Thermodynamic variations at the HPV E1-E2 interface correlate with clinical risk groups: an in-silico analysis

Abstract

Background

Human Papillomavirus (HPV) infection is a leading cause of human cancers, including nearly all cases of cervical cancer. The viral life cycle and its associated oncogenic potential are critically dependent on the replication of the viral genome, a process driven by the viral E1 protein. Its central role and high degree of sequence conservation within the enzymatic core provide a structural rationale for targeted therapies.

Methods

A comprehensive in silico analysis was conducted to investigate the structural and functional roles of the HPV E1 protein using a multi-scale computational workflow. E1 sequences from the Papillomavirus Episteme database underwent multiple sequence alignment to delineate evolutionary relationships and identify conserved residues. Structural modeling using AlphaFold3 generated high-resolution models of E1 in complex with E2 and as hexameric assemblies. These models were subjected to thermodynamic profiling to calculate predicted binding energies (ΔG) and interface geometry and solvation. Finally, a logistic regression framework evaluated the predictive power of these parameters in distinguishing oncogenic risk.

Results

The C-terminal helicase was identified as the most highly conserved region, with near-universal invariance in key catalytic motifs essential for ATP binding and hydrolysis. In contrast, the N-terminal regulatory region exhibited greater sequence divergence. Structural mapping onto the E1 DNA-binding domain (DBD) and the hexameric helicase assembly revealed that conserved residues cluster near DNA-recognition surfaces of the DBD and the interaction surface for the E2 regulatory protein. Thermodynamic analysis revealed an energetic distinction between High-Risk (HR) and Low-Risk (LR) types. LR E1-E2 complexes exhibited higher predicted stability (ΔG =  − 25.65 kcal/mol) compared to HR complexes (ΔG =  − 20.95 kcal/mol; P < 0.001). Receiver Operating Characteristic analysis demonstrated that E1-E2 binding energy has strong discriminative ability within this dataset (AUC = 0.875).

Conclusions

The extensive conservation of E1’s core enzymatic machinery highlights its indispensable function. However, thermodynamic variations at the E1-E2 interface correlate with distinct clinical risk groups. These in silico findings provide a framework for understanding functional sites in the E1-E2 complex and identify the recruitment interface as thermodynamically distinct between classes of pathogenicity. These results motivate future experimental validation of E1-E2 binding kinetics.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12985-026-03221-2.

Background

Human Papillomaviruses (HPVs) are a diverse family of small, non-enveloped DNA viruses that represent a significant global health burden [1]. These viruses infect the keratinocytes of stratified epithelia, causing a wide spectrum of clinical manifestations ranging from benign hyperproliferative lesions, such as cutaneous and anogenital warts, to malignant carcinomas [2, 3]. The International Committee on Taxonomy of Viruses has classified the over 200 identified human-infecting HPV types into five principal genera: Alphapapillomavirus, Betapapillomavirus, Gammapapillomavirus, Mupapillomavirus, and Nupapillomavirus, which exhibit distinct tissue tropisms and pathogenic potentials [2, 3]. Among these, the mucosal-infecting Alphapapillomaviruses are the most clinically significant [4, 5]. They are further categorized based on their oncogenic potential into HR types, such as HPV16 and HPV18, which are the etiological agents of virtually all cervical cancers and a substantial fraction of other anogenital and oropharyngeal cancers, and LR types, such as HPV6 and HPV11, which are primarily associated with benign condylomata [6, 7].

The HPV genome is a circular, double-stranded DNA molecule of approximately 8 kilobases (kb) [2, 3]. This compact genome is organized into three regions: an early (E) region encoding regulatory proteins (E1, E2, E4, E5, E6, E7), a late (L) region encoding the L1 and L2 capsid proteins, and a non-coding Long Control Region (LCR) that contains the viral origin of replication (ori) and key regulatory elements [3, 8]. The viral life cycle is tightly linked to the differentiation program of the host keratinocyte. Upon infection of basal epithelial cells, the virus establishes its genome as a stable, low-copy-number extrachromosomal plasmid, or episome [9, 10]. As the infected cell divides and migrates towards the epithelial surface, the virus enters a productive phase characterized by a massive amplification of the viral genome, expression of late genes, and assembly of new virions [11]. This entire process is orchestrated by a limited set of viral proteins that utilize the host cell’s molecular machinery.

Central to the viral life cycle is the replication of its DNA genome, a process for which the virus encodes only two essential proteins: E1 and E2 [1, 12]. The E1 protein, the subject of this report, is the primary engine of viral replication [13]. It is the largest and most conserved protein encoded by papillomaviruses, reflecting its fundamental enzymatic function [14]. E1 is a multifunctional, ATP-dependent DNA helicase belonging to the AAA + (ATPases Associated with diverse cellular Activities) Superfamily 3 (SF3) [15, 16]. Its activities are diverse: it recognizes and binds the viral origin, melts the DNA duplex, and then unwinds the DNA in the 3’ to 5’ direction to provide a template for the host’s replication machinery [1]. To accomplish this, E1 recruits a host of cellular replication factors to the viral replication fork, including DNA polymerase α-primase, DNA polymerase ε, Replication Protein A (RPA), and Topoisomerase I [1, 17].

Despite its enzymatic efficiency, the E1 protein possesses low specificity for the viral origin of replication and requires a critical partnership with the viral E2 protein [1]. E2 functions as a "molecular matchmaker," binding with high affinity to specific DNA sequences (E2 binding sites, or E2BS) within the LCR and, via a direct protein–protein interaction, recruiting E1 to its cognate binding site (E1BS) located within the ori [18]. The formation of this E1-E2-ori ternary complex is the first step in initiating viral DNA replication, ensuring that the helicase activity of E1 is precisely targeted to the viral genome [1]. This regulatory checkpoint prevents aberrant DNA unwinding at non-viral sites, which would be catastrophic for the host cell.

Given E1’s essential role in the viral life cycle, understanding the molecular determinants of its function is of great importance. This study adopts a multi-scale analytical approach, mirroring the "analytical funnel" framework used in the parent study on the E2 protein, to systematically dissect the HPV E1 protein [18]. We begin with a broad phylogenetic survey of E1 proteins across the major human-infecting HPV genera to establish the evolutionary context. We then narrow our focus to a detailed sequence and structural analysis of E1’s conserved functional domains: the DNA-Binding Domain (DBD) and the Helicase/ATPase domain to identify universally conserved residues critical for function. Finally, we zoom in on the specific molecular interfaces that govern the E1-E2 interaction and conduct a comparative analysis between E1 proteins from HR and LR HPV types. Through this integrated approach, we aim to construct a comprehensive model of E1 function, describe the structural basis of its conservation, and highlight the features that establish its therapeutic interest for the development of broad-spectrum antiviral therapeutics. While the enzymatic machinery of the E1 helicase is under strict evolutionary pressure to remain conserved, we hypothesize that the regulatory interface with the E2 protein exhibits distinct thermodynamic variations. We define this variation specifically as diversity in the predicted interface stability across different genotypes, rather than conformational adaptability within a single complex. We propose that this thermodynamic variability correlates with clinical risk groups, distinguishing HR genotypes from LR variants in silico.

Methods

This study utilizes a multi-stage bioinformatic pipeline to link evolutionary conservation with thermodynamic stability (Fig. 1).

Fig. 1.

