure prophylaxis (PrEP) agent^4,5, two individuals who acquired HIV-1 during LEN-based PrEP carried the LEN-resistant N74D variant⁵. Given that N74D is rarely observed among LEN-untreated HIV-1-infected individuals¹⁴, this is most consistent with de novo evolution rather than selective transmission.

While both in vitro and clinical studies have revealed LEN-resistant mutations, the extent of HIV-1’s ability to acquire higher-level resistance remains incompletely understood. It is plausible that the virus has not yet fully exploited this potential due to several factors. First, extensive studies on HIV-1 CA biology, including viral escape from antiviral pressures, have demonstrated the virus’s capacity to adapt to hostile conditions through mechanisms, such as altering binding sites or modulating capsid stability¹⁵. These findings underscore that while CA is generally genetically fragile¹⁶, it also possesses phenotypic flexibility that the virus can leverage to overcome antiviral interventions. Second, many DRMs identified during early stages of in vitro or in vivo evolution are single amino acid substitutions. However, recent studies, including findings from the CAPELLA trial, have revealed that LEN resistance can evolve into more complex mutational patterns involving multiple substitutions¹². This mirrors the resistance trajectories observed with other drug classes, such as diarylpyrimidine analog non-nucleoside reverse transcriptase inhibitors and integrase strand-transfer inhibitors, where resistance accumulates through combinations of mutations, particularly in the context of pre-existing variants^17,18. Currently, LEN is approved for use in a specific subset of people living with HIV-1, which has limited its widespread application. However, as its use expands in both treatment and prevention settings, the selective pressure on the virus will increase—potentially accelerating the emergence and evolution of resistance.

In this study, we aimed to uncover how HIV-1 circumvents LEN pressure at the earliest mutational steps and acquires high-level resistance. A multivariant selection and adaptation strategy allowed us to examine evolutionary trajectories and the underlying drivers of resistance. Our results identified known and previously unrecognized resistance pathways, offering insights into the adaptive potential of the HIV-1 capsid as well as diverse mechanisms and phenotypic consequences of LEN resistance.

Results

Mutational tolerance in conserved positions of HIV-1 CA

Among several DRMs reported so far, we focused on positions 67 and 74 in this study, as single-nucleotide mutations at these positions—Q67H and N74D—frequently emerge in individuals receiving LEN^1,11,12. We used site-saturation mutagenesis to probe less-accessible multi-nucleotide DRMs, which often emerge in clinical settings under selective pressure from other antiretrovirals, rather than those selected in vitro^17,19. As controls, we included two additional positions (4 and 18) that influence capsid stability^20,21 (Fig. 1). These amino acid residues are highly conserved across thousands of HIV-1 strains (Fig. 1b, right panel).

Fig. 1: Mutational tolerance and sequence conservation at four positions in HIV-1 CA.

a Schematic of the experimental workflow. Mutant libraries were used to infect MT-4 cells (day 0). At day 2, samples containing ~1% GFP⁺ cells were identified by flow cytometry and expanded to allow virus outgrowth in the absence of LEN. At day 5, GFP⁺ cells were analyzed by flow cytometry, and genomic DNA was harvested for next-generation sequencing (NGS) at day 5 or 7. Created in BioRender. Yamashita, M. (2026) https://BioRender.com/ cqlxtjd. b Sequence logos showing amino acid frequencies among viable mutants identified in this study (left) and among naturally occurring 12,261 HIV-1/SIVcpz strains (right) at the corresponding positions. Source data are provided as a Source data file.

The multiplex variant selection assay consists of two stages: the first phase selects for infection conditions where ~1% of target cells are infected with HIV-1 as determined by virus-encoded GFP reporter, followed by the second viral outgrowth phase in the absence of LEN. (Fig. 1a). Sequence analysis of variants selected 5 days post-expansion revealed highly variable degrees of mutational tolerance in these four positions, with an unexpected diversity of viable variants found in three of the four mutant libraries using an LAI backbone (Fig. 1b, left panel and Supplementary Fig. 1). The frequency of each individual variant ranges from WT-level to stop codon levels, which would yield noninfectious virus and serve as assay background (Supplementary Fig. 1). Position 74 showed the highest mutational tolerance, while position 18 was highly intolerant, with only one viable variant (the WT arginine). Positions 4 and 67 exhibited an intermediate phenotype, with approximately one-half to one-third of the variants detected at frequencies above 1%. Variant frequencies in both Q67X and N74X mutant libraries across two independent experiments (Supplementary Fig. 2) were highly concordant, supporting the robustness of the assay. Although the occurrence of variants revealed by our experiments differs substantially from those in vivo, the second most abundant variant identified in vitro was also the second most abundant in vivo for all positions examined except for position 4 (Supplementary Fig. 3). For instance, the second most prevalent variant in the R18 mutant library encodes a lysine residue—the same substitution that ranks second in sequence databases—potentially reflecting the importance of a positive charge at this position for inositol hexakisphosphate (IP6) binding and dNTP influx^21,22,23. Another key observation relevant to LEN (discussed further below) is the high frequency of Q67H in vitro, which, although much less common overall, is the second most abundant variant observed in vivo (Supplementary Fig. 3). These observations underscore the unexpectedly high mutational tolerance of the LEN-binding region, which contrasts with its strong sequence conservation in natural isolates.

Contrasting LEN resistance potentials of positions 67 and 74

The unexpectedly high mutational tolerance at positions 67 and 74—key residues involved in LEN binding²⁴ —has important implications for the evolution of LEN resistance. To address this, we examined how LEN affects the enrichment of each variant in the Q67 and N74 mutant libraries (Supplementary Fig. 4a). Both exhibited shifted dose–response curves, consistent with reduced sensitivity to LEN. While the Q67 library displayed greater overall resistance (Supplementary Fig. 4b), the N74 library exhibited higher infectivity in a drug range which suppresses WT infection by more than 10-fold. (Supplementary Fig. 4b). Sequence analysis of variant abundance following drug selection revealed strikingly distinct mutational profiles between positions 67 and 74 (Fig. 2). In the N74 mutant library, aspartic acid (N74D) became dominant after selection with 3.1 nM LEN. This well-characterized mutation confers LEN resistance in vitro and has been observed in individuals treated with LEN^11,12. A similar charge-altering substitution, N74E (glutamic acid), also increased in frequency and conferred resistance when tested in a cloned virus (Figs. 2a and S4c). A 2D plot comparing amino acid frequencies in the presence and absence of LEN demonstrated a clear distinction, with N74D and N74E deviating from the other variants (Fig. 2a). The Q67 mutant library exhibited a contrasting resistance pattern (Fig. 2b). This library contained fewer viable variants (~7) than the N74 mutant library. However, most of these viable variants in the Q67 library increased in abundance under the LEN-treated condition, surpassing the WT glutamine (Q67) (Fig. 2b). LEN resistance of these variants was validated using molecularly cloned viruses (Supplementary Fig. 4d).

Fig. 2: Variants selected upon lenacapavir treatment at positions 67 and 74 of HIV-1 CA.

A mutant library of HIV-1 targeting either position 74 (a) or 67 (b) was allowed to spread in the presence or absence of 3.1 nM lenacapavir for 5 days, after which viral DNA was harvested for NGS analysis. The results are shown as sequence logos (left) and 2D plots (right). In the 2D plots, each amino acid is color-coded according to the log₂ fold change in frequency from untreated to lenacapavir-treated samples. UT untreated, LEN lenacapavir. Source data are provided as a Source data file.

Q67 variants enriched following LEN treatment included those carrying the well-known LEN-resistant Q67H mutation. However, Q67H was outpaced by Q67W, Q67F, Q67Y, and, most of all, Q67N (Fig. 2b). Indeed, this ranking was reflected in dose–response analyses with cloned viruses, confirming that these variants (Q67F/N/W/Y) exhibited higher IC₅₀ values than Q67H. (Supplementary Fig. 4d). Among these variants, a Q67N substitution was observed in one participant from the CAPELLA study, whereas the Q67H change occurred in multiple participants within the same study¹². The higher occurrence of Q67H compared to Q67N in clinical settings may be attributed to the high replicative capacity of Q67H in LEN-free conditions, differences in selective pressure in vivo, and codon-based accessibility: Q67H arises from a single nucleotide change from WT, whereas Q67N requires two. We next assessed the viral infectivity and replicative capacity of Q67H and Q67N in primary CD4⁺ T cells. Single-round infectivity was comparable between Q67H and Q67N, while Q67H showed a modest trend toward higher replicative capacity relative to Q67N that did not reach statistical significance. As controls, M66I and N74D exhibited markedly impaired infectivity and replication in primary CD4⁺ T cells (Supplementary Fig. 5). Overall, these results reveal distinct resistance potentials of key sites and highlight fitness and resistance profiles of several previously undescribed, less-accessible multi-nucleotide substitutions.

