Unraveling C‑Peptide’s Role in MIDY: A Structural Perspective

Abstract

Proinsulin folding requires dynamic positioning of the C-peptide to guide A- and B-chain alignment and disulfide pairing. Mutant INS-gene-induced diabetes of youth (MIDY) arises when single-residue substitutions disrupt this process. We mapped the conformational free-energy landscapes of wild-type (WT) proinsulin and seven MIDY variants using metadynamics and molecular dynamics simulations. WT exhibits a deep free-energy minimum at compact conformations. In contrast, MIDY mutants display a continuum of destabilization: E­(A4)K retains near-WT stability, Akita (C­(A7)­Y), V­(B18)­A, and R­(Cpep + 2)C show moderate loss of the native basin, while H­(B5)­D, L­(A16)­P, and Y­(B26)C collapse the closed–open barrier and populate misfolded open states >50% of the time. Structural analyses reveal that WT and E­(A4)­K preserve robust A–C docking, with the C-peptide flexibly engaging the A-chain groove. Destabilizing mutants progressively erode these native A–C contacts while forming compensatory, non-native B–C interactions. Per-residue energy decomposition highlights the loss of canonical salt bridges and emergence of aberrant electrostatic and hydrophobic hot spots, correlating with the collapse of the folding free-energy barrier. Secondary-structure analysis further shows that mutants rigidify the normally disordered C-peptide, increasing helical or strand propensity in a mutation-specific manner. Collectively, these findings establish a continuum from near-native stability to overt misfolding, mechanistically linking single-site mutations to altered folding landscapes and aggregation risk in MIDY. The results highlight the C-peptide as a dynamic linchpin of proinsulin folding and suggest that restoring its flexible docking could provide a therapeutic avenue.

Introduction

C-peptide, a 31-amino acid polypeptide, is a molecule of remarkable structural versatility and functional significance, playing a key role in the proteostasis of pancreatic β-cells. Far from being a mere byproduct, its primary function is closely linked to insulin biosynthesis, where its precise positioning and structural organization facilitate the proper folding of proinsulin. , This involves guiding the formation of the three essential disulfide bonds (interchain B7–A7, interchain B19–A20, and intra-A-chain A6–A11) necessary for mature insulin’s structure, particularly by aiding in the alignment of folding intermediates. , This essential role is what makes mutations in the INS gene relevant to diseases like mutation-induced diabetes of youth (MIDY), which features proinsulin misfolding. , Indeed, while the insulin core within proinsulin adopts a conformation similar to that of mature insulin, it exhibits greater flexibility due to the presence of the C-peptide, which is crucial for its role as a folding guide. ,, This function, however, would mean that the C-peptide does not participate in insulin receptor binding, making it dispensable after proinsulin is processed into insulin. Beyond this foundational intracellular function, C-peptide emerges as a biologically active signaling molecule. A defining structural characteristic of C-peptide is its dynamic nature, meaning that under physiological conditions, it does not adopt a single, stable three-dimensional tertiary structure; instead, it exists as an ensemble of conformations. This inherent flexibility underpins the C-peptide’s diverse signaling capabilities, allowing it to bind to an as-yet-uncharacterized G-protein-coupled receptor. This binding can initiate intracellular cascades that contribute to beneficial effects in mitigating diabetic complications. −

The “flexible” nature of C-peptide is why two-dimensional (2D) nuclear magnetic resonance (NMR) spectroscopy has been the primary tool for its structural elucidation (PDB ID: 1T0C), capturing its local and transient features. Despite its overall random coil nature, C-peptide exhibits specific, albeit transient, local substructures. These include a β-turn in its N-terminal region and, crucially, a well-defined β-turn within its C-terminal pentapeptide (EGSLQ), a key “active site” responsible for its stereospecific binding to cell surface receptors and subsequent activation of signaling pathways. , The C-peptide is also notable for its ability to prevent proinsulin from amyloid fibrillation. Furthermore, evolutionary analysis highlights a remarkable evolutionary and structural flexibility of the C-peptide, exhibiting much higher sequence variation than the tightly conserved A- and B-chains, which influences its structural role and contributes to fewer disease mutations.

The genetic mutations in the INS gene (predominantly in the sequences encoding A- and B-chain) are the central problem in MIDY, which leads to a misfolded proinsulin that accumulates in pancreatic β cells’ endoplasmic reticulum (ER), causing cellular stress in diabetes. , While the link between proinsulin mutations and misfolding is well established, the specific role of the C-peptide in this pathological process is not fully understood. Existing structural data on C-peptide primarily focus on its isolated form, a random coil state that may not fully represent its behavior within the full-length, folding proinsulin molecule. The AlphaFold model of proinsulin, which shows varying degrees of helicity in the C-peptide across different species (Figure ), suggests that its conformation is more dynamic and structurally defined within the proinsulin context than previously thought. Therefore, we sought to use molecular dynamics (MD) simulations to elucidate the structural dynamics of the C-peptide within both wild-type (WT) proinsulin and various MIDY mutants. Our goal was to understand how these disease-causing mutations alter the proinsulin folding landscape and, critically, how this change directly impacts the conformational behavior of the C-peptide.

1.

Proinsulin models from various species, generated by AlphaFold. The B-chain, A-chain, and C-peptide are colored red, blue, and gray, respectively. Note the varying degrees of helicity present in the C-peptide.

Our study revealed that under normal conditions, interactions between C-peptide and the A or B chain of proinsulin are essential in maintaining C-peptide’s highly unstructured state; however, in the presence of MIDY mutations, proinsulin adopts a more open conformation. This structural alteration leads to weakened interactions between the C-peptide and the A- and B-chains, consequently forcing the C-peptide to adopt a local secondary structure, primarily the α-helix. This finding offers a crucial molecular-level understanding of how genetic mutations impacting the proinsulin folding landscape can directly influence the conformational behavior of the C-peptide, thereby linking its dynamic structure to the complex pathogenesis of diabetes.