Computational workflow for the evolutionary and thermodynamic analysis of the HPV E1 protein. The study employs a "funnel-based" analytical strategy divided into three phases: (1) Evolutionary Analysis: Retrieval of HPV E1 sequences from the PaVE database, followed by Multiple Sequence Alignment (MAFFT) and phylogenetic reconstruction to establish evolutionary relationships. Post-hoc assessment (GUIDANCE2) ensured alignment reliability. (2) Structural Architecture: A stratified subset of 24 representative genotypes (12 HR, 12 LR) was subjected to high-resolution structural modeling via AF3. These models underwent rigorous thermodynamic profiling using physics-based (FoldX), geometric (PDBePISA), and contact-based (PRODIGY/Area-Affinity) algorithms to quantify interface stability. (3) Functional Consequence: The resulting thermodynamic metrics were utilized in Receiver Operating Characteristic (ROC) and univariate logistic regression analyses to evaluate whether structural stability is associated with oncogenic risk classification within this dataset

Sequence data collection

Complete E1 protein sequences for 194 human papillomaviruses, representing the Alpha, Beta, and Gamma genera, were retrieved from the Papillomavirus Episteme (PaVE) database (https://pave.niaid.nih.gov) [19]. The selection was restricted to strains officially classified as HPV genotypes to ensure a focus on human-infecting viruses while capturing a broad range of genetic diversity. Genera such as Mupapillomavirus and Nupapillomavirus were excluded from this analysis due to their high sequence divergence, which compromises the reliability of multiple sequence alignments, and the limited number of identified genotypes, which could statistically skew evolutionary insights. This methodology follows the data collection strategy employed in the reference study on the E2 protein [18].

Multiple sequence alignment and phylogenetic analysis

All retrieved E1 sequences were subjected to multiple sequence alignment (MSA) using MAFFT v7 (strategy E-INS-i) with default parameters [20]. The E-INS-i iterative refinement method was selected specifically for its ability to handle sequences with large unalignable regions, such as the intrinsically disordered N-terminal regulatory domain of E1, which varies significantly in length and composition across genera [21]. Post-alignment quality control was performed using GUIDANCE2 to verify column reliability (Additional file 1) [22]. To capture conservation at distinct evolutionary depths, three specific alignment datasets were generated: (1) Alpha-Ref, comprising reference genomes of the Alphapapillomavirus genus to identify core lineage-specific features; (2) Alpha-Var, comprising reference genomes plus variant sublineages (e.g., clinically significant variants like HPV52 B2) to assess intra-type conservation; and (3) ABG, a combined alignment of Alpha, Beta, and Gamma reference genomes to identify universally conserved residues. The resulting MSAs were imported into Jalview (Version 2.11.5.1) for visualization [23]. To calculate residue-by-residue conservation scores, amino acid sequences were first categorized into seven distinct physicochemical groups to determine Shannon Entropy. These raw entropy values were then mathematically inverted, scaled to an 11-point biological index, and adjusted for alignment gaps using a sequence-weighted penalty [24]. Fully invariant positions and positions with strong physicochemical conservation were subsequently identified and annotated.

Structural modeling, thermodynamic assessment, and statistical analysis

Full-length E1 and E2 sequences were aligned to the HPV18 E1-E2 crystal structure (PDB: 1TUE) [25] to define homologous interaction domains. Additionally, hexameric E1 helicase assemblies were modeled by aligning sequences to the homologous hexameric crystal structure of BPV E1 (PDB: 2GXA) [26], corresponding to the ATPase/Helicase domain. Three-dimensional structures were generated for a stratified subset of 24 HPV genotypes selected to represent distinct IARC associated risk categories. This dataset comprised 12 High Risk types (Group 1 carcinogens) and 12 types broadly classified as Low Risk (IARC Group 3) based on the most recent IARC Monograph classifications [27]. (Additional file 2) Group 3 genotypes were utilized as the comparative cohort because human papillomaviruses are not formally classified under IARC Group 4 as non-carcinogenic. As a result, Group 3 represents the baseline for types lacking established clinical evidence of oncogenicity. To ensure structural accuracy at the complex interface, structural modeling was performed using AlphaFold3 with the automated template-search feature enabled [28]. This configuration allows the algorithm to automatically leverage available experimental structures on the backend, including the homologous crystal structure of the HPV18 E1-E2 complex (PDB 1TUE). In this capacity, AF3 functions as an advanced homology modeling pipeline, leveraging the reliable structure of the known E1-E2 complex while utilizing its diffusion-based architecture to accurately resolve genotype-specific side-chain conformations across divergent risk groups. To further ensure fold accuracy, structures were subjected to a quality control filter in which residues with a predicted Local Distance Difference Test (pLDDT) score below 70 were trimmed to remove low-confidence disordered regions. AlphaFold3 has demonstrated high accuracy for protein–protein interfaces, with reported ipTM > 0.8 typically corresponding to confident structural predictions. The retained interfacial domains exhibited high structural reliability, with all models exceeding an ipTM of 0.80 except for the HPV61 E1-E2 complex (ipTM = 0.77), confirming their suitability for thermodynamic profiling.

While contact-based scoring functions (PRODIGY & Area-Affinity) [31, 32] were calculated (data provided in Supplementary Materials), they were excluded from the primary risk prediction model. Preliminary analysis revealed that while HR and LR types maintain similar residue contact counts (quantity), the thermodynamic quality of the interface, specifically atomic packing and van der Waals complementarity, differed significantly. Therefore, physics-based algorithms (FoldX) and solvation-aware geometric analysis (PDBePISA) were prioritized to capture these subtle energetic distinctions.

Comparative analysis between high-risk and low-risk cohorts was performed using independent Welch’s t-tests, with Cohen’s d utilized to determine effect size. To evaluate the predictive capacity of structural stability on oncogenicity, a univariate logistic regression model was constructed using the FoldX E1-E2 binding energy as the sole predictor. The baseline model performance was evaluated via Receiver Operating Characteristic Area Under the Curve (ROC AUC), and Odds Ratios were calculated per 1 kcal/mol change in binding energy. Bootstrap validation with 10,000 resamples was performed to generate 95% Confidence Intervals. To verify mathematical robustness and mitigate overfitting on the 24-genotype array, the model was subjected to Leave-One-Out Cross-Validation and repeated fivefold cross-validation. Furthermore, model significance was verified via a 1,000-shuffle permutation test. Outlier sensitivity and phylogenetic influence were assessed through Leave-One-Clade-Out cross-validation and targeted exclusion protocols [33].

Results

Evolutionary context and domain architecture of the HPV E1 protein

The study began with a survey of the evolutionary relationships and conserved structural organization of the HPV E1 protein across the primary human-infecting papillomavirus genera. The Alphapapillomavirus clade contains the most clinically significant types, including both HR (e.g., HPV16, HPV18) and LR (e.g., HPV6, HPV11) mucosal viruses, while the Beta and Gamma genera are predominantly associated with asymptomatic cutaneous infections [34, 35].

The E1 protein maintains a remarkably conserved modular domain architecture, which is essential for its complex, stepwise role in viral DNA replication [13]. This architecture, illustrated schematically in Fig. 2, can be divided into three principal functional regions based on sequence analysis and experimental data from UniProt and related studies [13]. The N-terminal region (NTR, approx. residues 1–207) is the most variable segment and contains regulatory motifs, including a Nuclear Localization Signal (NLS), a Nuclear Export Signal (NES), and a cyclin-binding (RXL) motif that tethers E1 activity to the host cell cycle [13]. The central DNA-Binding Domain (DBD, residues 208–354 in HPV18 E1) is responsible for the initial, sequence-specific recognition of the viral origin of replication [13]. The C-terminal portion of the protein constitutes the large Helicase/ATPase domain (Residues 355–655), which is the most highly conserved region across all papillomaviruses [16, 36]. This enzymatic core can be further subdivided into an oligomerization domain and the AAA + ATPase motor that powers DNA unwinding [26].

Fig. 2.