HIV-1 can develop micromolar-level resistance to LEN

We next used in vitro viral adaptation to investigate the ability of HIV-1 to develop resistance to LEN. Unlike in vivo studies, in vitro systems offer the advantage of allowing viral evolution under controlled conditions with frequent sampling, providing detailed insights into evolutionary trajectories. While a previous study investigated LEN resistance in vitro, it used drug concentrations below the upper limit of physiological levels (~100 nM)¹. To accelerate the adaptation process up to and beyond this point, we started with Q67N, the most abundant variant in our selection experiment (Fig. 2b). We monitored the proportion of virus-infected cells using a replication-competent GFP-reporter virus. Adaptation was performed using a conventional dose-escalation scheme, where our initial dose titration determined that a LEN concentration of 10.9 nM maintained a low but substantial level of GFP-positive cells (~5%) for the Q67N variant (Fig. 3a). After approximately 10 days of culture, GFP-positive cells rose above 10%. In response, we increased the LEN concentration to 21.9 nM, which was soon followed by a more pronounced breakthrough (>50% GFP-positive cells). The supernatant from this culture was used to initiate a second round of adaptation, now requiring 43.8 nM LEN to maintain a steady level of virus infection. Adhering to this scheme, the culture progressed through 175, 350, 700, and finally 2800 nM of LEN in about 40 days. Remarkably, the virus exhibited breakthrough even at micromolar LEN concentrations—over 10,000-fold higher than the IC₅₀ of LEN against the WT virus.

Fig. 3: Adaptation of HIV-1 Q67N CA mutant virus to LEN.

a Dose escalation strategy for viral adaptation to LEN. The graph displays the stepwise increase in LEN concentration (nM) over 40 days (days post selection: dps). The level of virus-infected cells was maintained at a constant level (~5%). When infection levels increased, cells were expanded with increasing and variable LEN concentrations to identify doses that maintained consistent infectivity. This process was repeated until the LEN dose reached 2800 nM. Each arrow denotes a time point when the virus was harvested from the culture supernatant for LEN sensitivity analysis in (b). b Sensitivity of adapted virus variants to LEN. Sensitivity to LEN was assessed in three independent experiments. The parental virus is denoted as Q67N–0 nM, while adapted variants are labeled according to the LEN concentration at which they were harvested. Infectivity was quantified by measuring the number of GFP-positive cells using a GFP reporter virus system. Source data are provided as a Source data file.

To quantify changes in LEN sensitivity of the adapted viruses, we generated dose–response curves using viral supernatants harvested at different time points (Fig. 3b). The starting Q67N virus exhibited ~17-fold resistance relative to WT. After 18 days of adaptation, the virus showed further resistance, with a 37-fold increase in IC₅₀. Virus harvested at 175 nM LEN showed a dramatic increase, with an IC₅₀ of ~490 nM (~1500-fold over WT). The most resistant variant exhibited an IC₅₀ >1 µM. These results demonstrate that HIV-1, given the right combination of mutation and selective pressure, can evolve resistance even at micromolar concentrations of LEN.

Emergence of M66I and CTD mutations at CA hexamer interfaces during HIV-1 adaptation to LEN

Analysis of CA sequences from adapted viruses revealed a stepwise accumulation of both known and novel amino acid substitutions, which increased in frequency alongside rising drug concentrations (Fig. 4a). Throughout the adaptation process, the original Q67N mutation remained consistently present. A key substitution that emerged at 21.9 nM LEN was M66I—a well-characterized resistance mutation previously linked to high-level LEN resistance in both in vitro studies and in HIV-1 variants from LEN-treated individuals^1,12. The frequency of this substitution increased from 55 to 64% between 11 and 14 days post selection. An independent adaptation experiment similarly selected for M66I, reaching 71.4% at 21.9 nM (Supplementary Fig. 6). These parallel outcomes highlight M66I as a recurrent solution to LEN pressure by viruses carrying the Q67N substitution.

Fig. 4: Mutational landscape of Q67N adaptation to lenacapavir.

a Bubble plot showing the frequency of individual amino acid substitutions at increasing concentrations of LEN during viral adaptation. Each column represents a different LEN concentration, and each row corresponds to a specific amino acid change. Bubble size indicates the proportion (%) of each variant in the viral population; variants present at <1% were excluded from the plot. Amino acids are colored according to their chemical properties: polar in green; basic in blue; hydrophobic in yellow. b Rendering of seven HIV-1 CA hexamers (3J34) highlighting the positions of lysine substitutions that emerged at dimeric (yellow) and trimeric (magenta) interfaces. Top right inset: close-up view of the dimeric interface showing the location of N183K and E187K substitutions. Bottom right inset: trimeric interface rotated 70° backward relative to the top view, highlighting the locations of E75K, T210K, and E213K. Source data are provided as a Source data file.

By 175 nM LEN, the adapted population showed three major differences from the earlier 21.9 nM population (Fig. 4). First, the frequency of M66I decreased to 63.2%, coinciding with the emergence of another high-level resistance mutation, L56I (36.6%). The linkage analysis revealed that these two changes were mutually exclusive, indicating that nearly all Q67N-containing variants (99.5%) carried either L56I or M66I, but not both (Supplementary Fig. 7). Second, T107N, a well-documented DRM¹, appeared at moderate frequency (7.0%). Third, a novel substitution, E187K, which has not been described as a DRM for LEN, emerged at high frequency (42.1%). Interestingly, E187K appeared alongside several other CTD mutations, most involving substitutions to lysine: N183K, T210K, and E213K. While the frequency of E187K remained relatively stable across increasing LEN concentrations (42.1% at 175 nM, 36.9% at 700 nM, and 34.1% at 2800 nM), the other CTD substitutions—particularly E213K—continued to increase, reaching 46.7% at 2800 nM. Importantly, these CTD substitutions, along with E75K (another charge-altering substitution), are located at the interfaces between CA hexamers; N183K and E187K map to the dimeric interface, while E75K, T210K, and E213K reside at the trimeric interface (Fig. 4b and Supplementary Fig. 8). Thus, viral adaptation selected for amino acid substitutions both at the drug-binding pocket and at the inter-hexamer interfaces of the CA CTD, which likely cooperate to enable high-level LEN resistance.

Q67N facilitates M66I emergence by providing low-level resistance and fitness compensation

We first assessed the effects of M66I, which emerged early during adaptation at low LEN concentrations. Experiments using infectious molecular clones revealed that the acquisition of M66I by Q67N increased the IC₅₀ of LEN by more than 500-fold (Fig. 5a). M66I has been shown to severely compromise viral infectivity^1,10, as confirmed by our particle-normalized infectivity assay (Fig. 5b), and it is typically accompanied by additional DRMs in clinical samples¹². This suggests that the infectivity defect of M66I likely limits its adaptation unless compensatory mutations are present. Indeed, Q67N substantially restored the infectivity of M66I (Fig. 5b). Furthermore, other low-level DRMs, such as Q67H and A105T, also improved the infectivity of M66I when combined (Fig. 5b). Strikingly, low-level DRMs in combination with M66I are more susceptible to LEN than M66I on its own (Fig. 5c). This was particularly evident for A105T (Fig. 5c). These results suggest that the role of Q67N shifts from a primary resistance determinant to facilitating the emergence and stabilization of the high-level resistance mutation M66I. Thus, Q67N and other low-level DRMs likely serve as enablers of stronger resistance pathways in the evolution of drug-resistant HIV-1.

