Effect of Proline Mutations on the Monomer Conformations of Amylin

Abstract

The formation of human islet amyloid polypeptide (hIAPP) is implicated in the loss of pancreatic β-cells in type II diabetes. Rat amylin, which differs from human amylin at six residues, does not lead to formation of amyloid fibrils. Pramlintide is a synthetic analog of human amylin that shares three proline substitutions with rat amylin. Pramlintide has a much smaller propensity to form amyloid aggregates and has been widely prescribed in amylin replacement treatment. It is known that the three prolines attenuate β-sheet formation. However, the detailed effects of these proline substitutions on full-length hIAPP remain poorly understood. In this work, we use molecular simulations and bias-exchange metadynamics to investigate the effect of proline substitutions on the conformation of the hIAPP monomer. Our results demonstrate that hIAPP can adopt various β-sheet conformations, some of which have been reported in experiments. The proline substitutions perturb the formation of long β-sheets and reduce their stability. More importantly, we find that all three proline substitutions of pramlintide are required to inhibit β conformations and stabilize the α-helical conformation. Fewer substitutions do not have a significant inhibiting effect.

Introduction

Amylin, or islet amyloid polypeptide (IAPP), is a peptide hormone that is coproduced with insulin by pancreatic β-cells (1,2). Human IAPP (hIAPP) consists of 37 residues, with a disulfide bridge between residues 2 and 7. The amyloid formation of hIAPP has been closely associated with the development of type II diabetes (1,3–7). It is believed that hIAPP amyloid deposits cause the loss of β-cells, thereby reducing insulin secretion (6,8–10).

Other species, including nonhuman primates, cats, and the degu, are also known to develop islet amyloid (3,11–14). In contrast, rat and mouse IAPPs are not amyloidogenic (15–17). As shown in Fig. 1, rat IAPP (rIAPP) contains 37 residues and differs from hIAPP at only six of these residues. Hence, rIAPP is often used as a proxy to study the mechanism and inhibition of hIAPP fibril formation (18,19). Out of these six residues, rIAPP has three proline residues, often referred to as β-sheet breakers, that greatly attenuate fibril formation (16,17). Based on this information, pramlintide, a synthetic human amylin analog, was designed to exhibit much reduced amyloidogenic properties. Pramlintide, also known as symlin, includes three proline substitutions at residues 25, 28, and 29, as seen in rIAPP (15,20). It has been used clinically in amylin replacement treatment for both type I and type II diabetes patients who take insulin (3,21,22).

Figure 1.

Sequence alignment with numbering based on hIAPP for various hIAPP derivatives, and IAPP of rat and species that form islet amyloid (11–14). Sequences of hIAPP, pramlintide, and two other hIAPP mutants are listed above the green dashed line. The proline mutations are marked with red, underlined fonts. The species variants of IAPP are given below the green dashed line, with the residues that differ from those in hIAPP marked in blue, underlined fonts.

Due to its amyloidogenic nature, it is difficult to study the conformational states of hIAPP in solution. Early in vitro circular dichroism (CD) studies found that hIAPP is mostly unstructured (23,24). More recent CD and NMR measurements, however, suggest that hIAPP exhibits a tendency to form an α-helix at the N-terminus region (residues $5-23$), and that the C-terminus remains unstructured (25–29). Cort et al. (30) suggested that both hIAPP and pramlintide form an α-helix at the N-terminus, with pramlintide exhibiting a stronger helical content than hIAPP. It is believed that the N-terminal helix is responsible for receptor binding (1).

Just as the monomer conformation is poorly understood, little is known about the hIAPP fibril structure. Kajava et al. (31) proposed a three-strand model, with β-strands assigned to residues $12-17$, $22-27$, and $31-37$. Luca et al. (32) used solid-state NMR (ssNMR) to propose a strand-loop-strand morphology for hIAPP monomers in a fibril, where residues $8-17$ and $28-37$ form β-sheets. Such a morphology is also supported by time-resolved two-dimensional IR (2DIR) and electron paramagnetic resonance (EPR) studies, with minor differences of one or two residues in the β-strand assignments (33,34). Wiltzius et al. (35) utilized fibril structures from the segments of residues $21-27$ and $28-33$ and relied on x-ray diffraction to propose a similar U-shape fibril structure, but with more closely interdigitated side chains compared with those inferred by ssNMR (32).

The central region of hIAPP, i.e., residues $20-29$, can independently form fibrils. That region has been proposed to be responsible for hIAPP amyloidogenic properties (36). The $20-29$ fragment also contains five out of the six differing residues between rIAPP and hIAPP, and it is often used as a model for studies of amyloid formation (37). Experiments have shown that one or more proline substitutions on such fragments at positions 25, 28, and 29 reduce fibril formation, with position 28 being the most critical (16,17). Additional studies have also demonstrated that fragments other than the central region, including $8-20$, $10-19$, $30-37$, and $8-37$, can form amyloid (38–40). Proline substitutions outside the central fragment also attenuate fibril formation (15,41).