Methods

Metadynamics Simulations

Folded structures of WT proinsulin and MIDY mutants were predicted using AlphaFold 3.0 with the multimer setting. As expected, the C-peptide was modeled with low confidence due to its inherent flexibility, highlighting the importance of explicitly probing protein dynamics. Due to the highly disordered nature of the C-peptide, the backbone root-mean-square deviation (RMSD) for the five Alphafold3 models, for each of the variants, are as follows: WT (7.763 A), E­(A4)K (7.553 A), H­(B5)­D (7.057), L­(A16)­P (8.944 A), R­(Cpep + 2)C (8.103 A), V­(B18)­A (7.287), and Y­(B26)C (6.787 A). The Alphafold3 starting structural models can be downloaded from https://github.com/sranga88/proinsulin-MIDY-AF3.

WT proinsulin and seven MIDY mutants, including H­(B5)­D, G­(B8)­V, V­(B18)­A, Y­(B26)­C, R89C [R­(Cpep + 2)­C], E­(A4)­K, C­(A7)Y [henceforth referred to as Akita] and L­(A16)P were studied. Here is an example of MIDY mutant nomenclature: V­(B18)­A indicates that Val at the 18th position on the B-chain was substituted with Ala.

To evaluate the stability of the AlphaFold-predicted conformations and to map alternative states accessible to proinsulin, we performed well-tempered metadynamics simulations. − Metadynamics is an enhanced-sampling approach in which a history-dependent biasing potential is added to accelerate exploration of metastable conformations and reconstruct the free-energy surface along a chosen collective variable (CV). The radius of gyration (R _g) of the protein backbone (residues 1–86) was selected as the CV to characterize global compactness. R _g was computed by using the Cα atoms. Sampling was confined between 5 and 20 Å through soft harmonic walls with lower and upper force constants of 2.0 and 3.0 kcal·mol^–1·Å^–2, respectively, preventing both excessive expansion and unrealistic compaction. This range allowed access to compact conformations (R _g ≈ 12 Å) as well as more extended states (R _g > 14 Å). Simulations were run using the Colvars module in NAMD 2.3 with the CHARMM36m force field. In the well-tempered formulation, Gaussian hills of 0.3 kcal·mol^–1 height and 1 Å width were deposited every 500 steps, with a bias factor corresponding to an effective bias temperature of 1590 K. This ensured gradual flattening of the free-energy surface while maintaining efficient exploration. Colvars and bias potentials were recorded every 500 steps, with the evolving bias saved as both hills trajectories and free-energy files for subsequent reconstruction of the potential of mean force (PMF). Each trajectory was extended to 200 ns (200,000,000 steps with a 2 fs time step), sufficient to achieve convergence of the PMF along R _g.

All systems consisted of the solvated protein in explicit TIP3P water boxes under periodic boundary conditions. Temperature was maintained at 310 K via Langevin dynamics with a damping coefficient of 3 ps^–1 applied to non-hydrogen atoms, while pressure (1 atm) was controlled by the Langevin piston method with a piston period of 100 fs and decay time of 50 fs. Non-bonded interactions were treated with a 12 Å cutoff, a 10 Å switching function, and a 13.5 Å pairlist distance. Long-range electrostatics were handled by using particle mesh Ewald (PME) with 1.0 Å grid spacing. All bonds involving hydrogen were constrained with the rigidBonds option, enabling a 2 fs time step. Non-bonded interactions were updated every step, and PME calculations were performed every second step. The converged free-energy surfaces were used to identify metastable states, with the relative free energy of each state inferred from the negative of the accumulated bias. Representative structures from free-energy minima were used for hydrogen-bond analysis and visualization in VMD.

Protein Aggregation Prediction

Protein aggregation propensity of human proinsulin (sequence below) was evaluated in silico using three sequence-based algorithms: Aggrescan, AggreProt, and Waltz. The amino acid sequence was obtained from UniProt and formatted in a single-letter code before analysis. Single-point mutations were introduced by manually editing the wild-type amino acid sequence to generate the corresponding MIDY mutant protein sequence.

Human proinsulin amino acid sequence (excluding the signal peptide) used for aggregation prediction:

FVNQHLCGSHLVEALYLVCGERGFFYTPKTRREAEDLQVGQVELGGGPGAGSLQPLALEGSLQKRGIVEQCCTSICSLYQLENYCN

Aggrescan was used to predict aggregation “hot spots” along the sequence, returning residue-level aggregation scores and contiguous aggregation-prone regions according to its intrinsic aggregation propensity scale. AggreProt was employed to generate a per-residue aggregation profile and to identify aggregation-prone segments using its deep-learning-based predictor trained on experimentally characterized peptides. On the other hand, Waltz was applied to the same sequence to predict short amyloid-forming motifs, providing scores for hexapeptide windows and classifying segments as amyloidogenic or nonamyloidogenic. Default settings were used for all three web servers, including organism-agnostic parameters and standard thresholds for defining aggregation-prone or amyloidogenic regions. Predicted aggregation-prone regions and scores from the three methods were exported and compared qualitatively to assess the consensus and method-specific differences along the sequence.

Results

MIDY Mutations Show a Disruption in the Compact State in the Free-Energy Landscape

In this study, we have investigated the structural dynamics of the WT proinsulin and seven MIDY mutants (mentioned earlier in the Methods section). These mutants have previously been reported to show distinct (mis)­folding behavior.