Fig. 5: LEN resistance and infectivity of M66I-carrying variants.

a LEN dose–response analysis of cloned mutant viruses carrying either Q67N or M66I + Q67N. Infectivity of VSV-G-pseudotyped viruses in MT-4 cells was measured across increasing concentrations of lenacapavir and normalized to the maximum infectivity of each virus. One representative result from two independent experiments is shown. b Effects of low-level LEN DRMs on M66I infectivity. Relative infectivity of variants carrying M66I along with different LEN DRMs was assessed by infecting TZM-bl cells. Luciferase activity was normalized to viral input, as determined by levels of virion-associated RT. Infections were performed in triplicate and repeated twice, with six data points shown (n = 6). Data are represented as the mean with error bars representing the standard error of the mean (SEM). Statistical analysis was conducted using one-way ANOVA followed by Dunnett’s multiple comparisons test to calculate adjusted p values comparing M66I to the other mutants. ****p < 0.0001; **p < 0.01; *p < 0.05. c LEN dose–response analysis of the M66I single mutant and double mutants carrying M66I in combination with Q67H, Q67N, or A105T, performed as described in (a). Shown are the average values with error bars denoting SEM from multiple experiments: n = 4 for M66I and M66I + Q67H, and n = 3 for M66I + Q67N. For M66I + A105T (n = 2), data are shown without error bars and no statistical comparisons were performed. Comparison of dose–response curves using the extra sum-of-squares F test revealed a statistically significant difference among the three datasets (M66I, M66I + Q67H, and M66I + Q67N). Source data are provided as a Source data file.

E187K mutation augments high-dose LEN resistance

After acquiring M66I, the virus gained further resistance while approaching the maximum physiological concentration of LEN (100–200 nM)^5,25. The key amino acid substitution that significantly increased in frequency was E187K (Fig. 4), a change not previously reported in LEN resistance literature. E187K emerged at similar frequencies in combination with either M66I or L56I. An infectious molecular clone of M66I + Q67N + E187K confirmed increased resistance to high-dose LEN (Fig. 6a). The effect of E187K on resistance appears to be context-dependent as E187K alone displays WT LEN sensitivity, if not further susceptibility (Fig. 6b). E187K did not modulate increased resistance when combined with another individual key DRM at a time (Supplementary Fig. 9a). Both double and triple mutants impair the infectivity (Fig. 6c), also supporting the notion that these changes arose to increase the resistance, but not viral fitness. Interestingly, N183K, which emerged after high-dose selection and resides close to E187K, also did not affect LEN sensitivity (Supplementary Fig. 9b). Thus, the resistance-enhancing effect of E187K is unlikely to result from altered drug binding.

Fig. 6: E187K enhances LEN resistance of the M66I + Q67N variant.

Sensitivity of cloned variants to LEN was examined by infecting MT-4 cells with VSV-G-pseudotyped viruses. The left panel (a) compares three variants containing amino acid changes that emerged during adaptation of the Q67N variant, while the single E187K mutant was compared to WT in (b). One representative result from two independent experiments is shown. c Relative infectivity of HIV-1 capsid mutants compared with WT. TZM-bl cells were infected with WT or the indicated capsid mutants, and infectivity was quantified by luciferase activity normalized to reverse transcriptase (RT) input. Infectivity is shown relative to WT. Two independent experiments were performed, each in triplicate (n = 6). Bars represent mean values, with error bars indicating SEM. Statistical analysis was performed using ordinary one-way ANOVA followed by Dunnett’s multiple-comparisons test comparing each mutant to WT. ****p < 0.0001; ***p < 0.001. Source data are provided as a Source data file.

E187K enhances core stability and counters core collapse at high LEN doses

E187K is adjacent to residues M184 and W185, which constitute part of the dimeric interface between CA hexamers. Thus, E187K may influence capsid stability and disassembly, contributing to enhanced LEN resistance. To investigate this, we assessed capsid stability and integrity using cell-free assays. Capsid stability was assessed by measuring core yield after isolation using a spin-through method (Fig. 7a). The results showed that E187K exhibited a ~3-fold increase in core yield compared to the WT virus (Fig. 7b, c). We also used a previously established in vitro core disassembly assay with minor modifications^26,27. Isolated cores were incubated for 1 h at 4 or 37 °C, then spun down to measure the levels of pelletable cores, which represent cores with intact integrity. Incubation at 37 °C drastically reduced the pelletable cores (Fig. 7d, e). However, incubation with 100 µM of IP6 counteracted this reduction and restored the level of pelletable cores (Fig. 7e). Importantly, the E187K variant exhibited a significantly smaller reduction in pelletable cores upon incubation at 37 °C without IP6 compared to the WT virus (Fig. 7e).

Fig. 7: E187K enhances core stability and counteracts LEN-induced core disruption in the context of M66I + Q67N, even at high drug concentrations.

a Schematic of core isolation by ultracentrifugation. Created in BioRender. Yamashita, M. (2026) https://BioRender.com/e05xj1i. b Immunoblot analysis of p24 CA levels in viral lysates and isolated core fractions from WT and E187K mutant viruses. Shown is one representative blot from two independent experiments. c Quantification of core yield from the indicated viruses. Core yield was measured using a core-associated RT activity assay. Bars represent the mean ± SEM, and each dot represents a result from an independent experiment. Statistical analysis was performed using an unpaired two-tailed t-test comparing WT to each mutant. Left panel: n = 6. Right panel: WT (n = 12), M66I + Q67N (n = 9), and M66I + Q67N + E187K (n = 12). d Schematic illustrating the core disassembly assay. Created in BioRender. Yamashita, M. (2026) https://BioRender.com/e05xj1i. e Isolated cores were incubated at 4, 37, or 37 °C with 100 µM IP6 and pelleted by microcentrifugation. Pelletable core levels were quantified using a core-associated RT activity assay and normalized to the 4 °C condition for each sample. Data are presented as mean ± SEM. Left panel: n = 7. Right panel: WT (n = 6), M66I + Q67N (n = 4), and M66I + Q67N + E187K (n = 6). Statistical significance was assessed using two-tailed paired (left panel) and unpaired (right panel) t-tests. f Schematic of LEN-dependent core destabilization. Created in BioRender. Yamashita, M. (2026) https://BioRender.com/6iqgm3z. g Dose-dependent effects of LEN on HIV-1 core stability. Isolated viral cores were treated at 4 °C with increasing concentrations of LEN and pelleted by microcentrifugation. Core integrity was assessed by measuring core-associated RT activity in the pellet fraction. Values are expressed as percent pelletable cores relative to the no-drug control. Data represent mean ± SEM from three independent experiments (n = 3). Asterisks indicate statistically significant differences compared to WT (M66I + Q67N in red; M66I + Q67N + E187K in blue). The asterisk at the highest LEN concentration shown in black indicates a comparison between the double and triple mutant. n = 3. *** p < 0.001; **p < 0.01; *p < 0.05. Statistical significance was assessed using a one-way ANOVA followed by Tukey’s multiple comparisons test. Source data are provided as a Source data file.

Additionally, we compared the triple mutant (M66I + Q67N + E187K) with the precursor double mutant (M66I + Q67N). Interestingly, the double mutant (M66I + Q67N) already showed higher core yields compared to the WT virus, and there was no significant difference in core yield between the double and triple mutants (Fig. 7c). Similarly, core disassembly phenotypes observed by both mutants showed a lower magnitude of disassembly at 37 °C compared to WT, with no significant difference between the double and triple mutants (Fig. 7e).

We then directly tested the effects of LEN on core integrity by incubating isolated cores in increasing concentrations of LEN (Fig. 7f). As expected from known core-disrupting activity by LEN^28,29, LEN treatment resulted in a steady decrease in pelletable cores of the WT virus up to 20 nM, at which the curve plateaued (Fig. 7g). Consistent with their LEN resistance phenotypes, both the double and triple mutants preserved pelletable cores at low to mid-nanomolar concentrations (4, 20, and 100 nM; Fig. 7g). Remarkably, the triple mutant differed from the double mutant at the highest concentration; at 500 nM, the level of pelletable cores was significantly higher for M66I + Q67N with E187K than without it (Fig. 7g). Taken together, these results indicate that E187K, which stabilizes viral cores, counteracts the core-collapsing activity induced by high LEN doses in a context-dependent manner.

Convergent evolution drives shared selection of M66I and CTD lysine increase

We have shown so far that adaptation of the Q67N mutant to elevated LEN doses selected for well-described high-level DRMs, including M66I and L56I, as well as previously unreported mutations in the CTD of CA (Fig. 4). To investigate how the initial DRMs influence evolutionary trajectories in drug resistance, we performed viral adaptation using a pooled set of LEN-resistant Q67 variants (excluding Q67N)—specifically, Q67H, Q67F, Q67W, and Q67Y (Fig. 8). As described above, Q67H is one of the most frequently selected mutations in LEN-treated individuals¹², while the other three variants emerged as LEN-resistant mutants in our multiplex selection experiments (Fig. 2d).