Structural architecture and conservation of the HPV E1 protein. A global view of the HPV-18 E1 monomer modeled using AlphaFold3, illustrating the modular domain organization and distribution of evolutionary conservation. The protein is composed of three distinct functional regions: the intrinsically disordered N-terminal regulatory domain (Dark blue), the central DNA-Binding Domain (DBD, Green), and the C-terminal Helicase/ATPase domain (Cyan). The insets provide high-resolution views of each domain, with residues exhibiting complete physicochemical invariance or conservation across the Alpha, Beta, and Gamma genera rendered as red sticks. Note the high density of conserved residues forming the hydrophobic core and catalytic pocket of the Helicase (Left) and the DNA-interaction surface of the DBD (Right), contrasting with the sparse conservation observed in the flexible N-terminal linker (Middle)

The extreme conservation of the E1 protein, particularly its enzymatic C-terminus, stands in contrast to the higher sequence variability observed in other viral regulatory proteins like E2, E6, and E7 [18]. This disparity in evolutionary constraint is a direct reflection of their distinct biological roles. E1’s function as a molecular motor necessitates a rigid, precisely folded structure to coordinate ATP hydrolysis with DNA translocation and to maintain interactions with many highly conserved host replication factors, such as DNA polymerases and RPA [37]. Any significant deviation in these interaction surfaces would be deleterious to viral replication [13]. Conversely, proteins like E6 and E7, which target host tumor suppressors (p53 and pRb), are engaged in a continuous evolutionary "arms race" with the host, driving faster sequence evolution [38]. Similarly, the flexible linker of the E2 protein allows for greater structural plasticity in its regulatory functions [18]. This deep conservation of E1’s core machinery across all HPV types makes it a plausible target for the development of broad-spectrum antiviral drugs, as an inhibitor targeting a conserved site is likely to be effective against a wide array of pathogenic HPV types [39].

The most highly conserved features include the bipartite Nuclear Localization Signal (NLS) centered on the basic cluster at residues 86–88 (Kr-K/R-K), and a Leucine-rich Nuclear Export Signal (NES) at residues 109/113, both of which are essential for nucleocytoplasmic shuttling [40]. Additionally, the Cyclin Binding Motif (RXL) at residues 127–129 exhibits strong physicochemical conservation, indicating E1’s role in coupling viral replication to the host cell cycle [41, 42]. Notably, the extreme N-terminus (residues 16–35) retains a conserved hydrophobic patch implicated in the recruitment of the host factor p80/UAF1, a known regulator of viral replication [43–45]. These findings confirm that while the N-terminus is evolutionarily variable, it maintains a strict ‘motif-based’ conservation strategy to ensure host interaction.

Analysis of conserved residues in the N-terminal regulatory region

While the C-terminal of E1 is well conserved, the N-terminal region (residues 1–207) is intrinsically disordered, serving as a flexible docking platform for host cellular factors [46]. Despite its high overall sequence divergence, our conservation analysis identified specific islands of invariance corresponding to critical Short Linear Motifs (SLiMs) (Fig. 3). Detailed residue conservation scores and structural annotations are provided in Additional file 5.

Fig. 3.

Conservation mapping of the HPV E1 N-Terminal Regulatory Region. Conservation scores are derived from Shannon Entropy calculations and scaled from 0 to 11. Residues with scores of 9.0 or greater are highlighted to indicate strict evolutionary conservation. The analysis is stratified into three evolutionary tiers: Alpha-Reference genotypes (A, top panel, N = 65), Alpha-Variant sublineages (B, middle panel, N = 183), and a comprehensive pan-viral alignment of Alpha, Beta, and Gamma genera (C, bottom panel, N = 194). Key structural and functional motifs such as the p80/UAF1 interface, Nuclear Localization Signal (NLS), Nuclear Export Signal (NES), and Cyclin RXL binding site are annotated with dashed boundaries. Significant residues are color-coded by physicochemical properties, while low-conservation regions are rendered in grey

Analysis of conserved residues in the E1 DNA-binding domain (DBD)

The E1 DBD is the structure that confers the initial specificity for the viral genome, recognizing a palindromic E1 binding site (E1BS) within the origin of replication [47, 48]. The crystal structure of the monomeric HPV18 E1 DBD (PDB: 1R9W) [49] provides a high-resolution view of this domain, revealing a fold characterized by a central five-stranded antiparallel β-sheet that is flanked on either side by α-helices [49]. E1 binds to the E1BS as a dimer, a process mediated by a hydrophobic interface formed by the α3 helix of each monomer [13, 50]. Unlike many DNA-binding proteins that use a canonical helix-turn-helix motif, E1 utilizes an extended loop and an adjacent α-helix to form a continuous, positively charged surface that engages the DNA primarily through shape complementarity and interactions with the phosphate backbone, rather than extensive base-specific contacts [51] (Fig. 4).

Fig. 4.

Evolutionary conservation of the HPV E1 DNA Binding Domain. Shannon Entropy conservation scores are plotted across three taxonomic levels (panels A, B, and C) to evaluate intra-type and pan-genera stability. Residues maintaining a conservation score of 9.0 or higher are emphasized and color-coded according to their structural side-chain properties including hydrophobic, polar uncharged, acidic, and basic groups. Essential functional interfaces required for viral replication, including the Hydrophobic Core, Dimerization interface, and specific DNA Backbone and Phosphate contacts, are mapped across the domain sequence

Analysis of conserved residues in the C-terminal helicase/ATPase domain

The C-terminal domain of E1 contains the enzymatic machinery that powers viral replication. This domain assembles into a hexameric ring that encircles and translocates along single-stranded DNA (ssDNA) [26]. The crystal structure of the bovine papillomavirus (BPV) E1 helicase domain bound to ssDNA and ADP (PDB: 2GXA) serves as an outstanding model for this molecular machine. The upper "collar" ring is formed by the oligomerization domains, while the lower "motor" ring is composed of the six AAA + ATPase domains that drive translocation. ssDNA passes through the central channel of this AAA + ring, where it is engaged by a series of DNA-binding hairpins [26].

The unwinding of DNA is fueled by the hydrolysis of ATP at the interfaces between the AAA + subunits. This process follows a coordinated, sequential mechanism often described as a "spiral staircase" or "hand-over-hand" model [26]. In this model, each of the six subunits cycles through three states (ATP-bound, ADP-bound, and nucleotide-free), inducing a wave of conformational changes that propagate around the ring. These changes reposition the DNA-binding hairpins, which sequentially grip and release the ssDNA backbone, effectively pulling the strand through the central pore one nucleotide at a time [26]. This complex enzymatic function is mediated by a set of short, highly conserved sequence motifs within the AAA + domain that are hallmarks of SF3 helicases and P-loop NTPases in general [36]. The specific residues comprising these motifs are detailed in Fig. 5. Complete residue-level conservation scoring for these catalytic motifs is detailed in Additional file 5.

Fig. 5.

Structural and evolutionary conservation of the HPV E1 Helicase and ATPase Domain. The three-tiered plot details Shannon Entropy sequence conservation for Alpha-Reference (panel A, N = 65), Alpha-Variant (panel B, N = 183), and pan-genera (panel C, Alpha, Beta, Gamma, N = 194) datasets. Applying a stringent 9.0 conservation cutoff visually isolates the core catalytic residues from highly variable background regions. Critical enzymatic and structural motifs including Walker A, Walker B, the Arginine Finger, Sensor 1, and E2 binding interfaces are demarcated

The enzymatic activity of the E1 helicase relies on a coordinated network of canonical motifs. Central to this machinery is the Walker A motif (P-loop), defined by the consensus GxxGxGK[T/S] [52]. This flexible loop binds the β and γ phosphate groups of the ATP molecule, an interaction anchored by an absolutely invariant Lysine residue [26]. This is functionally coupled with the Walker B motif (consensus hhhh-DE), which positions a critical Aspartate residue to coordinate the Mg²⁺ cofactor required for catalysis, while an adjacent Glutamate residue activates the attacking water molecule necessary for ATP hydrolysis [26]. To translate this chemical energy into mechanical motion, the Sensor 1 and 2 motifs detect the presence of the γ-phosphate, triggering the distinct conformational changes required for hydrolysis and subsequent DNA translocation [53].

