Systematic literature searches were conducted in PubMed, Medstrade Foreign Language Retrieval Platform, China National Knowledge Infrastructure (CNKI), Wanfang Data, and Chinese Core Journals Database (up to 31 December 2024) using Boolean operators: (“HIV-1” OR “AIDS”) AND (“China” OR “Chinese”) AND (“subtype” OR “genotype”). Articles in English/Chinese were screened for nationwide HIV-1 subtype prevalence data. Single-province studies were excluded to minimize regional bias. Reference lists of systematic reviews were cross-checked to identify additional studies.

A cohort of 472 HIV-infected individuals diagnosed at Shenzhen Third People’s Hospital (2023-2024) was enrolled. Whole blood samples (10 mL) were processed using a high-purity viral RNA extraction kit (plasma-derived). Nested RT-PCR targeting a 1,062-bp pol gene fragment (HXB2 coordinates 2253-3314) was performed with One-Step RNA PCR reagents, ExTaq polymerase, and subtype-specific primers. Amplicons were sequenced by Dongguan HyLabs virology laboratory. Sequence quality was validated using the LANL HIV Sequence Quality Control Tool (https://www.hiv.lanl.gov/content/sequence/QC/index.html, accessed on 1 July 2023–20 August 2025). Recombinant strains were identified via RIP and jpHMM algorithms in the HIV Database. Phylogenetic analysis (MEGA 11, Kimura 2-parameter model, 1,000 bootstrap replicates) confirmed subtypes. Drug resistance profiles were determined through Stanford HIV Drug Resistance Database submissions.

HEK293F human embryonic kidney cells (obtained from ATCC, CBP-60437) were maintained at 37°C under 5% CO₂ in SMM 293-TII Expression Medium (Serum free, complete medium, M293TII, Sino Biological), supplemented with 1-5% SMS 293-SUPI Expression Medium Supplement (M293-SUPI, Sino Biological). TZM-bl cells (a HeLa-derived indicator cell line expressing β-galactosidase and luciferase reporter genes under the control of the HIV promoter) and 293T cells (kindly provided by Prof. Lin Cheng) were cultured in medium supplemented with: 10% fetal bovine serum (FBS; A5669701, GIBCO), 1% penicillin-streptomycin solution (10,000 U/mL; 15140122, GIBCO) and maintained in a humidified incubator at 37°C with 5% CO₂.

HIV-1 gp140 sequences (subtypes B, CRF01_AE, CRF07_BC, CRF08_BC, CRF55_01B; Los Alamos HIV Database) were cloned into pcDNA3.1 (V79020, Thermo Fisher) with C-terminal FLAG tags. Human CD4 (UniProt P01730) was cloned with a His₆ tag. Constructs were transformed into E. coli TOP10 (C404010, Thermo Fisher), and plasmid integrity was confirmed via Sanger sequencing (Genewiz).

HEK293F cells (2×10⁶ cells/mL) were transfected with 1 μg/mL plasmid using EZ Trans (AC04L092, Life-iLab; DNA:reagent=1:5). Culture supernatants harvested at 120 hr were filtered (0.45 μm), loaded onto Anti-FLAG M2 Affinity Gel (A2220, Sigma-Aldrich), and eluted with 3×FLAG peptide (150 μg/mL, F3165, Sigma-Aldrich). Purified gp140 was concentrated (30 kDa Amicon Ultra, UFC9030, Merck), subjected to size-exclusion chromatography (Superose 6 Increase 10/300 GL, 29091559, Cytiva), and trimer fractions (420–450 kDa) were collected. Purity (>95%) was confirmed by SDS-PAGE (PG610, Epizyme). The PageRuler™ Unstained Protein Ladder (26614, Thermo Scientific) was employed as a molecular weight standard for protein size estimation during SDS-PAGE.

The HIV attachment inhibitor temsavir (BMS-626529) was purchased from MedChemExpress (HY-15440). The compound was dissolved in dimethyl sulfoxide (DMSO, 276855, Sigma-Aldrich) at a stock concentration of 10 mM, aliquoted, and stored at −80°C until further use. The screened compounds were purchased as prepared solutions from Shanghai Topscience. The specific information is shown in Supplementary Table 1.