Fig. 8: Adaptation of Q67F/H/W/Y variant pool to lenacapavir.

a Dose escalation strategy for viral adaptation to LEN. The graph displays the stepwise increase in LEN concentration (nM) over 60 days (days post selection: dps). Each arrow denotes a time point when the virus was harvested from the culture supernatant for LEN sensitivity analysis in (b). b Infectivity was assessed in the presence of increasing concentrations of LEN using a GFP reporter virus system. The number of GFP-positive cells was quantified to determine the antiviral sensitivity of each adapted virus variant. c Bubble plot showing the frequency of individual amino acid substitutions at increasing concentrations of LEN during viral adaptation. Each column represents a different LEN concentration, and each row corresponds to a specific amino acid change. Bubble size indicates the proportion (%) of each variant in the viral population. Two samples were harvested at 12.5 nM, with the early time point labeled as 12.5a and the later one as 12.5b. Amino acids are colored according to their chemical properties: polar in green; basic in blue; acidic in red, hydrophobic in yellow. d Lenacapavir dose–response analysis of cloned variants carrying triple or quadruple mutations. Representative data from one of two independent experiments is shown. Source data are provided as a Source data file.

A similar dose-escalation scheme (Fig. 8a) to Q67N was employed for viral adaptation of the pooled variants (Q67F/H/W/Y) and successfully selected for variants with increased LEN resistance (Fig. 8b). NGS analysis of viral sequences revealed both shared and distinct evolutionary patterns in the pooled variants relative to Q67N evolution (Fig. 8c and Supplementary Fig. 10). Selection at the initial 12.5 nM time point (shown as 12.5a in Fig. 8c) primarily enriched the Q67W variant (62.9%), followed by Q67H (22.9%) and Q67Y (14.2%), with Q67F completely eliminated. However, from this point forward, all emerging variants retained the M66I + Q67H combination and accumulated additional mutations (Supplementary Fig. 10). Both substitutions reside within the drug-binding site and were maintained at higher drug concentrations. A notable de novo mutation that emerged during selection of the M66I + Q67H background was the well-known LEN resistance mutation N74D, which became nearly fixed at 200 nM; continued selection at 400 nM subsequently led to acquisition of A77V, resulting in the M66I + Q67H + N74D + A77V genotype. Importantly, the combination of these three mutations (M66I, Q67H, and N74D), together with another low-level DRM (A105T), exactly matches the mutations observed in one of the LEN-treated individuals from the CAPELLA study¹².

Acquisition of high-level resistance to LEN via these changes was confirmed using cloned viruses (Fig. 8d). The two adapted variants found here (M66I + Q67H + N74D, with or without A77V) exhibited an increased level of resistance (Fig. 8d). The levels of resistance match with those of variants identified in two specific participants in the CAPELLA study (M66I + N74D + A105T, with or without Q67H; Supplementary Fig. 11a). This magnitude of the increase in LEN resistance by these Q67H variants is well above the resistance level of Q67H alone, which has a 4.7-fold increase in IC₅₀. Similar to the viral adaptation of Q67N, the combinations of these DRMs restored viral infectivity, particularly when compared to the M66I mutation (Supplementary Fig. 11b). This supports the idea that compensatory mutations were selected to enhance viral replicative capacity rather than to further increase drug resistance.

Further selection at 400 and 800 nM revealed both shared and distinct evolutionary trajectories between Q67N and Q67H (Figs. 4 and 8). Specifically, the CTD lysine substitutions E187K and E213K—previously observed during Q67N adaptation—emerged and increased in frequency with escalating drug doses (Fig. 8c). Different from the Q67N adaptation, M66I + Q67H acquired additional binding-site mutations before (N74D) or alongside (A77V) CTD substitutions; the variant M66I + Q67H + N74D + A77V comprised 76.2% of reads before E187K was even detected. These results reveal both convergent and divergent evolution during the viral adaptation of Q67 mutants to high-LEN concentrations, sharing the initial selection of M66I and later novel CTD substitutions.

Combinatorial mutations drive restoration of HIV-1 infectivity in nondividing cells during in vivo LEN resistance evolution

One key property impaired by the high-level DRM M66I is HIV’s ability to efficiently infect nondividing cells, as shown by Briganti et al.³⁰ and confirmed by our results (Fig. 9). The severe infectivity defect in nondividing cells was shared by another high-level LEN resistance mutation, L56I, which likewise shows strong impairment in nondividing cells (Supplementary Fig. 12a). As described above, we found that low-level DRMs can compensate for the fitness cost imposed by M66I. We therefore asked whether any of these mutation combinations could also restore the ability of M66I-containing variants to infect nondividing cells. Q67N retains the ability to efficiently infect nondividing cells; however, neither it nor the E187K mutation rescued the M66I mutation’s infectivity in nondividing cells (Fig. 9a, b). Other low-level DRMs, such as Q67H and A105T, also failed to do so (Supplementary Fig. 12b).

Fig. 9: Distinct evolutionary trajectories of M66I-carrying variants yield differential infection outcomes in nondividing cells.

a, c Representative flow cytometry plots showing GFP expression in TZM-bl cells infected with HIV-1 variants carrying the indicated capsid mutations, in the absence (top row, −Aph) or presence (bottom row, +Aph) of aphidicolin. Percentages indicate the proportion of GFP-positive cells. b, d Quantification of infectivity in the presence or absence of aphidicolin. Infectivity was assessed by GFP expression and normalized to the untreated (−Aph) condition for each virus. Bars represent the mean ± SEM of biological replicates. Statistical analysis in (b) was performed using an unpaired two-tailed t-test; ****p < 0.0001. WT (n = 8), Q67N (n = 6), M66I + Q67N (n = 4), and M66I + Q67N + E187K (n = 4). In (d) (n = 4), statistical significance was assessed by one-way ANOVA followed by Dunnett’s multiple comparisons test comparing M66I to the other mutants. ****p < 0.0001; **p < 0.01; *p < 0.05. Source data are provided as a Source data file.

In marked contrast, M66I infectivity in nondividing cells was rescued to varying degrees by triple (M66I + Q67H + N74D) and quadruple (M66I + Q67H + N74D + A77V) mutants that emerged in our Q67F/H/W/Y adaptation, as well as triple (M66I + N74D + A105T) and quadruple (M66I + Q67H + N74D + A105T) mutants seen in two specific CAPELLA study participants¹² (Fig. 9c, d). The recovery of infectivity in nondividing cells was more pronounced in the two clinically observed mutants. In particular, the quadruple mutant (M66I + Q67H + N74D + A105T)¹² almost completely restored infectivity in nondividing cells (Fig. 9c, d). Together, these results highlight the divergent evolution driven by the initial mutation, which has a profound effect on the phenotypic outcomes of adapted variants, and underscore the importance of multiple amino acid replacements in fully restoring the ability to infect nondividing cells—an effect critically observed in an in vivo viral evolution setting.

LEN-dependent infectivity conferred by E213K

In addition to E187K, another glutamic acid-to-lysine (E-to-K) substitution in the CTD-E213K—also commonly emerged at high LEN doses in two adaptation experiments (Figs. 4 and 8). Various E213K-containing molecular clones were constructed but found to be noninfectious. As these E213K variants emerged upon replication in high LEN concentrations, we hypothesized that variants carrying E213K may depend on LEN for infection. Indeed, this was the case. The triple mutant virus M66I + Q67H + E213K generated in transfections with increasing concentrations of LEN showed increasing infectivity, contrasting the dose-dependent reduction of the WT (Fig. 10a). Namely, the triple mutant displayed little to no infectivity from 10 pM to 1 nM LEN, followed by a sharp, dose-dependent increase in luciferase activity peaking at 1 µM LEN. This LEN dependency appeared to be largely specific to this mutation combination, except for the M66I + E213K double mutant, which exhibited moderate infectivity at 100–1000 nM LEN (Supplementary Fig. 13a).