Crucially, the activity of the individual subunits is synchronized by the Arginine Finger. In this "in-trans" mechanism, a critical Arginine residue from one subunit extends into the ATP-binding pocket of the adjacent subunit, promoting efficient hydrolysis and linking the catalytic state of the entire hexameric ring. The absolute conservation of these catalytic residues across all human papillomaviruses highlights the extreme evolutionary pressure to maintain a functional helicase motor, marking this enzymatic core as a plausible viral target.

Structural dissection of the E1-E2 interaction interface

The interaction between the E1 helicase and the E2 transactivator is the master checkpoint for the initiation of viral DNA replication, ensuring that this process begins only at the viral origin [1]. The crystal structure of the complex between the HPV18 E1 helicase domain and the E2 transactivation domain (PDB: 1TUE) provides a detailed image of this critical regulatory partnership as replicated by AF3 (Fig. 6) [25].

Fig. 6.

Steric fit of the HPV E1-E2 regulatory complex. Surface representation of the E2 Transactivation Domain (beige) bound to the E1 Helicase domain, shown as a ribbon. The E1 structure is colored by evolutionary conservation (Red: Conserved, Teal: Variable), illustrating that the interaction is anchored by highly conserved residues. The E2 surface forms a binding pocket that accommodates the E1 helix, physically occluding the E1 oligomerization interface. This binding pose effectively masks the E1 oligomerization surface, maintaining the protein in a monomeric, replication-inactive state during the initial recruitment phase. The magnified inset highlights the critical salt bridge formed between E1 Arg454 and E2 Glu43

The structure reveals that the N-terminal transactivation domain of E2 binds to a composite surface on the E1 helicase domain. This binding event serves a crucial inhibitory function: it physically occludes a surface on E1 that is essential for E1-E1 oligomerization [25, 54]. By acting as a shield, E2 effectively holds E1 in a monomeric, replication-inactive state. This "matchmaker" function ensures that E1 is safely delivered to the origin without engaging in helicase activity elsewhere in the genome [25]. The subsequent dissociation of E2 is an ATP-dependent process, which suggests that only when E1 is properly loaded onto the origin and binds ATP can it shed its E2 inhibitor and begin the process of assembling into a functional hexameric helicase [25]. This regulatory mechanism represents an allosteric switch. E2 acts as a negative allosteric regulator of E1’s oligomerization and ATPase activity. Once the E1-E2 complex is tethered to the origin DNA, the high local concentration of E1, coupled with the cellular ATP pool, favors a conformational shift in E1 upon ATP binding. This shift simultaneously weakens E1’s affinity for E2, causing its release, and exposes the E1-E1 interaction surface, thereby triggering hexamer assembly [55].

To define this regulatory interface at the atomic level, the specific residues involved in the E1-E2 interaction were identified from the 1TUE structure using computational analysis. The interface is extensive, stabilized by a network of salt bridges, hydrogen bonds, and hydrophobic contacts. This analysis confirms the key interactions noted in the reference E2 study, such as the salt bridges formed between E2 residues Glu24 and Glu43 and E1 residues Arg447/Arg622 and Arg454, respectively [25].

Thermodynamic stability of the E1-E2 interface differentiates risk groups

While static crystal structures suggest a conserved binding mode, our thermodynamic analysis revealed differences between HR and LR HPV types. The E1–E2 interface in LR types is predicted to be significantly more stable, with a mean ΔG of − 25.65 kcal/mol, compared to HR types, which exhibit a mean ΔG of − 20.95 kcal/mol (Table 1). Within the analyzed dataset HR types display a looser interface, potentially conferring the structural variability required for rapid replication firing (Welch’s t-test, P < 0.001; Cohen’s d = 1.68). It is notable that this thermodynamic stratification was only captured by physics-based algorithms (FoldX) and geometric surface analysis (PDBePISA), which measure the precise atomic packing and solvent exclusion at the interface. In contrast, contact-based scoring functions (PRODIGY, Area-Affinity), which rely primarily on the largely conserved residue contact counts and interface chemistry, failed to distinguish between HR and LR types (see Additional file 6). This discrepancy indicates that the functional difference lies not in the quantity of residues buried, but in the quality and complementarity of the steric fit, a feature best resolved by energetic simulation. Furthermore, side-chain repacking and charge stability controls simulated at pH 6.5 yielded a mean absolute computational energy difference of only 0.49 kcal/mol compared to neutral pH. This confirms that the predicted thermodynamic rank-ordering between high-risk and low-risk cohorts is robust and invariant to physiological protonation states (Additional File 3 and 4).

Table 1.

Thermodynamic and geometric assessment of E1-E2 interface stability across 24 HPV genotypes

HPV type	FoldX	PDBePISA
	ΔG	BSA	ΔGSolv
HR
HPV16	-25.54	1173.8	-7.6
HPV18	-21.37	853.8	-3.4
HPV31	-19.66	984.9	-5.6
HPV33	-20.63	1018.4	0.8
HPV35	-20.77	1147.1	-2.5
HPV39	-19.17	958.8	-5.3
HPV45	-23.19	1020.0	-7.4
HPV51	-23.98	740.1	-4.8
HPV52	-21.55	940.8	0.8
HPV56	-16.55	939.6	-5.9
HPV58	-19.08	869.6	-2.7
HPV59	-19.85	1065.8	-7.3
HR Mean	-20.95	976.1	-4.2
HR StDev	2.43	122.5	2.9

LR
HPV6	-27.89	1175.7	-8.6
HPV11	-27.23	1242.3	-6.5
HPV40	-22.22	940.0	-7.0
HPV42	-25.27	1105.4	-9.7
HPV43	-25.91	1194.0	-12.3
HPV44	-28.85	1319.5	-7.2
HPV54	-25.46	1085.8	-1.5
HPV61	-18.25	873.8	-7.3
HPV71	-26.82	1209.5	-9.6
HPV72	-23.31	1320.8	-3.6
HPV81	-27.15	875.7	-8.3
HPV89	-29.39	1253.8	-3.6
LR Mean	-25.65	1133.0	-7.1
LR StDev	3.13	160.0	3.0

Comparison of physics-based energy calculations (FoldX) and geometric surface analysis (PDBePISA). FoldX ΔG indicates the repaired Model 0 (highest-ranked AF3 seed) stability. PDBePISA metrics (Buried Surface Area and Solvation Energy) quantify the tightness of the interface packing. Contact-based metrics and clade specific classifications are available in Additional file 6. ΔG and ΔGˢᵒˡᵛ are reported in kcal/mol; Buried Surface Area (BSA) is reported in Ų. Ensemble statistics across all 5 seeds in Additional file 4

Thermodynamic stability was assessed using physics-based energy calculations, at temperature 298 K, ionic strength 0.05 M, pH 7.0, unless otherwise specified (FoldX 5.1) [29] and geometric surface analysis (PDBePISA) [30]. To validate thermodynamic robustness and quantify structural modeling uncertainty, FoldX AnalyseComplex was performed across all five independent structural models generated by AlphaFold3 for each genotype, yielding 120 total calculations. The primary binding energy reported in Table 1 corresponds to the highest-ranked model (Model 0), while the full five-seed ensemble average (mean ± SD) is reported in additional file 4. Model 0 was utilized for the primary comparison to preserve a single, self-consistent structure per genotype for all downstream geometric and atomic calculations, while the multi-seed ensemble statistics were leveraged to confirm biophysical robustness (Additional File 4). This multi-seed approach accounts for potential sidechain and backbone variations introduced during the diffusion-based structural generation phase. Side-chain repacking and charge stability controls were performed at both pH 7.0 and pH 6.5 across all 24 genotypes to ensure energetic calculations were not artifacts of default protonation states.