CD4 (10 μg/mL) was immobilized on Ni-NTA biosensors (160016, GatorBio) for 60 s. Association/dissociation kinetics between gp140 (6.25–200 nM) and CD4 were measured at 25°C using a GatorPrime system. For inhibition assays, gp140 (250 nM) was pre-incubated with temsavir (50 μM, HY-15440, MedChemExpress)or candidate compounds (Supplementary Table S2). Data were analyzed using Gator One v2.3 (1:1 binding model, global fitting). Inhibition rates were calculated as: 1–[Response(gp140+inhibitor)/Response(gp140)].

HIV pseudovirions were generated by transfecting 293T cells (1×10⁶ cells/well) with 4 μg of full-length infectious HIV plasmid (pNL4–3 luc R-E- plasmid + Env plasmid) (kindly provided by Lin Cheng) using Lipofectamine 2000 transfection reagent (11668030, Thermo Fisher Scientific) in OPTI-MEM reduced-serum medium (31985070, Gibco). Transfected cells were cultured at 37°C for 48 hours. Virus-containing supernatants were filtered through 0.45-μm filters to remove cellular debris and stored at -80°C until use.

For infection, 500 μL of pseudovirus (MOI = 0.25~1) was co-incubated with small-molecule drugs at room temperature for 2 hours, then added to TZM-bl cells (5×10⁵ cells/well) for infection and drug inhibition. After 4 hours, unbound viruses were removed by washing cells five times with phosphate Buffered Saline (PBS) buffer, followed by the addition of fresh medium. Supernatants were harvested for ELISA after 48 hours, and cellular RNA was extracted for HIV nucleic acid detection.

Cell viability was evaluated using the Cell Counting Kit-8 (CCK-8; 100-120, Goonie) according to the manufacturer’s instructions. Briefly, cells were seeded in 96-well plates at a density of 5×10³ cells per well and incubated overnight to allow attachment. Following the respective treatments, 10 μL of CCK-8 reagent was added to each well, and the plates were incubated at 37°C for 2 hours. The absorbance was measured at 450 nm using a microplate reader (BioTek, USA). All experiments were performed in triplicate and repeated at least three times independently. Cell viability was calculated as a percentage relative to the untreated control group.

Using the HIV-1 Gag p24 ELISA Kit (GE0001, Service), prepare standards and reagents according to the manufacturer’s instructions prior to the experiment. Add 100 μL of sample (standard or test sample) and 50 μL of detection antibody to each well sequentially, completing all additions within 15 minutes. Mix well and incubate with shaking for 2 hours at room temperature. Wash the plate 5 times and pat dry, then add 100 μL of detection antibody working solution to all wells and incubate with shaking for 1 hour at room temperature. Wash again 5 times and pat dry, add 90 μL of TMB substrate to each well and incubate for 10–30 minutes at room temperature in the dark. Finally, add 50 μL of stop solution to all wells. Measure the absorbance within 10 minutes using dual wavelengths of 450 nm and 630 nm. The calibrated OD value is obtained by subtracting the 630 nm measurement from the 450 nm measurement.

Total RNA was isolated from cells using TRIzol Reagent (15596026CN, Thermo Fisher Scientific) according to the manufacturer’s protocol. cDNA was synthesized with the PrimeScript™ RT Reagent Kit (Perfect Real Time) (RR037A, Takara Bio) following the manufacturer’s instructions. HIV-1 RNA levels were quantified by real-time PCR targeting the Nef gene, with the following primers: Forward: 5’-GGTGGGTTTTCCAGTCACAC-3’;​Reverse: 5’-GGGAGTGAATTAGCCCTTCC-3’. PCR amplification was performed using TB Green^® Premix Ex Taq™ II (Tli RNaseH Plus) (RR820A, Takara Bio) on a QuantStudio 7 Flex Real-Time PCR System (Viia 7DX, Applied Biosystems). All HIV-1 copy numbers were normalized against endogenous TBP mRNA levels in the same samples, amplified with primers: Forward: 5’-CACGGCACTGATTTTCAGTTCT-3’; Reverse: 5’-TTCTTGCTGCCAGTCTGGACT-3’. Expression alterations were quantified via the 2^−ΔΔCT method, with comparisons referenced to negative controls.