The one-dimensional potentials of mean force (PMFs) projected onto the radius of gyration (R _g) of proinsulin (Figure ) reveal a clear spectrum of destabilization across the seven MIDY mutants relative to WT. For each variant, we define ΔG _{closed→open} = G _open – G _closed, where G _closed and G _open are the free energies at the minima of the compact and open basins, respectively (see blue and red dots in the free-energy landscape in Figure ). We further define the mutation-induced change relative to WT as

$$\mathrm{\Delta }{G}_{\mathrm{closed}\to \mathrm{open}}={\mathrm{\Delta }G}_{\mathrm{closed}\to \mathrm{open}}^{\mathrm{mutant}}-{\mathrm{\Delta }G}_{\mathrm{closed}\to \mathrm{open}}^{\mathrm{WT}}$$

We have also unified the notation throughout the manuscript to consistently use the arrow form ΔΔG _{closed→open}, and removed the ambiguous “ΔΔG _{closed–open}” notation. Thus, it is now unambiguous that ΔG _{closed→open} = G _open – G _closed and that ΔG _{closed→open} = ΔG _{closed→open} – ΔG _{closed→open} always denotes the difference in this quantity between a mutant and WT.

2.

Free-energy landscape for protein stability. The potential mean force curves (PMFs) from metadynamics simulations for WT proinsulin (A) (blue curve) and seven MIDY mutants (B–H). All variants show a defined minima corresponding to a compact state (R _g ∼ 12.5 Å). The WT shows a defined deep minimum at 12.5 Å. The mutants, on the other hand, show a relatively destabilized compact state and a greater likelihood to sample the more open configurations. E­(A4)K shows behavior closer to the WT, while H­(B5)­D, L­(A16)­P, and Y­(B26)C show a significant destabilization and access the open state with equal or greater likelihood than the compact states.

To further quantify the equilibrium population of expanded conformations, we define the probability of the open state, P(open), from the one-dimensional PMFsF(R _g). After shifting each PMF so that its global minimum is zero, we evaluate Boltzmann weights w(R _g) = exp­[−β·F(R _g)], with β = 1/k _B T, and integrate these over radii of gyration corresponding to open conformations (here, R _g ≥ 14 Å).

$$P(\mathrm{open})=\frac{{\int }_{R_{\mathrm{g}}\ge 14\AA }{\mathrm{e}}^{-\beta F(R_{\mathrm{g}})}\mathrm{d}{R}_{\mathrm{g}}}{{\int }_{\mathrm{all}{R}_{\mathrm{g}}}{\mathrm{e}}^{-\beta F(R_{\mathrm{g}})}\mathrm{d}{R}_{\mathrm{g}}}$$

where β = 1/(k _B T). The integrals are evaluated numerically using the trapezoidal rule on the PMF grid.

In Figure A–G, we compare the features of the conformational landscape of the WT with those of the MIDY mutants. In WT, a deep, narrow minimum at R _g ≈ 12.5 Å lies ∼27 kcal·mol^–1 below the unfolded baseline (R _g > 16 Å), and a steep barrier of >20 kcal·mol^–1 separates it from higher-R _g conformers. The free-energy profile for E­(A4)K almost entirely overlaps with that of WT, retaining a compact minimum at R _g ≈ 12.5 Å and a barrier only ∼2 kcal·mol^–1 lower. The open state remains sparsely populated, indicating near-native stability. In the case of the Akita mutant, the compact well shifts slightly inward (R _g ≈ 12.3 Å) and shallows significantly, with ΔG _{closed→open} showing a more than one-third reduction. Two emergent local minima at R _g > 14.5 reflect increased sampling of semiopen states, and a lower barrier to open states compared to the WT and E­(A4)­K. V­(B18)­A exhibits a comparable ΔG _{closed→open} (−7.35 kcal·mol^–1) but a more pronounced lowering of the barrier (10 kcal·mol^–1). A shallow secondary well at R _g > 15 Å emerges, indicating more frequent sampling of the open configurations. The free-energy landscape for R­(C-pep2)C displays a broadened compact basin spanning R _g ≈ 11.5–13 Å and a local minimum corresponding to an open state at around 15 Å. The barrier is depressed significantly, to 11 kcal·mol^–1, compared to 46.65 for E­(A4)­K, suggesting a higher propensity of open-state excursions. In the case of H­(B5)­D, the native well is markedly flattened and broad (R _g ≈ 11–14 Å), with only a minor barrier (∼5 kcal·mol^–1). This mutant interconverts freely between compact and open configurations, sampling open states ∼40% of the time. L­(A16)P exhibits a distinct minima at R _g ≈ 12.5 and 14.5 Å, and the barrier is effectively abolished. Extended conformers are highly favored, indicating a strong propensity for open states. The mutant Y­(B26)C shows a deep native minimum but also a nearly isoenergetic open well at R _g ≈ 17 Å. The shallow interwell barrier (∼7 kcal·mol^–1) permits ease of sampling of the open state (∼65% population).

Overall, the eight PMFs underscore a continuum from near-WT stability (E4AK) through intermediate destabilization (Akita, V­(B18)­A, R­(CPEP + 2)­C) to overt misfolding phenotypes (H­(B5)­D, L­(A16)­P, and Y­(B26)­C). Thus, single amino acid substitutions in proinsulin variably remodel its free-energy landscape, from modest modulation of folding landscape to wholesale collapse of the native basin.