Decomposition of the energy terms indicated that this stability is driven primarily by optimized Van der Waals packing (ΔG_VdW =  − 18.34 for LR vs. − 14.83 for HR; P < 0.01), indicating that LR types rely on highly optimized interfacial packing to maintain a stable recruitment checkpoint. In contrast, HR types display a weaker predicted interface, which may confer more rapid transition to active replication states or integration events. To identify the specific residues driving this thermodynamic variation, a systematic residue-level energy decomposition was conducted for all twenty-four genotypes, with positions mapped to the homologous coordinates of HPV18. This analysis revealed that the interface is comprised of a conserved functional core and a variable peripheral region (Additional File 4 and 7). Core residues, including E1 isoleucine 461 and E2 isoleucine 20, contribute strong stabilizing energy across all genotypes. The thermodynamic differences between risk groups are instead localized to specific peripheral positions. For example, E1 Tyr585 and E1 Arg589 contribute stabilizing energy in low-risk genotypes but remain neutral in high-risk genotypes. In contrast, E2 glutamate 24 provides stronger relative stabilization in the high-risk cohort, whereas E1 lysine 476 introduces a destabilizing clash penalty of approximately 0.5 kcal/mol in high-risk genotypes that is absent in low-risk genotypes. Detailed residue contributions are provided in Additional File 4.

To confirm that this phenomenon is specific to the recruitment interface and not a general property of HR viral proteins, we performed a control analysis on the E1-E1 hexameric interface. No significant difference was found between HR and LR hexamer stability (P = 0.34) (Table 2), confirming that the thermodynamic divergence is specific to the E1-E2 regulatory checkpoint. However, identical thermodynamic principles should not be assumed for both contexts. The recruitment phase relies on a flexible binary interaction, whereas the processive helicase motor utilizes a highly constrained multi-subunit ring. This structural divergence implies that evolutionary pressures shape the interface energetics of these two complexes independently.

Table 2.

Thermodynamic and geometric assessment of E1-E1 hexamer interface stability

HPV type	FoldX	PDBePISA
	ΔG	BSA	ΔGSolv
HR
HPV16	-30.24 ± 1.20	1842 ± 20	-12.7 ± 0.7
HPV18	-37.67 ± 1.65	2157 ± 44	-15.2 ± 1.4
HPV31	-30.33 ± 0.28	1704 ± 24	-10.2 ± 0.3
HPV33	-28.54 ± 1.22	1796 ± 12	-9.0 ± 1.5
HPV35	-35.39 ± 1.66	1824 ± 28	-11.4 ± 1.5
HPV39	-24.30 ± 2.19	1775 ± 17	-11.0 ± 0.9
HPV45	-37.10 ± 1.36	1890 ± 23	-11.4 ± 0.5
HPV51	-35.49 ± 2.32	1841 ± 21	-14.9 ± 1.2
HPV52	-34.28 ± 2.06	1872 ± 27	-11.2 ± 3.0
HPV56	-40.49 ± 0.66	1983 ± 32	-12.6 ± 1.1
HPV58	-25.31 ± 1.18	1715 ± 22	-11.9 ± 0.8
HPV59	-31.50 ± 1.46	1899 ± 41	-10.2 ± 0.8
HR Mean	-32.6	1858.1	-11.8
HR StDev	5.0	122.2	1.8

LR
HPV6	-34.33 ± 0.72	1844 ± 47	-14.9 ± 1.2
HPV11	-38.78 ± 1.63	1973 ± 22	-15.0 ± 1.5
HPV40	-27.46 ± 1.72	1673 ± 30	-16.1 ± 1.3
HPV42	-38.72 ± 1.68	1926 ± 20	-12.0 ± 1.5
HPV43	-28.32 ± 0.78	1647 ± 15	-15.1 ± 1.8
HPV44	-42.22 ± 2.04	1870 ± 29	-14.3 ± 1.4
HPV54	-31.70 ± 1.40	1692 ± 22	-15.8 ± 0.8
HPV61	-29.22 ± 1.54	1853 ± 19	-10.5 ± 0.8
HPV71	-32.89 ± 1.05	1747 ± 30	-11.2 ± 0.5
HPV72	-36.41 ± 2.18	1838 ± 25	-11.1 ± 0.8
HPV81	-36.08 ± 1.80	1910 ± 34	-11.8 ± 1.1
HPV89	-37.16 ± 2.64	1836 ± 47	-11.7 ± 1.9
LR Mean	-34.4	1817.4	-13.3
LR StDev	4.6	104.7	2.1

Comparison of physics-based energy calculations (FoldX) and geometric surface analysis (PDBePISA). FoldX ΔG represents the stability of the repaired Model 0 hexamer assembly. PDBePISA metrics (Buried Surface Area and Solvation Energy) quantify the tightness of the interface packing. Contact-based metrics and clade specific classifications are available in Additional file 8. Unlike the single heterodimeric interface of E1–E2 (Table 1), the hexameric E1 helicase contains six equivalent inter-subunit interfaces. The reported Standard Deviation ( ±) here reflects the variance calculated across the six unique inter-subunit interfaces within this repaired Model 0 structure, capturing the structural heterogeneity of the ring. Full genotype-level data are provided in Supplementary Tables. ΔG and ΔGˢᵒˡᵛ are reported in kcal/mol; Buried Surface Area (BSA) is reported in Ų

E1-E2 binding energy correlates with clinical risk profiles

Given the thermodynamic separation observed between HR and LR genotypes, we tested whether E1–E2 interface stability alone could classify clinical risk using a univariate logistic regression model with FoldX binding energy (ΔG) as the sole predictor variable. The baseline model displayed strong discriminative ability within the 24-genotype dataset, achieving a Receiver Operating Characteristic Area Under the Curve (ROC AUC) of 0.875, with a bootstrap 95% confidence interval of 0.69 to 1.00 generated via 10,000 resamples (Fig. 7B). The estimated odds ratio was 1.72 per + 1 kcal/mol increase in ΔG (representing a progressively weaker binding interface), indicating an estimated 1.72-fold increase in the odds of a high-risk classification for each 1 kcal/mol weakening of the E1–E2 interaction complex (p = 0.008). Individual genotype-level model-fitted risk probabilities are supplied for all 24 strains within Additional File 4.

Fig. 7.

Thermodynamic stability of the E1-E2 interaction correlates with oncogenic risk. (A) Univariate logistic regression analysis of FoldX binding energy (ΔG) versus viral risk classification (n = 24). HR genotypes (blue circles) significantly associate with weaker binding energies (less negative ΔG; P = 0.008) compared to LR genotypes (orange circles). The solid line represents the logistic fit, and the shaded region indicates the 95% confidence interval. (B) Receiver Operating Characteristic (ROC) curve for the logistic model. The Area Under the Curve (AUC) of 0.875 demonstrates high discriminative power in distinguishing HR from LR types based solely on interface stability

To evaluate the structural reliability of the underlying computational pipeline, the modeling uncertainty across the AlphaFold3 structural seeds was analyzed. The mean inter-seed standard deviation across the 120 total calculations was 1.80 kcal/mol. In context, this structural variance is 2.6-fold smaller than the 4.68 kcal/mol inter-clade gap observed between the clinical risk groups. This clear separation demonstrates that the thermodynamic divergence captured by the model is likely driven by genuine biological differences rather than stochastic variations in the structural generation process.