Fig. 10: A LEN-resistant and -dependent variant carrying E213K.

a WT or triple mutant viruses were produced in 293T cells transfected with the indicated concentrations of LEN. Culture supernatants were directly transferred to TZM-bl cells without removing LEN from the producer cells or adding fresh LEN to the target cells. Infectivity was assessed by luciferase reporter activity. RLU, relative light units. Results represent the average of two independent experiments, each performed in duplicate (n = 4). Error bars indicate SEM. b The triple mutant (M66I + Q67H + E213K) was produced in the presence or absence of 1 µM LEN. Viruses were concentrated 10×, and LEN was removed during the concentration step. Fresh LEN was added to MT-4 target cells in selected conditions. GFP⁺ cells were counted 2 days post-infection. Data are presented as mean ± SEM. Data represent two independent experiments performed in triplicate (n = 6), except for the −/− condition (no LEN during production or infection), where two data points with 0% infection were excluded for clarity (n = 4). Statistical analysis was performed using one-way ANOVA. c Culture supernatants were harvested, and virion-associated RT activity was measured as an indicator of virus particle production. Data are shown relative to untreated WT. Virus-producing cells were either untreated or treated with 1 µM LEN. Results represent the average of two independent experiments, each performed in triplicate (n = 6). Error bars indicate SEM. Statistical analysis was performed using one-way ANOVA followed by Dunnett’s multiple comparisons test to calculate adjusted p-values. d Effects of LEN during virus production on core yields. Viruses were produced in the presence or absence of 1 µM LEN. Viral cores were isolated using a spin-through method and quantified based on core-associated RT activity serving as a measure of core yield. Results are shown relative to RT levels in the virus input and represent the average of three independent experiments (n = 3). Error bars indicate SEM. Statistical significance was determined using a two-tailed unpaired t-test.; ****p < 0.0001; ***p < 0.001; **p < 0.01; *p < 0.05. Source data are provided as a Source data file.

This initial experiment allowed residual LEN from virus-producing cells to carry over into target cells. To determine the stage at which the triple mutant is rescued by LEN, LEN was removed by concentrating virus stocks and resuspending them in LEN-free medium. M66I + Q67H + E213K viruses produced with LEN but used to infect cells without LEN showed 20–30-fold higher infectivity than viruses made and used without LEN (Fig. 10b). In addition, LEN added only to target cells enhanced infectivity: virus made without LEN showed an 8-fold increase, and virus made with LEN showed a 3.8-fold increase. These results suggest that LEN in producer cells plays a more dominant role in enabling infectivity of the triple mutant.

To further examine the role of LEN during late stages of replication, we measured virion-associated RT activity. As expected, 1 µM LEN reduced RT activity in WT virus particles (Fig. 10c). The triple mutant showed an overall reduction in particle production, but, notably, LEN did not reduce nor increase RT activity in this context—levels were nearly equivalent between treated and untreated samples. These results suggest that LEN-mediated enhancement of infectivity is not due to an increased particle production. When we examined core yields of viruses produced in the presence or absence of the drug, LEN present during virus production drastically reduced core yield in the WT virus, likely due to capsid collapse from CA hyper-stabilization (Fig. 10d). In contrast, the triple mutant displayed two notable features. First, in the absence of LEN, it produced fewer cores than WT. Second, LEN addition rescued core yield to levels comparable to WT virus produced without LEN (Fig. 10d). These findings suggest that LEN packaged into virus particles plays a critical role in proper capsid maturation and/or stability for the triple mutant carrying E213K. Altogether, our results demonstrate that LEN resistance at high doses can lead to strict LEN dependence for the production of infectious virus particles.

Capsid backbone influences lenacapavir resistance evolution

Having established that several amino acid substitutions identified using the LAI-based clone are also observed among resistance-associated changes in the CAPELLA study¹², we next examined whether the CA backbone influences the evolutionary trajectory of LEN resistance. To address this, we generated a chimeric infectious molecular clone in which most of the Gag-coding region was replaced with that of CH505, a subtype C transmitted/founder (T/F) strain. Using a dose-escalation adaptation strategy analogous to that employed for LAI, we initiated selection with a CH505 chimeric virus carrying Q67H, as this mutation was both highly prevalent and became dominant during prior adaptation.

Sequence analysis of Q67H-containing viruses revealed acquisition of a distinct set of secondary mutations (Fig. S14). Notably, selection favored K70R, resulting in the Q67H + K70R combination, which mirrors another mutation pattern observed in LEN-treated participants¹². Upon further escalation of drug pressure, and in contrast to our initial expectations, an additional novel substitution (A174T) emerged in the CTD of CA. Although limited in scope, these findings suggest that the CA backbone can shape LEN resistance pathways, while supporting our model that HIV-1 capsid resistance to lenacapavir can arise through both recurrent and novel evolutionary routes.

Discussion

In this study, we focused on two critical positions in CA that confer low-level LEN resistance and further investigated HIV-1’s capacity to adapt to LEN using a combination of targeted mutagenesis, multivariant selection, and viral adaptation. Mutational profiling of positions 67 and 74 revealed high mutational tolerance at both sites, but with contrasting resistance profiles. Viral adaptation reveals both shared and distinct evolutionary trajectories toward enhanced LEN resistance. These were driven by the stepwise accumulation of known and novel amino acid substitutions acting through diverse mechanisms—including reduced drug binding, increased viral fitness, and altered capsid integrity. We also found one variant heavily dependent on LEN. Together, these results provide insights into the evolutionary potential of HIV-1 CA for LEN resistance and its phenotypic consequences.

Our multiplex assay revealed unexpectedly high mutational tolerance in three out of four mutant libraries (positions 4, 67, and 74). This high mutational tolerance sharply contrasts with the strong sequence conservation observed in publicly available databases^6,7. Despite this mutational flexibility, WT amino acids dominate in all four libraries, likely due to a fitness advantage. Most HIV-1 sequence data are derived from Sanger sequencing, which primarily samples “major variants.” Therefore, the discrepancy between sequence conservation and mutational tolerance is not entirely unexpected. It is also possible that some viable variants identified in this study have a defect in replicating in primary CD4⁺ T cells, although limited analyses of HIV-1 mutants in primary CD4⁺ T cells show infectivity and replicative properties similar to those in MT-4 cells. Nonetheless, our data challenge the view that highly conserved viral sequences are strictly constrained and demonstrate that even evolutionarily conserved regions can accommodate resistance-associated variation under strong selective pressure. Experiments on other amino acid residues key for LEN binding are critical for deepening our understanding of HIV-1 adaptability to LEN.

The high mutational tolerance of residues within the drug-binding pocket (e.g., positions 67 and 74)^1,24 has critical implications for LEN resistance. Amino acid changes at these positions, such as Q67H and N74D, have been observed in clinical studies of LEN treatment¹² and confer resistance through adversely influencing inhibitor binding²⁴. In our experimental setting, this N74D substitution, along with a similar charge-altering change, N74E, were identified as DRMs. Our inability to detect N74S, which appeared alongside Q67H in one LEN-treated individual⁹, may be due to a potential auxiliary role in supporting DRMs at position 67, differences in drug pressure and/or cellular environment between in vitro and in vivo settings.

In sharp contrast, a broader range of mutations at position 67 was enriched during our multivariant selection under LEN pressure, often surpassing the well-characterized Q67H mutation. This supports the utility of site-saturation mutagenesis, which identified mutations that in vivo sampling and conventional resistance selection methods did not reveal. Three factors may be responsible for this discrepancy: resistance level, viral fitness, and codon usage. Although Q67H confers lower resistance to LEN than other variants, it remains the fittest and requires only a single nucleotide change from the WT sequence. In contrast, higher-resistance mutations, such as Q67N and Q67Y, require two nucleotide changes, while Q67F and Q67W require three. Notably, Q67N emerged in a LEN-treated individual¹², while Q67Y was identified through in vitro selection against GS-CA1, an LEN precursor compound⁸. Q67N and Q67Y differ by only one nucleotide from Q67H, suggesting that Q67H may serve as a stepping stone for the evolution of higher-level resistance.

Another key factor shaping viral resistance evolution is drug concentration. In our in vitro adaptation using a pooled virus population containing four variants (Q67F/H/W/Y), Q67H was preferentially selected in combination with M66I, rather than Q67Y or Q67N. While Q67N was highly enriched at 3.2 nM LEN in our selection assay, M66I emerged in viral adaptation experiments at concentrations just above 10 nM. This difference may reflect the use of a dose-escalation strategy in our adaptation scheme. If drug concentrations instead remain fixed or were changed only modestly in a manner that more closely mirrors pharmacokinetic dynamics in individuals receiving LEN, alternative resistance pathways may be selected. Indeed, the complex evolutionary patterns observed, including both M66I-containing mutational signatures and M66I-independent trajectories, may be driven by heterogeneity in drug exposure. The wide range of LEN concentrations observed among clinical trial participants suggests the existence of multiple resistance trajectories, underscoring the complexity of LEN resistance development.