We quantified the relative stability of the compact state by extracting ΔG _open – ΔG _closed from the PMFs for the MIDY-associated variants (Figure ). This result shows that a gradient of ΔG values stratifies the mutants into three classes:

• 1.WT-like stability: E­(A4)­K
• 2.Moderate destabilizers: Akita, V­(B18)­A, R­(Cpep + 2)­C
• 3.Severe open state propensity: H­(B5)­D, L­(A16)­P, Y­(B26)­C

3.

Stability of the compact state. The free-energy difference between the open and closed configurations of the WT and seven MIDY mutants under study. The WT shows the most stable compact state along with E­(A14)K and Akita. L­(A16)­P, Y­(B26)­C, and H­(B5)­D show a complete destabilization of the compact state.

As the ΔG diminishes, the equilibrium shifts increasingly toward open, misfolded conformations, providing a quantitative link between single-site perturbations and the propensity for proinsulin misfolding in MIDY.

Native, Transition, and Open-State Structures for MIDY Mutants Show Varying Degrees of Helicity Domain Packing

In Figure , we show representative structures from the free-energy basins for WT proinsulin and MIDY mutants, with the C-peptide in gray and the A- and B-chains in blue and red, respectively. The ensemble of states and the structural changes in the C-peptide that underlie them vary dramatically between the WT and each mutant, reflecting their graded destabilization of the native fold shown in Figures and . The WT proinsulin populates a single, well-defined compact state. The C-peptide is largely unstructured and docks snugly into the A–B interface, making extensive hydrophobic and backbone hydrogen-bond contacts. No semiopen or open conformers were observed at appreciable populations, consistent with its deep PMF minimum (Figure ) and large ΔG _{closed→open} (Figure ). E4AK behaves almost identically to WT in its native conformer except for the helical tendency in the C-peptide and a preservation of the domain packing. A minor subpopulation (open in Figure ) in which the N-terminal turn of the gray helix peels slightly away from the blue A-chain, although the remainder of the helix remains docked against both chains. In the H­(B5)­D native structure, the gray helix of the C-peptide is visibly shortened by one turn at its N-terminus but still engages the red B-chain. In the H­(B5)­D transition state, only two turns of the C-peptide helix remain with substantial unraveling and partial detachment. In its fully open state, H­(B5)­D shows a significant loss of C-peptide helicity, now an extended gray loop that no longer interacts with the blue A-chain or red B-chain. In L­(A16)­P, the closed state shows a characteristic kink in the gray C-peptide with the helix docked between A and B chains. The native state in R­(Cpep + 2)C shows nativelike domain packing with very little helical tendency in the C-peptide. In V­(B18), A native configuration, the gray C-peptide helix docks similarly to WT but with slightly looser packing near the red B-chain. Overall, across mutants, the C-peptide shows a large structural variability, with varying degrees of secondary-structure formation along with A and B-chain docking in the variants.

4.

Structures corresponding to the compact (native) state, transition, and open states for the WT and different MIDY mutants, with the C-peptide in gray, the A- and B-chains in blue and red, respectively. The conformational variation between the different states is an outcome of structural changes to the C-peptide.

MIDY Mutants Show Selective Disruption of the A–C Interface and Compensatory B–C Pairing

To probe how MIDY mutations reshape proinsulin’s interaction network, we counted all residue–residue contacts with >5% occupancy over the 1-μs equilibrium ensembles in three domain pairs in A–B (Figure , left), A–C (Figure , middle), and B–C (Figure , right). WT is shown in blue; mutants are in red. Over a 200 ns trajectory, a 5% occupancy corresponds to an average cumulative contact lifetime of: 0.05 × 200 ns = 10 ns. Contacts persisting for ≥10 ns are long-lived relative to the intrinsic picosecond–nanosecond time scale of thermal fluctuations in peptides. In contrast, contacts that occur <5% of the time correspond to <10 ns of total lifetime, typically appearing as very brief subnanosecond or nanosecond encounters scattered throughout the trajectory. These short-lived events reflect transient collisions and local breathing motions rather than stable structural interactions that influence global folding thermodynamics.

5.

Interdomain contacts for WT (blue) and different MIDY mutants (red bars). A-chain–B-chain contacts show negligible variation between mutants and the WT. A-chain–C-peptide interactions show a significant reduction in the mutants as compared to the WT proinsulin. WT and E­(A4)K show sparser B-chain–C-peptide interactions compared to those of the rest of the mutants.

As seen in Figure , the quantification of total interchain contacts reveals a clear correlation between interface disruption and the degree of ensemble destabilization inferred from the PMFs. In WT, we observe 37 A–C contacts, 11 B–C contacts, and 12 A–B contacts, forming a dense interchain network that stabilizes the compact state. L­(A16)­P, which exhibits the most dramatic PMF shift (minimum displaced to R _g ≈ 19.5 Å, ≈99% probability of R _g ≥ 14 Å), retains only 18 A–C contacts (−51%) and 5 B–C contacts (−55%), indicating extensive interface disruption, while A–B contacts are slightly increased (13 A–B contacts). H­(B5)­D also shows major losses, with 16 (−57%) resulting in their elevated open-state probabilities (≈42%). R­(CPEP + 2)C has substantially reduced B–C contacts (5, −55%) and a more than 50% loss in A–C contacts, also resulting in a destabilized PMF.

E­(A4)K is unique in maintaining 32 A–C contacts (−14%) and even increasing B–C contacts to 15 (+36%), aligning with its deep compact-state PMF and negligible probability of visiting R _g ≥ 14 Å. These quantitative results underscore that A–C contact loss is the dominant driver of open-state bias, with cumulative A–C counts inversely related to open-state probability, while mutants that preserve or enhance these contacts (E­(A4)­K) stabilize the compact ensemble. Paradoxically, some MIDY variants might retain or form additional B–C contacts, many of which could be non-native hydrophobic or spurious electrostatic interactions. These compensatory contacts likely stabilize non-native conformers once the primary A–C interface is compromised. Overall, the analysis of interdomain contacts suggests that the A–B core remains stable, MIDY mutations erode the A–C docking interactions in proportion to their free-energy destabilization, and concurrently promote aberrant B–C contacts. This rewiring of interdomain contacts underlies the graded shift from a tightly folded native state toward heterogeneous, misfolded ensembles in MIDY proinsulin variants.