Additionally, to verify the mathematical robustness of this logistic regression framework and mitigate the statistical risks of model overfitting on a 24-genotype array, a Leave-One-Out Cross-Validation (LOOCV) pipeline was implemented. The LOOCV protocol yielded a cross-validated classification accuracy of 75.0% and a cross-validated ROC AUC of 0.8472, confirming that the thermodynamic stability profiles of the recruitment interface maintain reproducible predictive power when generalizing to unrecognized viral genotypes. To further assess model generalizability and guard against overfitting, we also performed repeated fivefold cross-validation, which yielded a consistent mean ROC AUC of 0.8413 (95% CI: 0.7778–0.8681). Finally, to verify that the classification power is not due to random label assignment, a 1,000-shuffle permutation test was performed, yielding a highly significant empirical p-value of 0.0010. The logistic regression equation calculated from the full dataset was parameterized as: Log-Odds (HR) = 12.6394 + (0.5428 × ΔG) where the fitted intercept is 12.6394 and the slope coefficient for the FoldX Delta G predictor is 0.5428. Utilizing a standard classification probability threshold of 0.5, the cross-validated model achieved an operating sensitivity of 75.0% and a specificity of 75.0%. This predictive boundary produced a cross-validated confusion matrix consisting of 9 True Positives, 9 True Negatives, 3 False Positives, and 3 False Negatives. Individual genotype-specific cross-validated probability values (PLOOCV) are cataloged in Additional File 4. Furthermore, a post-hoc sensitivity analysis was conducted to systematically evaluate the numerical impact of outliers within the modeling. To achieve this, HPV16 (a hyper-prevalent high-risk genotype presenting an anomalously stable interface binding energy) and HPV61 (a low-risk genotype displaying low AlphaFold3 structural confidence metrics and a highly destabilized interface) were extracted from the data array. When filtering only the HPV16 reference outlier, the LOOCV AUC improved to 0.8864 with an operating sensitivity of 81.8% and a specificity of 83.3%. When filtering only the HPV61 outlier, the LOOCV AUC improved to 0.9091 with an operating sensitivity of 83.3% and a specificity of 81.8%. When the logistic regression analysis was completely re-fitted across this filtered 22-genotype subset, the equation adjusted to: Log-Odds (HR) = 28.3738 + (1.2189 × ΔG) with a filtered model intercept of 28.3738 and an increased slope coefficient of 1.2189. Under these outlier-free conditions, the cross-validated discriminative capacity of the interface increased significantly, yielding an elevated out-of-sample LOOCV ROC AUC of 0.9421 and an overall cross-validated accuracy of 81.8%. This filtered model architecture operated with a confusion matrix consisting of 9 True Positives, 9 True Negatives, 2 False Positives, and 2 False Negatives. The steady preservation of a positive slope coefficient across both the full and filtered datasets (+ 0.5428 versus + 1.2189) supports the calibration stability of the underlying biophysical signal: structural variations that weaken the E1–E2 regulatory checkpoint are uniformly associated with clinical high-risk oncogenic classification profiles across these viral clades.

To analytically address the outlier status of the HPV16 reference genome identified during sensitivity testing, we evaluated 5 seeds of select sublineages (A1,A4,D2 and D3). While the primary Model 0 reported in Table 1 yields a single-structure binding energy of -25.54 kcal/mol for the reference type, averaging across the multiple structural seeds (as reported in Additional File 4) weakens the reference A1 interface down to a five-seed ensemble mean of -23.32 kcal/mol (standard deviation 2.67). Furthermore, when evaluating the high-risk non-European sublineages, the A4 sublineage maintained an ensemble binding energy of -23.15 kcal/mol (standard deviation 0.60), while the hyper-carcinogenic D2 and D3 sublineages exhibited substantially weakened interface stabilities of -21.73 kcal/mol (standard deviation 2.17) and -22.10 kcal/mol (standard deviation 1.16). Residue-level energy decomposition indicates this destabilization is primarily driven by the loss of favorable interactions at Proline 157 (E1 protein) and Tyrosine 178 (E2 protein), which offsets localized energetic gains at Tyrosine 19 (E2 protein).

Clade-specific structural strategies

Beyond the binary HR/LR distinction, clade-specific analysis (Additional file 6) revealed unique structural strategies. Structural analysis of the E1-E2 interface density suggests distinct stabilization strategies between HR clades. Alpha-9 maintains a high bond density (~ 26.4 bonds/1000Ų), whereas Alpha-7 relies on a lower density (~ 21.7 bonds/1000Ų) interface, prioritizing geometric shape complementarity. To determine if E1 recruitment and hexamer assembly are evolutionarily coupled, we performed a global correlation analysis of their binding energies. Across the full dataset, no significant correlation was observed, suggesting that the E1-E2 interaction and the E1-E1 hexamer generally evolve as independent functional modules. This independence suggests that the thermodynamic variations observed in the recruitment interface represent a specific feature associated with viral risk profiles, unrelated to the stability requirements of the hexameric assembly.

To evaluate whether the observed thermodynamic differences between high-risk and low-risk genotypes might be influenced primarily by phylogenetic clustering, a clade-based sensitivity analysis was conducted. The human papillomavirus Alpha genus is highly stratified, with the majority of the clinical cancer burden associated with the Alpha-9 and Alpha-7 species groups. When evaluated independently against the low-risk pool (which maintained a mean Delta G of -25.65 kcal/mol), both the Alpha-9 clade (mean Delta G of -21.21 kcal/mol, p-value equals 0.0045) and the Alpha-7 clade (mean Delta G of -20.90 kcal/mol, p-value equals 0.0042) showed significantly weaker E1-E2 binding energies. Furthermore, to explicitly test model behavior across evolutionary boundaries, a Leave-One-Clade-Out cross-validation protocol was executed. This framework yielded a predictive accuracy of 75.0%, matching the standard cross-validation performance.

Discussion

The replication of the Human Papillomavirus genome is a masterfully orchestrated process that hinges on the function of the viral E1 protein, the sole enzyme encoded by the virus [13]. This study has employed a comprehensive bioinformatic and structural approach to dissect the E1 protein, revealing a sophisticated molecular machine whose core functions are under immense evolutionary pressure to remain conserved, yet whose regulatory features show adaptations that may contribute to the differential oncogenic potential among HPV types.

Our analysis confirms that the E1 protein is a multi-domain entity whose functions are executed in a precise, stepwise manner. The process begins with the critical E1-E2 interaction, a regulatory checkpoint that ensures replication initiates exclusively at the viral origin [1]. As revealed by the 1TUE crystal structure and our interface analysis, the E2 transactivation domain binds to and sterically occludes an oligomerization surface on the E1 helicase domain [25]. This act of molecular matchmaking delivers E1 to the origin in a monomeric, inactive state. The E1 DNA-binding domain, a structurally conserved module detailed in the 1R9W structure, then engages the E1BS [50]. This tethering event, coupled with the high cellular concentration of ATP, triggers an allosteric switch. ATP binding to E1 induces a conformational change that simultaneously promotes the dissociation of the E2 inhibitor and unmasks the E1-E1 interface, priming the system for the next step: helicase assembly [55]. E1 monomers then self-assemble into a double-hexameric ring around the origin DNA [56]. Once assembled, the enzymatic core, detailed by the 2GXA structure and our conservation analysis of the AAA + motifs, begins its work. The universally conserved Walker A, Walker B, and Arginine Finger motifs coordinate the binding and sequential hydrolysis of ATP, driving a wave of conformational changes that power the processive 3’-5’ translocation of the helicase along the DNA, unwinding the duplex to create the replication fork [13].