The HIV sequences were derived from Chinese samples in the Los Alamos HIV Sequence Database (http://www.hiv.lanl.gov; accessed on 1 July 2023–20 August 2025). Clustraw (https://www.genome.jp/tools-bin/clustalw; accessed on 14 July 2023–20 March 2024) was used for sequence comparison, Esprint 3.0 (https://espript.ibcp.fr/ESPript/ESPript/index.php; accessed on 14 July 2023–20 March 2024) was used for sequence alignment visualization and secondary structure prediction, and Jalview software to export different subtype consensus sequence colors for automatic generation. The Logo plots for HIV were made using the Analyze Align tool at the HIV database and are based on the WebLogo 3 program (https://weblogo.threeplusone.com; accessed on 1 October-28 December 2023) and the HIV-1 database global curated and filtered 2021 alignment published, including 5 HIV-1 gp120 protein sequence per person from 3, 878 individuals in China. The relative height of each letter within individual stack represents the frequency of the indicated amino acid at that position. The numbering of all the Env amino acid sequences is based on the prototypic strain of HIV-1. Use Jaview (https://www.jalview.org; accessed on 11 October-28 December 2023) to compare sequence edits and conduct percentage statistics.

All molecular dynamics (MD) simulations were performed using the GROMACS 202X software package. The following force fields were employed: AMBER99SB-ILDN for proteins, GAFF for ligands, and TIP3P for water molecules. The system was solvated in a cubic water box with a minimum distance of 1.2 nm between the protein surface and box edges. Simulation parameters included: PME method for long-range electrostatic interactions, and a 1.0 nm cutoff for van der Waals interactions. The protocol consisted of energy minimization, followed by 100 ps NVT ensemble equilibration, and finally 100 ns production simulation under NPT ensemble. Temperature (300 K) and pressure (1 bar) were maintained using V-rescale and Berendsen coupling methods, respectively. Trajectory analysis included: structural stability assessment via Cα-RMSD, and energetic evaluation through MM/PBSA binding free energy calculations (performed every 10 ns) with residue-wise decomposition.

The three-dimensional structure of the HIV-1 gp120 monomer was predicted using AlphaFold3 (https://alphafold.com; accessed from 25 June 2024 to 23 August 2024) with the following protocol: The gp120 amino acid sequence (signal peptide removed) obtained from the Los Alamos HIV Sequence Database (http://www.hiv.lanl.gov) served as input; five models were generated using default parameters without template mode, from which the model with the highest confidence (pLDDT > 90) was selected as the initial structure. Subsequent post-processing included energy minimization using the AMBER99SB force field (5000 steps of steepest descent algorithm) and addition of missing hydrogen atoms. The predicted model was then aligned with the experimental structure PDB: 3J70 (HIV-1 subtype B gp120-CD4 complex), with Cα-RMSD analysis confirming the accuracy of core structural regions. Finally, all models underwent stereochemical validation through Ramachandran Plot analysis (https://saves.mbi.ucla.edu/, accessed from 25 June 2024 to 20 December 2024), which demonstrated that >95% of residues occupied allowed regions, ensuring structural reliability.

CD4 (PDB: 1CDJ) was retrieved from the RCSB Protein Data Bank. GRAMM-X protein-protein docking server (https://gramm.compbio.ku.edu/; accessed on 29 August-22 October 2024) predicted gp120-CD4 complexes using a fast Fourier transform correlation algorithm that evaluates surface complementarity and electrostatic potential matching. Docking parameters: Receptor (CD4) retained crystallographic coordinates; Ligand (gp120) was stripped of water molecules. Grid resolution: 1.0 Å; angular sampling density: 15°; electrostatic weighting: 0.5 (balancing hydrophobic/electrostatic interactions); collision threshold: 50. The top-scoring complex underwent 100 ns MD simulation (GROMACS 2020.6, AMBER99SB-ILDN force field), with the final 20 ns equilibrium trajectory used for structural analysis.