Mutation-Specific Rewiring of the Proinsulin Contact Network

To quantitatively characterize how MIDY mutations perturb the interchain packing of proinsulin, we computed residue–residue contact maps from the full simulation ensembles of the WT and each mutant. Contacts were defined based on hydrogen-bond or heavy-atom proximity criteria, and their occupancies were normalized over the entire trajectory, yielding a per-residue-pair contact frequency expressed as a percentage of simulation frames. Only contacts with occupancies greater than 5% were retained for visualization to focus on persistent structurally meaningful interactions.

The resulting maps provide a dense matrix representation of all B-chain, C-peptide, and A-chain interactions, with circle size and color proportional to occupancy, allowing a direct visual comparison of contact patterns across variants. In the WT, the maps reveal a prominent cluster of high-occupancy A–C contacts, particularly between C-chain residues 58–66 and A-chain residues 74–84 (Figure , WT), which together form a structural “clamp” stabilizing the compact conformation. Comparing these maps across mutants highlights both subtle shifts and dramatic losses of these anchoring interactions, with some variants showing large-scale destabilization of the A–C interface and others displaying compensatory rewiring in neighboring regions. For further insight and a detailed quantification of the loss/gain of contacts with respect to the WT, we generated Δ-contact maps (mutant–WT) to highlight the most strongly lost (blue) and gained (red) contacts, enabling a residue-level view of network remodeling (Supporting Figures S1–S3). In these Δ-contact maps, a negative value represents destabilization or lower occupancy of a contact with respect to WT, while a positive value represents an increased occupancy. These quantitative analyses of contact frequency form the basis for the subsequent thermodynamic interpretation, in which we examine how disruption or reinforcement of specific interchain interactions correlates with the relative free energy of compact versus open states in the PMF profiles.

6.

WT and mutant residue–residue contact maps are shown for all variants, with circles sized and colored by occupancy or occupancy change. Panels (A)–(H) contain plots for the WT and MIDY mutants. Chain boundaries (B-chain: residues 1–29, C-peptide: 30–65, A-chain: 66–86) are indicated by dashed lines. WT maps reveal a dense A–C interface (marked in a red box), particularly involving C-chain residues 58–66 and A-chain residues 74–84. Mutants exhibiting strong destabilization in their PMFs (e.g., L­(A16)­P, Y­(B26)­C, and H­(B5)­D) show a weakened A–C clamp and elevated open-state populations. E­(A4)K and Akita compensate for A–C losses by rewiring new A–C/B–C contacts, preserving compact packing and maintaining a deep free-energy minimum. V­(B18)­A uniquely shows substantial A–B contact loss (C71­(A)–L6­(B)) in addition to A–C disruption, while R­(CPEP + 2)C gains alternative A–C contacts near A74 and A79 that partially rescue packing.

L­(A16)­P, from the PMF profiles, shows the highest propensity to access open states, with a P(open) ≈ 98.1%. This variant shows a large dismantling of the A–C clamp (Figure , L­(A16)­P), including a loss in occupancy of R65­(C)–Y84­(A) with respect to the WT (−62.7% with respect to WT, Supporting Figures S1–S3). Similarly, frequent A–C anchors in WT, such as L58­(C)–S77­(A), are reduced by 40.9% and E69­(A)–G49­(C) by 37.7% (Supporting Figures S1–S3). Some compensatory links appear (e.g., E69­(A)–K64­(C) +44.6%; R65­(C)–T27­(B) +35.0%; R65­(C)–P28­(B) +34.2% in Supporting Figures S1–S3), but these are dispersed and often involve the C-terminal C-peptide edge, which might not be as effective as a clamp stabilizing the compact state. The Δ-maps and the big A–C count drop explain the PMF’s dramatic shift toward expanded conformations. Below, we discuss the drop (or gain) in frequencies of key interactions for each mutant with respect to the WT. For example, a −5% for any contact in the following sections means a lowered frequency of the contact in the mutant with respect to the WT, suggesting destabilization of the contact in the mutant.

Y­(B26)C shows a strongly open ensemble, with P(open) ≈ 65.2%, ranking it among the most destabilized variants after L­(A16)­P. The dismantling of the A–C clamp is evident (Figure ), with large losses in L58­(C)–S77­(A) (−40.9%), E69­(A)–G49­(C) (−37.7%), and N83­(A)–T30­(C) (−36.0%), which are critical compact-state anchors in WT. Although some compensatory contacts appearnotably E33­(C)–K29­(B) (+23.8%), Q38­(C)–G66­(A) (+20.9%), and T27­(B)–T30­(C) (+17.6%), they are scattered and represent a rewiring rather than restoration of the A–C interface (Supporting Figure S3). This diffuse compensation fails to re-establish a cohesive clamp, explaining the large shift toward expanded conformations in the PMF.