The strong conservation of E1’s enzymatic domains across the Alpha, Beta, and Gamma genera shows the rigid structural and chemical constraints required for its function. The ATPase active site, formed at the interface of adjacent subunits, is a source of absolutely invariant residues from multiple motifs (Walker A/B, Sensor 1/2, Arginine Finger) that must be perfectly positioned for catalysis [57]. This deep conservation reflects not only the demands of the enzymatic reaction itself but also the necessity of maintaining precise interaction surfaces for a suite of equally conserved host replication factors, such as DNA polymerase α/ε and RPA, which E1 must recruit to build a functional replisome [58]. In contrast, the greater sequence divergence in the N-terminal regulatory domain likely reflects evolutionary adaptation to different cellular contexts and host-specific regulatory factors, such as cyclins, which differ between the mucosal and cutaneous environments favored by different HPV genera [59]. The sequence divergence of the N-terminal domain may also have implications for the viral manipulation of the DNA Damage Response (DDR). Since nuclear accumulation of E1 is known to trigger an ATM-dependent DNA damage signal [40], variations in N-terminal phosphorylation sites or nuclear localization kinetics between genotypes could tune the intensity of this checkpoint activation, balancing viral replication needs with host cell survival. Our conservation analysis provides direct evidence for this mechanism by revealing an interesting yet somewhat counterintuitive trend regarding the Nuclear Export Signal (NES) and the p80/UAF1 interaction motif. While these motifs exhibit strong physicochemical conservation across the pan-viral ABG alignment, they demonstrate notable sequence variance within the Alpha genus specifically (Fig. 3). Previous studies utilizing mutant HPV31 genomes have demonstrated that an inability to export E1 from the nucleus under physiological conditions directly results in increased levels of DNA damage [40]. Therefore, the genus-specific sequence plasticity of the NES likely allows high-risk mucosal types to finely tune these exact nuclear export kinetics, ensuring just enough E1 accumulations to trigger the necessary ATM-dependent DDR. Similarly, the variation in the p80/UAF1 motif highlights an evolutionary adaptation of the E1 protein to accommodate the specific host factors of mucosal epithelia [43, 44].

This framework provides context for understanding the significant thermodynamic differences we identified between HR and LR E1 proteins. Our data challenges the notion that these proteins are functionally identical. Rather, they exhibit a thermodynamic trade-off. It is important to emphasize that these FoldX values represent computational scores rather than direct, experimentally measured binding free energies, and that the biological relevance of these findings rests entirely on relative differences between genotypes rather than on absolute energy values. Consequently, this observed gap serves as a comparative metric for risk group association, independent of the systematic biases inherent in empirical scoring functions. LR types appear to prioritize a high-affinity, rigid E2 interaction (− 25.65 kcal/mol) that likely enforces strict copy number control, a ‘lock-and-key’ mechanism suitable for benign persistence. HR types, conversely, appear to have evolved a weaker interface (− 20.95 kcal/mol) that is significantly weaker (Fig. 7A). As indicated by our logistic regression model (OR = 1.72), this instability is strongly associated with oncogenic classification. It is important to emphasize that this logistic regression model is not a standalone clinical diagnostic classifier. Rather, these findings provide a theoretical comparative framework that requires prospective biochemical validation. We hypothesize that the observed thermodynamic instability manifests kinetically as a more rapid off-rate for E2, allowing the HR E1 to transition more aggressively into the active hexameric state. In biological switches, thermodynamic stability is often the primary determinant of residence time [60, 61]. Assuming association rates (kon) for homologous protein interfaces are diffusion-controlled and conserved, we propose that the calculated differences in Gibbs free energy (ΔG) are predominantly driven by the dissociation rate (koff). This serves as a hypothesized proxy for the lifetime of the regulatory complex [62]. To validate this central thermodynamic and kinetic hypothesis, experimental testing using Bio-Layer Interferometry (BLI) or Surface Plasmon Resonance (SPR) is required to map real-time binding kinetics across a representative panel. This panel must include high-risk weak-interface examples (HPV18 or HPV45), low-risk strong-interface examples (HPV6 or HPV11), and the specific outliers identified in this study (HPV16 and HPV61). Although the functional necessity of the core E1 Arg454 and E2 Glu43 salt bridge was previously observed in the crystal structure of the HPV18 complex [25], these physical assays are needed to validate the model and should incorporate comparative mutagenesis at E1 Arg454 and related E2 acidic residues across these clades. Tracking the dissociation rates of these cross-clade variants will provide the empirical proof needed to link computational interface stability with actual complex lifetimes. Additionally, using long-term molecular dynamics simulations to characterize dynamic interface equilibration remains an informative future objective. This biophysical mechanism contrasts sharply with the classic role of the E6 and E7 oncoproteins, whose contribution to oncogenesis is driven by functional differences in targeting host suppressors p53 and pRb [10]. For E1, its association with risk categories appears to be structurally encoded in the biophysics of its regulatory checkpoint. However, the influence of shared evolutionary history becomes apparent during Leave-One-Clade-Out cross-validation accuracy, which dropped to 75%. This reduction in performance reflects a limitation for lineage-blind classification models when entire evolutionary branches are omitted from model training. Intriguingly, this outcome likely indicates that E1-E2 thermodynamic properties may be evolutionary features that are deeply conserved within specific lineages. Because clinical oncogenicity is structurally clustered within the Alpha-9 and Alpha-7 lineages, these findings suggest that the biophysical classification signal is closely linked to the evolutionary history of these clades, representing an interesting avenue for further structural and phylogenetic studies.

Within this framework, it is important to address the thermodynamic behavior of two notable outliers in our dataset: HPV16 and HPV61. While HPV16 is the most universally prevalent high-risk genotype, the E1-E2 interface for the reference genome evaluated in this study (HPV16REF, corresponding to the A1 sublineage) exhibited a binding energy (-25.54 kcal/mol) more characteristic of low-risk types. Comprehensive genomic analyses have demonstrated that HPV16 is not a monolith but comprises multiple sublineages with vastly different oncogenic penetrance [63]. Epidemiological data indicate that the A1 sublineage represents a moderate oncogenic baseline. Conversely, non-European clades such as the D lineage (D2, D3) and the A4 sublineage are hyper-carcinogenic and drive a disproportionate burden of severe adenocarcinomas and early-onset disease [64].

To analytically address this intratypic diversity and the outlier status of the HPV16 reference genome, our ensemble modeling (detailed in Sect. "E1-E2 binding energy correlates with clinical risk profiles") demonstrated that averaging multiple structural seeds weakens the reference A1 binding energy. More importantly, evaluating the hyper-carcinogenic non-European sublineages (D2 and D3) revealed substantially weakened interface stabilities that closely align with the thermodynamic profile of other mucosal high-risk genotypes. While intratypic variation appears to modulate interface thermodynamics, the reference HPV16 genome represents an exception to the rule with stronger binding despite HR status. It is necessary to account for this intratypic structural diversity when utilizing in silico models to correlate oncogenic risk. On the other end of the spectrum, the low-risk genotype HPV61 presented an anomalously weak binding energy (-18.25 kcal/mol) compared to its phylogenetic peers. This deviation is likely an artifact of structural modeling limitations rather than a true biological feature. As noted in our quality control assessment, the HPV61 E1-E2 complex was the only structure to fall below the optimal confidence threshold (ipTM = 0.77). Because the FoldX algorithm is highly sensitive to precise atomic packing [65], minor inaccuracies in the predicted three-dimensional coordinates can artificially destabilize the computed binding energy and skew thermodynamic classifications.