M66I repeatedly arose early across three independent dose escalations of Q67 mutant variants. These in vitro results recapture the in vivo appearance of this high-level DRM¹². Notably, substitutions such as M66I are not frequently observed in the absence of LEN pressure. However, given the high mutational supply generated during untreated HIV-1 replication, single-nucleotide substitutions are expected to arise stochastically in vivo but are unlikely to be detected unless they reach sufficient frequency, which would be constrained by fitness and competition with the WT virus. While double mutants could in principle be generated at very low frequency, their persistence in the absence of selection would be expected to be rare due to associated fitness costs. This, combined with the apparent inability of M66I alone to adapt³⁰, suggests that the evolutionary trajectory plays a critical role in LEN resistance evolution involving M66I. Both Q67H and Q67N, as well as another low-level DRM (A105T), compensate for M66I, with double mutants showing higher infectivity than the M66I single mutant. Consistent with the evolutionary timeline, acquisition of M66I markedly enhances resistance for both Q67H and Q67N. However, interestingly, these double mutants are no more resistant to LEN than the M66I single mutant. These findings suggest a scenario in which low-level DRMs (i.e., Q67H and Q67N) act as initial evolutionary intermediates with moderate resistance but, more importantly, serve as compensatory mutations that improve the fitness of highly resistant but severely impaired M66I variants. As shown by the evolution of variants, such as M66I + Q67H + N74D + A77V in vitro, or similar combinations observed in the phase 2/3 CAPELLA study^12,13, the accumulation of additional mutations after M66I acquisition appears primarily aimed at restoring viral fitness.

Relevant to this accumulation of low-level compensatory DRMs is the restoration of infectivity in nondividing cells, in other words, active nuclear entry of HIV-1 capsid through nuclear pores. Previous studies have shown that single compensatory mutations, which can reduce CypA or CPSF6 binding, can restore infectivity of cell-cycle-dependent CA mutants in nondividing cells^15,31. However, this was not the case in the adapted LEN-resistant variants selected in vitro. Remarkably, during HIV-1 evolution, the acquisition of multiple mutations near the drug-binding pocket eventually restores infectivity in nondividing cells without substantially altering drug resistance. This observation aligns with our prior work studying HIV-1 evolution in HLA-B27–harboring individuals, where earlier CTL escape variants that lost the ability to infect nondividing cells eventually regained this capacity during later stages of evolution³². These case studies of HIV-1 evolution point to the importance of the virus’s ability to transport viral capsids through nuclear pores, enabling infection of cells independent of cell division in vivo. LEN by itself blocks HIV-1 nuclear entry^1,2, an indispensable step in the viral life cycle that allows infection of non-cycling cells. These compensatory DRMs (i.e., N74D, A77V, A105T) all block capsid interactions with CPSF6, a key host factor that allows HIV-1 to use specific nuclear pore proteins, pointing to the use of an alternative nuclear entry pathway used by such LEN-resistant variants^32,33,34,35. One key property of the capsid that appears to correlate with nuclear entry and infection of nondividing cells is its elasticity³⁶. LEN by itself reduces core elasticity, a potential mechanism of LEN-mediated block of nuclear entry³⁶. It will be interesting to investigate how resistance and compensatory mutations that emerged during viral adaptation affect core elasticity in the presence or absence of LEN.

While the combination of M66I and Q67N confers significant LEN resistance, we observed a drastic shift in viral dose–response curves when comparing virus populations selected at 21.9 versus 175 nM LEN. A key amino acid substitution that emerged at 175 nM LEN and enhanced the resistance of the M66I + Q67N double mutant was E187K in the CA-CTD, a novel change not observed in previous studies. The absence of E187K in prior studies may reflect the rarity of sustained selection under high LEN exposure; although physiologically active LEN concentrations can approach ~200 nM, such levels are unlikely to be commonly or persistently achieved in vivo. Our cell-free assay effectively recapitulated LEN’s activity to disrupt capsid integrity^{28,29,37,38,39}. Crucially, only the E187K-carrying triple mutant exhibited significant resistance against the loss of pelletable cores at the highest concentration of LEN (500 nM). These findings point to neutralization of LEN’s core-disrupting activity at high concentrations as a key selective pressure for E187K.

A notable feature observed after viral selection at high LEN concentrations was the dominance of lysine substitutions, three of which result in a loss of negative charge from glutamic acid residues. N183K and E187K are both located on H9 of CA, with N183 directly involved in LEN binding^1,2. However, the N183K mutation by itself did not confer LEN resistance. Critically, all lysine changes map to regions at or near the dimeric (N183K, E187K) or trimeric (E75K, T210K, E213K) interfaces between CA hexamers. It is likely that introducing lysine residues at these interfaces subtly alters the integrity of the capsid lattice, counteracting LEN’s effects on the viral capsid (see further discussion below). In fact, the E187K mutation alone increased core yield, suggesting an inherent core-stabilizing effect. This observation presents a paradox, given that LEN’s core-disrupting activity is thought to originate from its ability to induce hyper-stabilization of the viral core^{2,28,29,37,40,41,42}. One possible explanation is that stabilization of individual CA hexamers can be accompanied by hexamer flattening, which introduces lattice-level strain and destabilizes the assembled core²⁹. Consistent with this view, previous work showed that increased core stabilization, either through mutations or IP6, is associated with enhanced LEN resistance^2,38.

We discovered one LEN-resistant variant that became LEN-dependent. We did not observe intermediate evolutionary steps toward dependence. The LEN-dependency relies on this specific combination of amino acid substitutions (M66I + Q67H + E213K), although M66I + E213K showed a moderate level of LEN dependence. While drug dependence involving HIV-1 CA has been observed with other inhibitors, such as cyclosporine A (CsA)⁴³, PF74⁴⁴, and PF-46396, a maturation inhibitor⁴⁵, this is, to our knowledge, the first report of such a phenomenon arising from resistance to LEN. CsA-dependent mutants require CsA in target cells, whereas maturation inhibitor–dependent variants depend on the drug in virus-producing cells. LEN dependence did not resemble either. While moderate target cell effects were observed, LEN in producer cells was the primary requirement for the LEN dependence, despite no significant change in the level of produced particle. This suggests that the defect in virus particles generated in the absence of this triple mutant (M66I + Q67N + E213K) is qualitative rather than quantitative, likely involving improper maturation and/or capsid instability.

E213 is located at the trimeric interface of the CA hexamer, a region known to be involved in modulating core stability^{26,46,47,48,49,50,51}. We found that E213K reduced core yield, although to a considerably lesser extent than other destabilizing substitutions, such as K203A and Q219A. It has been suggested that balanced repulsion of negatively charged patches at the trimeric interface is critical for optimal lattice stability⁵². The E213K substitution introduces a positive charge at a site that normally bears a negative charge, potentially disrupting local electrostatic balance and thereby reducing core stability. Other charge-altering substitutions (i.e., E75K or T210K) at the trimeric interface may act similarly. Although not addressed here, prior work has shown that residues at the trimeric interface, including E75 and E213, can interact with the RRR motif of Nup153⁵³, suggesting that this region may also contribute to nuclear import. Critically, the LEN addition in producing cells increased core yield for the triple mutant. Thus, a straightforward explanation for E213K selection is that E213 offsets LEN’s stabilizing effects; in the absence of LEN, the triple mutant fails to undergo proper capsid maturation or exhibits reduced stability, rendering it noninfectious.

While resistant variants carrying the decisive E213K substitution emerged only under high concentrations of LEN not used in clinical settings, E213K does exist as a naturally occurring polymorphism⁷, albeit at a very low frequency (0.08% among 12,261 HIV-1/SIVcpz sequences in the Los Alamos HIV-1 sequence database). Given the strong therapeutic potential of LEN and the possible further optimization of capsid-targeting antivirals, the emergence of such drug-dependent variants—while not expected in the immediate future—may eventually occur and raises important clinical questions. A logical strategy to eliminate drug-dependent variants is treatment interruption—in the absence of the drug, the virus cannot propagate. However, this is complicated by the very beneficial long-acting nature of current LEN formulations. In terms of transmission, a LEN-dependent virus is at a clear disadvantage when encountering a new, untreated host. However, broad clinical deployment of capsid inhibitors could create a permissive environment for rare breakthrough infections and onward spread. Such complex scenarios raised by drug-dependent variants underscore the importance of continuous monitoring for emerging drug-resistant variants and their phenotypic characteristics.