Temsavir 3D coordinates were acquired from DrugBank. Protein preparation (water removal, polar hydrogen addition, charge assignment) and flexible residue/rotatable bond definition utilized AutoDock Tools 1.5.6. Semi-flexible docking was executed in AutoDock Vina 1.2.3 (exhaustiveness=25, num_modes=10) within a 30×30×30 ų grid box. The optimal pose was selected based on hydrogen bonding and binding free energy. Redocking validation yielded RMSD < 2 Å versus the crystallographic pose.

Protein-protein interfaces were characterized using PDBePISA (http://www.ebi.ac.uk/pdbe/prot_int/pistart.html; accessed on 5 October-30 December 2024) for buried surface area and binding energy quantification. PyMOL (https://PyMOLwiki.org/index.php/InterfaceResidues; accessed on 25 November 2024–23 March 2025) visualized interaction networks, with hydrogen bonds and steric clashes systematically compared across subtypes.

From GRAMM-X’s top 100 outputs: Hierarchical clustering (PyCluster, RMSD cutoff=5.0 Å) revealed consistent binding modes among the top 10 conformations. The centroid of the largest cluster (ranked #1) was selected as the representative structure. Interface properties (optimal ΔG, contact area) were evaluated via PDBePISA, with docking reliability confirmed by comparison to the known complex (PDB:3J70). This dominant conformation served as the MD starting structure.

Homology models of gp140 proteins from major HIV-1 subtypes (CRF01_AE, CRF07_BC, CRF08_BC, CRF55_01B, and B) were constructed using SWISS-MODEL, with >96% residues in favored regions of Ramachandran plots. Ligand-binding sites were predicted using MOE’s Site Finder. Protein structures were energy-minimized in Schrödinger Suite (Protein Preparation Wizard) with bond optimization, hydrogen addition, and disulfide/hydrogen bond assignment. MOE and Schrödinger Suite were used under academic licenses from the School of Pharmacy, Fudan University. License configurations follow institutional policies. A chemical library (13,819 compounds) was prepared using LigPrep, generating tautomers, ionization states, and up to 32 conformers per molecule. Molecular docking was performed via Glide’s standard-precision mode, retaining top 50% scoring compounds. Candidates showing≥1.5-order affinity differences across subtypes were prioritized. Protein-ligand interaction fingerprints (PLIF) identified key binding residues for CRF55_01B. Structurally similar compounds were clustered, retaining the highest-affinity representative per cluster.

Statistics were analyzed using GraphPad Prism version 6.01 (GraphPad, San Diego, CA, USA). All data are presented as mean ± SEM. Comparisons between two groups were performed using Student’s t-test. The statistical significance of differences between groups was determined by two-way ANOVA. A p-value below 0.05 was regarded as significant. Experiments were performed in triplicate with biological replicates of n = 3 per condition.

To determine the differences in the adhesion processes of different HIV-1 subtypes, search relevant literature on the distribution of HIV-1 in China (981 Chinese and 205 English literatures). To avoid regional differences, we selected only 19 literatures that statistically covered the national distribution. By counting the top 5 HIV-1 subtypes in these literatures, we found that subtypes B (100%, 19/19), CRF01_AE (94.7%, 18/19), CRF07_BC (89.5%, 17/19), CRF08_BC (68.4%, 13/19), and CRF55_01B (42.1%, 8/19) were most frequently mentioned in the Chinese region (Figure 1A), and had the highest prevalence rates, which were 11.9%, 38%, 32.2%, 9%, and 3.7% respectively (Figure 1B). Clinical validation using 2024 subtyping data from Shenzhen Third People’s Hospital (n=472) demonstrated concordance: CRF07_BC (47.25%, 223/472), CRF01_AE (25%, 118/472), CRF55_01B (12.07%, 57/472), CRF08_BC (5.93%, 28/472), and B (3.6%, 17/472) (Figure 1C).