H­(B5)­D is similarly prone to more open conformations, with P(open) ≈ 42.2%, and shares the same hallmark A–C contact losses (L58­(C)–S77­(A) −30.87%; E69­(A)–G49­(C) −36.45%; N83­(A)–T30­(C) −28.2%). Several new contacts form, such as E35­(C)–S9­(B) (+31.2%, B–C), E59­(C)–S74­(A) (+27.3%, A–C), and K29­(B)–V39­(C) (+19.0%, B–C), partially patching B–C edges but leaving the primary A–C glue insufficiently restored (Figure and Supporting Figure S1–S3). This results in a shallower compact basin and a PMF shifted toward more frequent opening events, consistent with the Δ-map signature of a weakened A–C network (Supporting Figures S1–S3). R­(CPEP + 2)­C, while considerably more compact than Y­(B26)C and H­(B5)­D, still displays a measurable shift toward opening with P(open) ≈ 0.42%. This variant loses several of the strongest A–C stabilizers, including R65­(C)–Y84­(A) (−62.7%), L58­(C)–S77­(A) (−32.8%), and E69­(A)–G49­(C) (−24.8%). Importantly, R­(CPEP + 2)C compensates by forming a new set of A–C contacts clustered around A-chain residues 74–79e.g., E43­(C)–S74­(A) (+35.6%), S52­(C)–Y79­(A) (+18.5%), and G46­(C)–S74­(A) (+5.3%)effectively rerouting the A–C interface to a different locus. This partial rescue is consistent with its moderately destabilized but compact PMF profile.

V­(B18)­A shows only mild opening (P(open) ≈ 0.11%) but exhibits an interesting pattern of losses and gains. It loses the canonical A–C anchors (L58­(C)–S77­(A) −26.9%; E69­(A)–G49­(C) −23.9%) and uniquely disrupts a strong A–B contact (C71­(A)–L6­(B) −56.7%), which may loosen the N-terminal A–B junction (Figure and Supporting Figures S1–S3). New contacts emerge, particularly G49­(C)–Y79­(A) (+37.3%), R65­(C)–T27­(B) (+36.4%), and G66­(A)–T30­(C) (+29.7%), partially stabilizing the system. Nevertheless, the combined A–B and A–C reductions leave the compact state less stable, explaining the modest elevation in open-state probability.

Akita remains mostly compact, with P(open) ≈ 0.12%, despite losing the usual A–C stabilizers (L58–S77, E69–G49, N83–T30; in the range of −36% to −41%). This variant compensates robustly, introducing high-occupancy B–C and A–C contacts, such as E33­(C)–K29­(B) (+49.4%), E69­(A)–K64­(C) (+31.8%), and N3­(B)–G47­(C) (+10.4%). The Δ-map reveals that while the identity of contacts is shifted, their number and geometry are largely restored, maintaining a deep compact-state free-energy well and keeping the PMF close to the WT.

E­(A4)K is the most compact-like of all variants, with an essentially negligible P(open) ≈ 10–8. Although it loses some of the canonical WT interactions (L58­(C)–S77­(A) −22.8%; E69­(A)–G49­(C) −29.7%; and an A–B F25­(B)–Y84­(A) −38.0%), E­(A4)K more than compensates by gaining a dense network of A–C and B–C interactionsincluding R32­(C)–Y26­(B) (+18.9%), Q38­(C)–V68­(A) (+17.6%), and E35­(C)–K69­(A) (+16%), which reinforce interchain packing (Figure and Supporting Figures S1–S3). This rewiring yields a PMF that is indistinguishable from WT, explaining why its compact ensemble remains overwhelmingly favored.

Across mutants, loss of the A–C clamp (especially contacts centered on C-chain 58–66 with A-chain 74–84 and A69 with C49 and 30) is the strongest predictor of destabilization toward open states. Mutants like L­(A16)­P, Y­(B26)­C, H­(B5)­D lose multiple high-occupancy A–C bonds without forming equally stabilizing replacements, yielding large positive Δ-maps (gains) that are scattered and negative Δ-maps (losses) at canonical anchors, consistent with high P(open) and shallow compact basins (Figure and Supporting Figures S1–S3).

In contrast, E­(A4)K (and to a lesser degree Akita) replaces lost A–C/B–C links with coherent, compensatory networks that preserve cross-chain cohesion and thus maintain compact PMFs. V­(B18)­A and R­(CPEP + 2)C fall in betweenboth exhibit the signature A–C losses but acquire specific A–C/B–C gains that partially rescue packing, leading to modest open-state populations. In short, the pattern (which identifies the critical A–C bonds lost and whether equally strong, well-placed alternatives replace them) explains the rank order of destabilization observed in the PMFs.

Per-Interaction Stability Contributions Reveal Differential Loss of A–C Contacts and Rewiring of B–C Interactions

To dissect the molecular basis of the shifts in PMFs, we focused specifically on interchain electrostatic interactions. Residues were classified as belonging to the B-chain (1–29), C-peptide (30–65), or A-chain (66–86), and we retained only A–B, A–C, and B–C contacts for analysis, thereby isolating the domain–domain interactions that mediate packing of proinsulin. To quantify how each interaction relates to compact-state stability, we computed the Pearson correlation coefficient between the electrostatic energy of the pair (−E _elec) and the PMF-derived compact-state stability metric: ΔG _{open→closed}. This approach ensures that more stabilizing interactions (more negative energies) correspond to higher values on the correlation axis. The top 20 interchain interactions were selected based on the magnitude of their correlation (|r|), and these were used to construct Figure A (stacked electrostatic contributions across variants) and Figure B (interaction-specific stability correlations).

7.