Placing the papillomavirus E1-E2 matchmaker framework into a broader virological context highlights unique evolutionary adaptations in replication initiation. While small DNA viruses such as the Polyomaviridae, Parvoviridae, and Geminiviridae encode autonomous single-chain initiators that merge origin recognition and helicase translocation, papillomaviruses split these functions between two independent proteins. Autonomous initiators like the simian virus 40 large T antigen [66], parvovirus NS1 [67], or geminivirus Rep [68] combine these functions within a single multidomain protein. This high level of mechanistic consolidation reduces external regulatory control and drives aggressive replication that can prove directly cytopathic to the host cell. In contrast, the binary E1-E2 recruitment network acts as a barrier to premature replication, creating a distinct regulatory checkpoint that prevents unguided helicase activity and supports a stable, non-cytopathic latent phase [69].

From a viral fitness perspective, this weaker E1-E2 binding in high-risk types may represent an evolutionary adaptation that enhances replication dynamics within the restrictive environment of mucosal epithelia. By shedding the E2 repressor more readily, high-risk viruses can rapidly assemble the active hexameric helicase, accelerating genome amplification and increasing the likelihood of viral persistence and eventual oncogenesis. It is important to acknowledge alternative explanations for these observations, such as the possibility that these stability differences are merely a consequence of overall sequence divergence rather than a direct mechanistic cause of oncogenicity. However, our control analysis, which shows no significant difference in E1–E1 hexamer stability, supports the interpretation that this effect is specific to the E1–E2 regulatory interface. This control indicates that the observed interface variability is specifically selected for at the E1-E2 regulatory checkpoint rather than being a general artifact of viral evolution. Furthermore, the thermodynamic stability of the E1-E2 interface is just one component of a multi-factorial process. The progression to malignancy is highly complex and is primarily driven by viral persistence, tissue tropism, immune evasion, genomic integration patterns, and the classic transforming effects of the E6 and E7 oncoproteins on p53 and pRb. As our lab’s previous work has shown, the E2 protein binds to four distinct E2-binding sites within the Long Control Region with highly variable, sequence-dependent affinities. This specific binding hierarchy dictates the sequential activation of DNA replication and the necessary repression of the E6 and E7 oncogenes [70]. Ultimately, these variations in E1-E2 interface stability likely operate in tandem with these E2-DNA binding dynamics to shape the overall clinical trajectory of the infection.

The indispensable and highly conserved nature of the E1 protein makes it a possible therapeutic target. The detailed structural and functional understanding synthesized in this report highlights several distinct plausible strategies for inhibiting its function. First, the ATPase active site represents a viable focus for broad-spectrum antiviral development, a strategy supported by recent virtual screening efforts targeting this pocket [39]. Small molecules acting as competitive or allosteric inhibitors could theoretically disable the helicase motor across divergent genotypes due to core sequence conservation. While these invariant motifs highlight a clear therapeutic opportunity, translating these structural insights requires caution. Because explicit small-molecule docking and selectivity comparisons against human host AAA + ATPases were outside the scope of this study, these interaction pockets remain predictive targets until verified by empirical cross-reactivity profiling. Second, the E1-E2 protein–protein interaction interface represents a unique, virus-specific target. Inhibitors that disrupt this interaction would prevent the initial recruitment of E1 to the origin, halting replication before it even begins [71]. Consequently, an immediate future direction of this work involves the de novo design of E1-inhibitory peptides targeting these highly conserved interfacial residues to therapeutically disrupt the E1–E2 complex. Finally, E1’s status as a conserved foreign antigen makes it an excellent candidate for immunotherapy. Therapeutic vaccines designed to elicit a T-cell response against E1 could provide broad protection and, crucially, may simultaneously unmask the virus to the immune system by neutralizing E1’s immunosuppressive functions [72, 73]. The observation that E1-specific T-cell responses are correlated with improved clinical outcomes in patients provides strong validation for this approach [74].

Conclusions

The primary barrier to understanding the E1-E2 interplay has been the biochemical difficulty of purifying replication complexes across the phylogenetic spectrum. By utilizing AlphaFold3 and FoldX, we have established a predictive framework that complements traditional biochemical assays by generating a comparative pan-genotype thermodynamic landscape. In this comprehensive in silico investigation, we have delineated the structural and functional landscape of the Human Papillomavirus E1 protein. Through a multi-scale analysis combining phylogenetics, multiple sequence alignment, and structural modeling, we have identified the core functional regions and universally conserved residues that are indispensable for E1’s role as the engine of viral DNA replication. The C-terminal helicase/ATPase domain exhibits profound conservation, particularly in the catalytic motifs responsible for ATP-driven DNA unwinding, while the central DNA-binding domain contains conserved residues essential for dimerization and origin recognition. Our analysis of the E1-E2 interaction interface provides a structural basis for the allosteric regulation that governs the initiation of replication. Crucially, our thermodynamic profiling highlighted a significant biophysical distinction between HPV genotypes: the E1–E2 recruitment interface in High-Risk types is thermodynamically weaker compared to the rigid interaction observed in Low-Risk types. This stability difference was strongly correlated with oncogenic potential (ROC AUC = 0.875), suggesting that thermodynamic instability may mirror regulatory dynamics associated with carcinogenic progression. A limitation of this study is the restricted sample size of 24 genotypes, which is inherently tied to the availability of definitive clinical data. The International Agency for Research on Cancer has only formally evaluated and classified a limited subset of HPV genotypes into definitive risk categories. While the thermodynamic divergence within this established dataset is statistically robust, further experimental validation is required to rule out single-predictor model overfitting as epidemiological classifications expand to include more diverse genotypes. The deep conservation of E1’s essential machinery across diverse and clinically significant HPV types highlights its therapeutic interest for the development of broad-spectrum antiviral drugs and immunotherapies to treat HPV-associated diseases. Future work integrating thermodynamic, kinetic, and sequence-based features will be essential to develop a comprehensive predictive framework for HPV oncogenic risk.

Abbreviations

AAA  + 

ATPases associated with diverse cellular activities

ABG

Alpha–beta-gamma alignment

AF3

AlphaFold 3

AUC

Area under the curve

BLI

Bio-layer interferometry

BPV

Bovine papillomavirus

BSA

Buried surface area

CI

Confidence intervals

CS

Conservation scores

DBD

DNA-binding domain

DDR

DNA damage response

E1

Early protein 1

E1BS

E1 binding site

E2

Early protein 2

E2BS

E2 binding site

HPV

Human papillomavirus

HR

High-risk

IARC

International agency for research on cancer

ipTM

Interface predicted template modeling score

kb

Kilobases

LCR

Long control region

LOOCV

Leave-one-out cross-validation

LR

Low-risk

MSA

Multiple sequence alignment

NES

Nuclear export signal

NLS

Nuclear localization signal

NTR

N-terminal region

OR

Odds ratio

ori

Origin of replication

PaVE

Papillomavirus episteme

PDB

Protein data bank

pLDDT

Predicted local distance difference test

ROC

Receiver operating characteristic

RPA

Replication protein A

SF3

Superfamily 3

SLiMs

Short linear motifs

SPR

Surface plasmon resonance

ssDNA

Single-stranded DNA

Author contribution

SF: Conceptualization, Analysis, Methodology, Visualization, Writing – original draft and editing. EEB-F: Conceptualization, Supervision, Funding acquisition. SBB: Conceptualization, Methodology, Supervision, Resources, Writing – review and editing, Funding acquisition.

Funding

This publication was made possible through support from the Angela Santoro Summer Undergraduate Research Award as well as the Delaware INBRE Summer Research Program, funded by the National Institute of General Medical Sciences (NIGMS) of the National Institutes of Health (P20 GM103446) and the State of Delaware. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Data availability

The structural models, sequence alignments, and custom analysis scripts utilized in this study are publicly available via Zenodo (10.5281/zenodo.19130708). All other datasets supporting the conclusions of this article are included within the manuscript and its supplementary files.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.