Research shows that fully glycosylated secreted gp140 expressed in mammalian cells can better mimic the envelope protein structure of native HIV conformation. Thus, we purified gp140 (Flag-tagged) from 5 HIV subtypes (B, CRF01_AE, CRF07_BC, CRF08_BC, and CRF55_01B) and human CD4 protein (His-tagged) (Supplementary Figures 1, 2). Bio-layer interferometry (BLI) was used to detect the binding affinity between gp140 at different concentrations and CD4. From the sensorgram, during the binding phase, as the concentration of the ligand gp140 increased, the signal rose rapidly in a concentration-dependent manner. The signal reached a relatively stable state after approximately 200 seconds. Data analysis yielded the equilibrium dissociation constants (K_D) of gp140 proteins from 5 HIV-1 subtypes and the CD4 protein. HIV-1 subtype B displayed the lowest K_D value (79 pM), indicating the strongest binding affinity. K_D values increased for CRF07_BC (559 pM) and CRF08_BC (931 pM), while CRF01_AE (4.61 nM) and CRF55_01B (8.76 nM) showed considerably higher K_D values, representing weaker binding (Figure 2).

To investigate the mechanisms underlying the differences in the adhesion processes among different HIV subtypes, we conducted a comparative analysis of the gp120 sequences of the five selected HIV-1 subtypes and found numerous differences (Supplementary Figure 3). Subsequently, we use Alphafold2 to predict the structures of the gp120 proteins of the five HIV subtypes. Then, these structures were compared with the known HIV-1 B subtype structure in the literature and analyzed in terms of charge distribution, buried area, binding free energy, and hydrogen bond distribution.

The analysis of charge distribution indicated that the binding interface between the CD4 and gp120 proteins was mostly in a region of positive charge distribution, while the corresponding site of gp120 was mostly negatively charged, facilitating their binding. However, no significant differences were observed among different subtypes (Supplementary Figure 4). By docking the gp120 proteins of different HIV-1 subtypes with the CD4 protein and analyzing the buried area at the binding site and the binding free energy, we found that subtype B, which had the strongest binding ability, had the largest buried area (1153.6 Ų). The binding free energy results showed that the binding free energy between the gp120 of subtype B and the CD4 protein was -5.6 kcal/mol, indicating a certain degree of binding stability. The buried areas and binding free energies of CRF07_BC and CRF08_BC subtypes were the next, while those of CRF01_AE and CRF55_01B had smaller buried areas and the lowest absolute values of binding free energy (Supplementary Figure 5, Table 1).

Next, we analyzed the hydrogen bonds at the binding interface between gp120 of different HIV-1 subtypes and CD4. We found that there is a more extensive hydrogen-bond network between the gp120 of subtype B and the CD4 protein (Supplementary Table 2). Moreover, we discovered that in the gp120 of subtypes B, CRF07_BC, and CRF08_BC, hydrogen bonds are formed between the backbone amide of Ser335 and the side chain of Asn52 of CD4, as well as between the hydroxyl group of Ser335 and the carbonyl oxygen of Lys46. In contrast, CRF01_AE and CRF55_01B, which exhibited poor binding abilities, failed to form hydrogen bonds at this position (Figure 3). Through analysis, we believe that the differences in the adhesion processes of different HIV subtypes may be attributed to sequence mutations in the gp120 proteins of different HIV-1 subtypes during evolution. These mutations lead to changes in the buried area at the binding interface with the CD4 protein and the hydrogen bonds at the interaction interface, consequently resulting in differences in binding free energy.

Subsequently, we used BLI to detect the inhibitory effect of Temsavir, the active precursor form of fostemsavir, the only FDA-approved HIV adhesion inhibitor, on the adhesion processes of different HIV subtypes. The results showed significant differences in the inhibitory effect of Temsavir on the adhesion processes of different HIV subtypes (Figures 4A–E). The highest inhibition rate was observed for subtype B (35.7%), followed by CRF07_BC (9%) and CRF08_BC (9%), while the inhibitory effects on CRF01_AE (1.27%) and CRF55_01B (0.96%) were the poorest (Figures 4F).