Key electrostatic interactions as predictors of stability. (A) Stacked bar plot of the mean electrostatic energy contributions (kcal·mol^–1) for the top 20 predictive interchain residue pairs across wild-type (WT) and seven MIDY variants (Akita, E­(A4)­K, R­(CPEP + 2)­C, V­(B18)­A, H­(B5)­D, Y­(B26)­C, L­(A16)­P). Each bar represents the sum of the contributions from the selected pairs, highlighting a net loss of stabilizing interactions in destabilized mutants (e.g., H­(B5)­D, Y­(B26)­C). (B) Pearson correlation coefficients (r) between the mean electrostatic energy of each pair and the compact-state stability metric ΔG _{closed→open}. Positive correlations indicate interactions that stabilize the compact ensemble when strengthened, whereas negative correlations highlight contacts whose strengthening is associated with compact-state destabilization. Together, the panels identify electrostatic hot spots that control proinsulin packing and reveal which contacts act as predictors of folding competence. (C) Visual layout of residues that are involved in interactions in stable variants (blue CPK) or destabilized variants (red CPK).

In Supporting Figure S4, scatter plots of energy versus ΔG _{open→closed} were generated for the interactions that are the strongest predictors of destabilization or stabilization. The combined analysis is summarized in Figure A, which displays the electrostatic contributions of the top 20 predictive interchain contacts across all variants. Figure B shows the Pearson correlations for these pairs, clearly distinguishing those that promote compaction (positive r) from those that favor opening (negative r). The interactions with negative r, such as Q41–Q63, E69–K29, and I67–T27, are stronger (more prevalent) in destabilized mutants such as H­(B5)­D and Y­(B26)C compared to the WT and E­(A4)­K. Interactions such as C72–H5, H5–S74, and C71–L6, on the other hand, are more favorable in the stable variants like the WT and E­(A4)K but became less favorable (electrostatics shifted by +2 to +10 kcal·mol^–1 relative to WT) in destabilized mutants such as H­(B5)­D and Y­(B26)­C, supporting their role as electrostatic anchors between domains. In Figure C, we provide a visual map of the residues involved in contacts that are markers of either destabilized native states (Q41, Q63, E69, and K29 in red) or stabilized compact states (C72, H5, and S74 in blue) with respect to the WT compact structure.

Together, these results demonstrate that MIDY mutations destabilize proinsulin primarily by weakening a small number of key interchain contacts while strengthening alternative ones that stabilize open or misfolded conformations. These findings provide a quantitative, residue-level map of electrostatic hot spots that control proinsulin compaction and suggest a rational strategy for rescue of mutations.

Proinsulin Shows Secondary-Structural Remodeling in MIDY Variants

Secondary-structural analysis of the C-peptide (residues 30–65) from free-energy minima structures from MD reveals that, although WT proinsulin’s linker is overwhelmingly unstructured (∼40% coil, ∼60% turn, virtually no α-helix or β-strand), each MIDY mutant shifts this balance toward defined secondary structure (Figure ). In Akita, the C-peptide acquires roughly 25% α-helix at the expense of coil and turn, indicating modest helix stabilization, while E­(A4)K raises the helical content to about 40%, reducing coil to 25% and turn to 35%, reflecting enhanced local helicity despite overall WT-like folding. L­(A16)P displays ∼30% α-helix plus a small (∼5%) β-strand population, suggesting a helix-to-strand transition with the remainder split between coil (40%) and turn (25%). R­(CPEP)­2C, by contrast, shows negligible helix but ∼7% β-strand alongside ∼45% coil and ∼48% turn, implying aberrant strand formation. H­(B5)­D gains about 25% helix at the expense of coil (30%) and turn (45%), consistent with partial overstabilization of local helical segments, whereas V­(B18)­A exhibits only ∼5% helix with coil and turn still dominating (∼45 and ∼50%, respectively), indicating minimal helix bias. Finally, Y­(B26)C displays the highest helical content (∼55%), dramatically converting the C-peptide into a more rigid helix and reducing coil to 15% and turn to 30%.

8.

Secondary-structural propensity of the C-peptide in proinsulin WT and MIDY mutants. The mutants show an increased helical propensity compared to the WT.

Collectively, these data demonstrate that the flexible, unstructured C-peptide of WT becomes progressively more helical or even strand-like in R­(CPEP + 2)C across the MIDY spectrum. This inherent structural plasticity is what allows the C-peptide to make and break packing contacts dynamically as the molecule explores the compact and partially open states. We propose that the loss of interchain stabilizing contacts (A–C and B–C interactions) gets compensated via local helical structures, which,

• 1.Reduces its ability to conform to the contours of the A- and B-chain surfaces,
• 2.Weakens key interchain stabilizing contacts (many identified in our correlation analysis), and
• 3.Favors conformational states in which the C-peptide is displaced outward, consistent with the increase in P(open) and the reduction in the compact-state free-energy basin.

Overall, the loss of A–C interactions is compensated for by the local secondary-structure formation in MIDY mutants.

Discussion

Protein folding diseases, also known as “conformational diseases”, arise when a protein’s structure is altered, causing it to form harmful aggregates. These aggregates can progress from oligomers to proto-fibrils and ultimately accumulate as amyloid fibrils, which damage tissue. Type 2 diabetes and Mutant INS-gene-induced Diabetes of Youth (MIDY) are such disorders that have its origins in the misfolding/incorrect folding of proinsulin. , While the conformational roots of loss-of-function in proinsulin are well established, the role of the C-peptide in shaping the proinsulin conformational landscape is underexplored.

The field of computational structural prediction and modeling has become a powerful tool for studying misfolding-associated diseases, accelerating research by allowing us to predict how structural changes impact a protein’s function. This is vital for understanding biological mechanisms and for developing new, structure-based drugs. For example, one study used these methods to examine nonsense mutations in GABRG2, revealing that different truncation points lead to distinct structural defects and varying severities of epilepsy. The same principle, linking a protein’s structure to its function and disease pathology, allows us to understand how proinsulin mutations lead to its misfolding in MIDY, which directly influences the conformational behavior and dynamics of the C-peptide (current study).