Finally, our mutagenesis work solely relied on the LAI CA backbone. Hence, the mutational tolerance we observed may not be universal across other strains or subtypes. Recent work, however, has shown that the pattern of early-emerging LEN resistance mutations is highly similar across different subtypes⁵⁴. In contrast, our in vitro adaptation using a chimeric virus harboring the capsid from a T/F strain of HIV-1 (CH505) to which we added the Q67H substitution too displayed a distinct evolutionary trajectory from that observed in our LAI-based adaptation, which recurrently selected for the strong M66I substitution. Thus, while the mechanistic basis for these differential evolutionary pathways remains to be defined, these results highlight the importance of capsid sequence context in LEN resistance evolution.

In conclusion, this study highlights the adaptability of HIV-1 to develop high-level LEN resistance. Insights into evolutionary trajectories, novel resistance mechanisms, and diverse phenotypic outcomes have practical implications for the use and further development of HIV-1 capsid inhibitors. Our findings underscore the central role of M66I, previously recognized in resistance evolution^30,55, and reveal the added contribution of CTD mutations in modulating LEN sensitivity. Rational targeting of these residues—whether through direct engagement or strategies that amplify subtle changes in capsid stability—may guide the continued refinement of this already powerful and promising class of HIV-1 capsid-targeting antivirals^24,56,57,58 as well as expand to a new target in the capsid⁵⁹.

Method

All research was conducted in accordance with relevant ethical regulations. This study did not involve human subjects research or clinical investigation. Peripheral blood mononuclear cells (PBMCs) used in this study were obtained from leukocytes collected from anonymous donors through the New York Blood Center. These materials were fully de-identified, and informed consent was not applicable for their use in this study.

Chemicals

Lenacapavir (HY-111964) was obtained from MedChemExpress. Aphidicolin (5047440001) was purchased from Sigma. Plasmocin® Prophylactic (ant-mpp) was acquired from InvivoGen.

HIV-1 CA mutant library construction

We used pBru3ori-GFP3, a replication-competent molecular clone of the LAI strain³². This reporter virus expresses GFP in place of the nef gene, while all other viral genes remain intact. To facilitate mutant library construction, we generated an intermediate plasmid in which a large portion of the gag-coding sequence was deleted. This design minimizes contamination from undigested vector, which could produce the WT virus and confound our analysis of mutant variants that are expected to be less fit than the WT virus.

For PCR-mediated mutagenesis, we designed pairs of 27-bp mutagenized oligonucleotides (sense and antisense) with the mutated codon (three nucleotides) positioned centrally. Each codon position was completely randomized (NNN). Oligonucleotide primers used for site-directed mutagenesis and other PCR-based applications were synthesized by Eurofins Genomics or Integrated DNA Technologies (IDT). Oligonucleotide sequences used in this study are listed in Supplementary Table 1.

PCR-mediated mutagenesis was performed using NarI or BssHII at the 5′ end and ApaI at the 3′ end. Oligonucleotide primers annealing 15–25 bp upstream or downstream of these sites were used. Separate PCR reactions amplified the 5′ and 3′ fragments using mutagenized oligonucleotides. Purified PCR products were mixed in a Gibson assembly reaction with vectors digested by NarI or BssHII and ApaI. In some cases, a second PCR step was performed to combine the two PCR fragments before cloning.

The assembled products were used for chemical transformation. We observed >300 colonies, whereas the control transformation with the digested vector alone yielded ~60 colonies. Each mutant library, where a single amino acid residue was randomized, had a theoretical size of 64. To achieve 99% probability of full coverage, 338 transformants were required⁶⁰. Plasmid DNA for each mutant library was prepared by scraping colonies from bacterial plates. Each plasmid DNA library was then used for transfection.

Generation of mutant clones

Individual proviral clones carrying single or multiple mutations were generated by either PCR amplification of Gag-encoding sequences from viral DNA of selected or adapted variants, or by PCR-mediated mutagenesis. PCR amplicons containing mutations were cloned into replication-competent or Env-defective infectious clones of GFP reporter viruses. A chimeric LAI-CH505 construct was generated by replacing the Gag-coding region between BssHII and ApaI with the corresponding sequence from the CH505 transmitted/founder virus (ARP-13550, BEI)⁶¹ using Gibson assembly. Site-directed mutagenesis was then used to introduce Q67H into the CH505 Gag backbone. Introduction of mutations was confirmed by Sanger sequencing (GENEWIZ) or by whole plasmid sequencing (GENEWIZ or Plasmidsaurus). Newly generated plasmids in this study are available from the corresponding author upon reasonable request.

Cell culture

Adherent cells, such as Lenti-X 293T cells (632180, Takara) and TZM-bl (HRP-8129, BEI)⁶², were maintained in Dulbecco’s modified Eagle’s medium (DMEM; Corning) supplemented with 10% fetal bovine serum (FBS; Sigma) and 1× penicillin-streptomycin (P/S; Corning). MT-4 cells (ARP-120, BEI) were cultured in RPMI 1640 medium (Corning) supplemented with 10% FBS, 1× P/S, and 2 mM L-glutamine (Corning). In certain experiments, Plasmocin® prophylactic (ant-mpp) was added to the culture medium. Primary CD4⁺ T cells were isolated from PBMCs derived from leukocytes collected from anonymous donors through the New York Blood Center, using the EasySep™ Human CD4⁺ T Cell Enrichment Kit (STEMCELL Technologies) and activated using Dynabeads™ Human T-Activator CD3/CD28 (Thermo Fisher Scientific).

Viruses

Stocks of replication-competent GFP reporter viruses were generated by transient transfection of HEK293T cells using polyethylenimine (PEI; Polysciences). For mutant libraries, plasmid DNA libraries were co-transfected with the Gag-Pol expression vector pCRV1-Gag-Pol⁶³ to supply WT Gag-Pol during the initial round of virus production, minimizing potential dominant-negative effects of CA mutations. In some cases, a VSV-G expression vector was also included to enhance the first round of infection. For the generation of pseudoviruses from Env-defective infectious clones, a VSV-G expression vector⁶⁴ was included.

High-titer virus stocks were generated by virus concentration. Briefly, 293T or Lenti-X 293T cells were seeded in 100-mm plates at 5 × 10⁶ cells per plate. Culture medium was either left unchanged or replaced with fresh medium one day after transfection. Supernatants were harvested 2 days post-transfection, filtered using a Steriflip filter unit (Millipore), and gently layered onto 5 mL of 20% (w/v) sucrose solution in polyallomer tubes (Beckman Coulter). Ultracentrifugation was performed at 100,000 × g for 2 h at 4 °C using an SW28 rotor. Virus pellets were incubated with 3.2 mL of culture medium for at least 1 h at 4 °C before resuspension. Virus aliquots were prepared and stored at −80 °C until use.

Multiplex variant selection

MT-4 cells were plated at 3 × 10⁶ cells per well in 12- or 24-well plates and infected with varying amounts of virus. In some experiments, spinoculation was used to enhance infection. GFP-positive cells were assessed 2- or 3-days post-infection by flow cytometry. Virus-infected cells were then diluted with uninfected target cells to achieve 1–3% GFP-positive cells, in the presence or absence of LEN at various concentrations. The virus was allowed to propagate for 3 days, and viral growth was quantified by flow cytometry. Infected cells were either harvested immediately or expanded for several additional days before harvesting for sequencing analysis.

Viral adaptation to lenacapavir

Viral adaptation was performed using a dose-escalation scheme. Briefly, MT-4 cells were infected with replication-competent viruses carrying different mutations at position 67 of CA. Two main viral adaptation experiments were conducted: one using a single Q67N mutant virus, and the other using a pooled mixture of four mutants (Q67F/H/W/Y). For the pooled adaptation, each mutant virus was initially infected separately, and similar levels of each variant were mixed 2–3 days post-infection. Viral adaptation was carried out to maintain approximately 5% virus-infected cells in the presence of LEN. The initial LEN concentration was 10.9 nM for Q67N adaptation and 12.5 nM for the pooled Q67F/H/W/Y adaptation. The proportion of virus-infected cells was monitored every 2–3 days using flow cytometry. Once an increase in GFP-positive cells was observed, infected cells were diluted with uninfected cells to maintain ~5% infection and exposed to increasing concentrations of LEN.