Through small-molecule docking studies of gp120 and Temsavir, we discovered that CRF01_AE and CRF55_01B failed to successfully dock with Temsavir (Supplementary Figure 6). To further explore the underlying reasons, we compared the sequence variations at relevant sites involved in Temsavir’s function (Supplementary Figure 7A) and conducted an in-depth analysis of the variation at position 375 in 3,878 sequences from China in the HIV database. The findings revealed that in the gp120 of CRF01_AE and CRF55_01B subtypes, the serine at position 375 mutated to histidine. This mutation generated substantial steric hindrance, which impeded the binding to Temsavir (Supplementary Figures 7B–F). This change was highly likely the key factor leading to the failure of CRF01_AE and CRF55_01B subtypes to dock with Temsavir successfully and, consequently, the lack of a significant inhibitory effect (Figure 5).

Next, we carried out a detailed analysis of the residues within a 4 Å distance between gp120 and Temsavir in five HIV subtypes. The results showed that, from the perspective of the overall spatial conformation, the distance between residues in subtype B was relatively large. Its binding pocket space was more expansive compared to other subtypes, and it was also farther from the spatial position of the small-molecule inhibitor. This unique spatial structure might be more conducive to the entry and binding of Temsavir. When comparing the residue mutations among subtypes, we found that in CRF07_BC and CRF08_BC subtypes, a mutation from lysine (K) to arginine (R) occurred at position 432. In CRF01_AE, mutations were present at sites T202K, S375H, K432Q, and M475I, while in CRF55_01B, mutations occurred at sites D113E, T202K, S375H, I424V, K432Q, and M475I. These mutations were likely important factors contributing to the differences in inhibitory effects among subtypes (Figure 6).

Given the limited efficacy of FDA-approved HIV adhesion inhibitors against common CRF subtypes in China, we performed molecular docking using Schrödinger software between 13,819 compounds and the gp140 protein of the CRF55_01B subtype (Standard Precision mode). After energy optimization and retaining the top 50% scoring molecules in each iteration, the compounds were subsequently docked against gp140 proteins from four other subtypes. Compounds exhibiting binding energy differences exceeding 1.5 orders of magnitude were selected (Supplementary Figures 8, 9). From these, 453 compounds with binding affinities below -6 kcal/mol were identified, among which 53 showed consistent differences of 1.5 orders of magnitude across all five subtypes, and 400 met this criterion in at least two subtypes.

Analysis using PLIF revealed that the 53 consensus differential compounds involved 14 residues and formed 31 interaction fingerprints, with frequent interactions observed at Lys119 and Asp208. In contrast, the 400 partial differential compounds engaged 24 residues and yielded 69 interaction fingerprints (Supplementary Figure 10). Under a 70% similarity threshold, the compounds were clustered into 252 groups, with the highest-affinity compound retained from each group. Following visual inspection of binding conformations, 22 of the 53 consensus differential compounds and 119 of the 400 partial differential compounds were retained.

To validate the screening results, the top 10 compounds were ranked based on their Glide scores (see Supplementary Table 1). The ability of these compounds to inhibit the interaction between CRF55_01B gp140 and CD4 was evaluated using bio-layer interferometry (BLI) (Figure 7, Supplementary Figure 11). The results indicated that five compounds exhibited significant inhibitory activity, including two FDA-approved drugs. Notably, T21299 and T10808 showed inhibition rates of 20% and 19%, respectively (Figure 7). Subsequently, the cytotoxicity of the compounds toward TZM-bl cells was assessed using the CCK-8 assay. The results indicated that none of the five compounds exhibited significant cytotoxicity after 48 hours of treatment, except for T21102 and T10808 at 100 μM, which showed evident toxicity (Supplementary Figure 12A). A non-cytotoxic concentration (10 μM) of each compound was selected for further antiviral evaluation. The compounds were pre-incubated with HIV pseudovirus for 2 hours before inoculation into TZM-bl cells. After 4 hours, the culture medium was replaced. At 48 hours post-infection, ELISA and qRT-PCR were performed to evaluate antiviral activity. The results demonstrated that both T21299 and T8489 exhibited strong anti-HIV activity (Supplementary Figures 12B, C). Furthermore, molecular docking was carried out between these five highly effective small molecules and gp140 proteins derived from various HIV subtypes to delineate their binding sites (Supplementary Figures 13-17, Supplementary Table 3).