The C-peptide plays a crucial role in proinsulin’s folding and its conformational dynamics. ,, Acting as the connecting link between the A and B chains, it ensures they are correctly positioned to facilitate the formation of three essential disulfide bonds. , Without the C-peptide, the A and B chains would not fold efficiently into the correct native structure. The current study emphasizes that the C-peptide’s interaction with the A and B chains is dynamic. A more “open” conformation of proinsulin corresponds to a state where the C-peptide is loosely packed against the A and B chains. Conversely, a “closed” conformation represents a state where the C-peptide is brought closer, resulting in more favorable contacts with the A or B chains and stabilizing the final correctly folded structure. MIDY mutations can disrupt this delicate balance, causing the C-peptide to adopt a configuration that favors the open state. This leads to a persistent misfolded state, increasing the likelihood of aggregation. Our metadynamics-MD analyses (Figures –) reveal a unified, quantitative mechanism by which MIDY-associated point mutations progressively remodel proinsulin’s folding landscape, driving a continuum from near-wild-type (WT) stability to outright collapse of the native basin. Mutations such as E­(A4)­K, which preserve or modestly enhance the depth of the compact-state free-energy well (ΔG _{closed→open} ≈20.4 kcal·mol^–1 vs WT 19.3 kcal·mol^–1), maintain intact C-peptide helicity (∼40%) and full retention of canonical A-chain–C-peptide and B-chain–C-peptide stabilizing contacts (Figures –). In contrast, severe variants (H­(B5)­D, L­(A16)­P, and Y­(B26)­C) collapse the free-energy barrier for opening up to near zero, erode multiple salt bridges (retaining only Arg²²–Asn⁸⁶), and convert the once-flexible C-peptide into a rigid helix (25–55% helicity) or even a β-strand in R­(C-peptide)₂C (Figure ). The graded destabilization of the native-state barrier, quantified by PMF shifts (Figure ), ΔG reductions (Figure ), and Boltzmann-weighted open-state sampling (negligible in WT to >50% in severe mutants), tracks precisely with the loss of A-chain–C-peptide docking (Figure ), fragmentation of the A–C docking network (Figures and ), and the dismantling of the electrostatic core (Figure ). The C-peptide is thus the “linchpin” of proinsulin integrity; its dynamic ability to unfold and refold (Figure ) could thus enable proper A–B chain assembly during protein folding. These mutations, along with aberrant post-translational modifications (disulfide bond mispairing, in this case) (Supporting Figure S5), can dramatically alter the conformational landscape of proinsulin, making it susceptible to oligomerization.

Further strengthening this idea, our aggregation prediction results (Figure S6), and previous experimental evidence, ,, suggest that proinsulin has strong aggregation tendencies. Therefore, perturbation to the free-energy landscape (Figure ) due to an altered interaction network (Figure ) could have significant implications for the fate of the peptide, leading to misfolding, aggregation, and loss of function in physiological settings.

Therapeutically, our findings suggest two complementary rescue strategies. First, electrostatic “molecular glues” (small molecules designed to rebridge key salt-bridge pairs) could rebuild the collapsed network in moderate mutants. Second, peptide-mimetic chaperones that transiently bind and stabilize the C-peptide helix at its native docking register may restore the dynamic range necessary for proper A–B docking, particularly in variants with helix-bias traps. Our study’s limitations include reliance on classical force fields and finite sampling, which may under- or overestimate barrier heights and secondary-structure propensities. We also omitted the oxidizing endoplasmic reticulum (ER) environment and the ER luminal pH. Future work should incorporate constant-pH MD and enhanced sampling of disulfide isomerization as well as in vitro validation of predicted small-molecule stabilizers and peptide chaperones. In sum, our integrated computational portrait illuminates how single-site mutations dismantle proinsulin’s finely tuned folding machinery, quantitatively linking biophysical perturbations to disease phenotypes and laying a rational foundation for targeted MIDY therapies.

Glossary

Abbreviations

WT

wild-type

C-peptide

connecting peptide

MIDY

mutant INS-gene-induced diabetes of youth

MD

molecular dynamics

PMF

potential of mean force

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.5c10573.

• Symmetric inter-residue contact maps for WT, E­(A4)­K, Akita, R­(CPEP + 2)­C, V­(B18)­A, H­(B5)­D, Y­(B26)­C, and L­(A16)­P, zoomed to A-chain vs C-peptide (Figure S1); symmetric inter-residue contact maps for WT, E­(A4)­K, Akita, R­(CPEP + 2)­C, V­(B18)­A, H­(B5)­D, Y­(B26)­C, and L­(A16)­P, zoomed to B-chain vs C-peptide (Figure S2); WT and mutant residue–residue Δ-contact maps are shown for all variants with circles sized and colored by occupancy change (Figure S3); correlation between residue–residue electrostatic interaction energies and the free-energy cost of opening for the six strongest-correlated interchain contacts (Figure S4); effect of post-translational modifications (disulfide bridges) on the conformational landscape of proinsulin (Figure S5); and aggregation tendency in proinsulin as predicted by three algorithms (Figure S6) (PDF)

S.R., P.Z., K.D., D.B., P.K., and A.A. generated research data. S.R. and A.A. wrote the manuscript. All authors reviewed the manuscript. P.Z., D.B., S.R., and A.A. revised/edited the manuscript and contributed to the discussion. All authors approved the final version of the manuscript.

We thank the Max Planck Institute for the Physics of Complex Systems (S.R.) and East Tennessee State University (A.A.) for their financial support.

The authors declare no competing financial interest.