For the Q67N adaptation, a breakthrough was observed after 15 days when the proportion of infected cells exceeded 50%. Culture supernatant was harvested, and new infections were initiated. This process was repeated until the LEN concentration reached 2.8 µM, at which point signs of cytotoxicity were noted. A similar dose-escalation approach was applied to the pooled Q67F/H/W/Y viruses. At each cell split, the supernatant and a portion of the cells were collected and stored at −80 °C.

NGS-based variant analysis

For the multi-variant selection assay, PCR primers targeting regions containing mutagenized positions (positions 4, 18, 67, and 74) were used for amplification. A second PCR was performed to incorporate sample barcodes and Illumina sequencing adapters for multiplexing. PCR products were gel-purified using a DNA extraction kit, and DNA concentrations were measured using a Qubit fluorometer. The resulting libraries were submitted to GENEWIZ for Illumina sequencing or AmpSeq using the Element AVITI sequencing platform, which yields estimated per-base error probabilities of <0.1% per nucleotide. NGS data were first demultiplexed, and variants at each position were extracted using cutadapt (https://doi.org/10.14806/ej.17.1.200). The frequency of each amino acid was then calculated using a custom R script. HIV-1 and SIVcpz sequences (n = 12,261) were obtained from the Los Alamos HIV database.

For CA sequence analysis of adapted variants, PCR-amplified products covering the full coding sequence of HIV-1 CA were submitted for Oxford Nanopore sequencing at Plasmidsaurus, which is characterized by raw single-read per-base error probabilities of approximately 1–5% per nucleotide. Sequencing reads were mapped to the reference sequence (the LAI strain) using Hisat2⁶⁵ and inspected using IGV software⁶⁶. The frequency of each mutation was extracted from BAM files using a custom shell script. Mutation calls were verified by manual inspection in IGV. A threshold of 5% was arbitrarily used to identify de novo mutations. For further analysis, we included mutations exhibiting a significant frequency increase. Amino acid substitutions were visualized on the CA hexamer structure (PDB: 3J34 and PDB: 6VKV) using UCSF ChimeraX⁶⁷.

For the linkage analysis, cutadapt was used to remove sequences shorter than 50 bp and those with a Phred quality score lower than 30. Burrows–Wheeler Aligner⁶⁸ was used for alignment, and Samtools⁶⁹ for indexing. Single-nucleotide variants (SNVs) were called using LoFreq⁷⁰, and bcftools⁶⁹ filtered out SNVs with a frequency less than 1% and those with Phred-scaled quality values less than 500. A Python script incorporating pysam⁶⁹ and vcfpy (DOI: 10.21105/joss.00085) was written in-house and used to parse through the selected reads for the selected SNVs. Reads that did not contain information for one or more of the SNVs were removed from the data set. The output contained groups of mutations and their number of occurrences in the remaining reads. This was manually manipulated in Microsoft Excel to remove any group of SNVs with a frequency of less than 1%. These low-frequency groups of SNVs for each time point were moved into the “other” category for that time point. Once all linkage information was determined, any group of SNVs which never rose above 5% at any time point was also moved to the “other” category for its respective time point.

Isolation and disassembly of HIV-1 cores

HIV-1 cores were obtained by ultracentrifuging intact virus particles through a detergent layer onto a 30% sucrose cushion⁴⁸. Specifically, 250 µL of 10-fold concentrated viral stocks were subjected to ultracentrifugation at 4 °C for 2 h using a Beckman SW50.1 rotor at 127,000 × g. The detergent layer contained 0.03% IGEPAL CO-630 in a Tris-based buffer (100 mM NaCl, 10 mM Tris-HCl, pH 7.4, 1 mM EDTA; STE buffer). The isolated cores were then resuspended in the STE buffer for subsequent analysis.

To examine core disassembly, 50 µL of purified cores were incubated with or without 100 µM IP6 for 2 h at either 4 °C or 37 °C. After incubation, core samples underwent microcentrifugation at 20,817 × g for 10 min at 4 °C. The resulting pellets were resuspended in 50 µL of RPMI 1640 medium supplemented with 10% FBS. A quantitative reverse transcriptase (RT) assay described below was then performed on both isolated viral cores and the pellet samples after incubation to evaluate core yield and disassembly. To determine the effects of LEN on viral cores, 50 µL of purified cores were incubated with LEN at final concentrations of 0.004, 0.02, 0.1, and 0.5 µM at 4 °C for 1 h. After the 1-h incubation, the core samples were subjected to a similar microcentrifugation process to separate the pelletable cores.

Quantitative reverse transcriptase assay

To quantify the amount of reverse transcriptase (RT), a product-enhanced RT assay utilizing quantitative PCR (qPCR) was performed⁷¹. In this assay, 5 µL of input materials (such as virus stocks or isolated cores) was mixed with 5 µL of 2× lysis buffer (40% glycerol, 0.25% Triton X-100, 100 mM Tris-HCl pH 7.5, and 50 mM KCl) in a 96-well round-bottom plate. The mixture was incubated at room temperature for 10 min to allow lysis. A pre-determined RT standard curve was generated using a virus stock with a known level of RT. To stop the lysis process, 90 µL of distilled deionized water was added.

The reaction mixture consisted of 10 µL of lysed supernatant and 10 µL of 2× reaction buffer, which contained 10 mM MgCl₂, 1× core buffer (20 mM Tris-HCl pH 8.3, 20 mM KCl, and 5 mM (NH₄)₂SO₄), 4 µg/mL BSA, 0.4 mM dNTPs, 1 µM forward primer (5’-TCCTGCTCAACTTCCTGTCGAG), 1 µM reverse primer (5′-CACAGGTCAAACCTCCTAGGAATG), 0.04 µg MS2 RNA (Sigma), 1× SYBR™ Green I Nucleic Acid Gel Stain (Themo Fisher Scientific), 0.02 µL Rox reference dye (Themo Fisher Scientific), and 0.4 U RiboLock RNase Inhibitor. Before initiating the reaction, 2 units of Taq polymerase were added.

Reverse transcription of MS2 RNA was followed by qPCR using a real-time PCR system. For assays performed in a 384-well plate format, half of the reaction volume was used. The thermal cycling conditions were as follows: reverse transcription at 42 °C for 20 min, initial denaturation at 95 °C for 2 min, followed by 40 cycles of amplification with denaturation at 95 °C for 5 s, annealing at 60 °C for 5 s, extension at 72 °C for 15 s, and a final step at 80 °C for 25 s.

Western blotting

Virus particle (2 µl) and core samples (18 µl) were lysed with 4X Bolt™ LDS Sample Buffer (Invitrogen) and resolved on NuPAGE Bis-Tris 4–12% polyacrylamide gels (Invitrogen). Electrophoresis was carried out at 100 V for 60 min in a mini-gel tank. Following electrophoresis, proteins were transferred onto a nitrocellulose membrane using a wet transfer system with 1× Bolt Transfer Buffer (containing 10% methanol). Membranes were blocked overnight at 4 °C. After blocking, membranes were washed three times for 5 min each in 1× TBS-T (Tris-buffered saline with 0.1% Tween-20), then incubated with mouse monoclonal anti-HIV-1 p24 antibody (clone 183-H12-5C, ARP-3537, BEI) at a 1:1000 dilution for 2 h at room temperature with gentle agitation in a shaker. After three additional washes in 1× TBS-T, the membranes were incubated with an IRDye 488-conjugated anti-mouse secondary antibody (LI-COR Biotechnology) diluted 1:10,000 in blocking buffer for 1 h at room temperature. Final three washes (3 times, 5 min each) were performed in 1x TBST. Blots were imaged using a Licor Odyssey system, and p24 signal intensities were quantified. Uncropped and unprocessed scans of the blots are provided in the Source data file.

Software and data analysis

Data were collected using commercially available software, including QuantStudio Design and Analysis Software, NovoExpress, LI-COR Image Studio, and Molecular Devices SoftMax Pro. Data analysis was performed using commercially available and open-source software together with custom scripts. Software used included GraphPad Prism, FlowJo, bcftools, Biopython, Biostrings, bwa, cutadapt, LoFreq_Star, pandas, pysam, samtools, stringr, and vcfpy.

Statistical analysis

Statistical analyses were conducted using Student’s t test or ANOVA, implemented in Prism software (GraphPad). Statistical significance was defined as p < 0.05. ****p < 0.0001; ***p < 0.001; **p < 0.01; *p < 0.05.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.