Taken together, the studies cited above suggest that the aggregation of hIAPP involves a significant portion of the peptide, not just individual fragments. With that view in mind, in this work we investigated the effect of the proline substitutions encountered in pramlintide on the solution conformations of IAPP using the full peptide sequence. Recently, we used molecular dynamics (MD) and Monte Carlo simulations combined with Fourier-transform infrared (FTIR) spectrometry to study the conformations of full-length rIAPP and hIAPP monomers in dilute solution (42,43). Our data revealed that rIAPP adopts primarily two conformations: N-terminal α-helix and unstructured coil. In contrast, hIAPP was found to adopt three conformational states: N-terminal α-helix, β-hairpin, and unstructured coil. The β-hairpin conformation was shown to act as a possible precursor for amyloid formation (43–45). Simulations and time-resolved 2DIR have also been used to examine the structure of hIAPP fibrils as well as fibril formation in mixtures of hIAPP and rIAPP (18,33).

Here, we use MD and Monte Carlo simulations in the context of an advanced sampling technique, bias-exchange metadynamics (BE-Meta), to study the conformations of hIAPP and pramlintide in solution (46). BE-Meta has been successfully applied to study protein folding and the nucleation of small amyloid peptides (46–48). We also explore the conformations of two hIAPP derivatives, one having a proline substitution at residue 25 (hIAPP[A25P]) and the other having two proline substitutions at positions 28 and 29 (hIAPP[S28P,S29P]), and examine their structural stability.

Materials and Methods

Simulation details

Fig. 1 gives the amino acid sequences of hIAPP and the mutants considered in this work, namely, pramlintide, hIAPP[A25P], and hIAPP[S28P,S29P]. The C-terminal of each peptide was capped with an NH₂ group, and a disulfide bond was formed to link the Cys2 and Cys7 residues. The amino acid side chains were ionized on the basis of their pKa values, and the resulting peptide had a net positive charge of +3 at pH 7. The corresponding chloride ions were added to the simulation box to neutralize the system. All peptides were modeled using the GROMOS96 53a6 force field along with the SPC water model (49–51). Such a combination was shown to provide agreement with experimental IR spectrum in previous studies of the rIAPP monomer, and the hIAPP monomer and its aggregates (18,33,42,43). The cutoff for van der Waals and short-range electrostatic interactions was 9 Å. Long-range electrostatic interactions were treated with a particle mesh Ewald sum (52,53). A linear constraint solver was used to keep all bonds at their respective equilibrium lengths (54). Temperature and pressure were controlled at 298 K and 1 bar using the Nose-Hoover thermostat and the Parrinello-Rahman barostat, respectively (55–58). All simulations were performed with an integration time step of 2 fs using the GROMACS 4.5 molecular simulation package (59). The system’s coordinates were saved every 10 ps for further analysis.

Collective variables

The peptide conformation was characterized using the ${\alpha }_{rmsd}$ and anti-${\beta }_{rmsd}$ collective variables (CVs) introduced by Pietrucci and Laio (60). The ${\alpha }_{rmsd}$ is defined as

$${\alpha }_{rmsd}=\sum _{\alpha }n\left(RMSD\right),$$ (1)

$$\text{and}n\left(RMSD\right)=\frac{1-{\left(RMSD/0.8\right)}^8}{1-{\left(RMSD/0.8\right)}^{12}},$$ (2)

where RMSD measures the root mean-square deviations, in angstroms, of the backbone N, C_α, C, O, and C_β atom positions for every six consecutive residues in the peptide from those in an ideal α-helix (i.e., the average α-helical structure from the STRIDE database) (61). The function $n\left(RMSD\right)$ switches smoothly between 0 and 1, and ${\alpha }_{rmsd}$ sums over all possible six-consecutive-residue blocks. Larger ${\alpha }_{rmsd}$ values indicate more ideal α-helical blocks. Similarly, the anti-${\beta }_{rmsd}$ order parameter measures the RMSD values of the backbone atoms for every pair of three-residue blocks in the peptide sequence from those found in an ideal antiparallel β-sheet. Large anti-${\beta }_{rmsd}$ values indicate more antiparallel β-strands in the structure.

For the particular case of hIAPP, we also used the number of native contacts, $n_{contact}$, in the α and β structures determined in our previous work (43) as two additional CVs to improve conformational sampling:

$$n_{contact}=\sum _{i,j}^{contact}{n}_{ij},$$ (3)

$$\text{and}{\text{n}}_{\text{ij}}=\{\begin{array}{ll}1\hfill & \text{for}{r}_{ij}\le 5.5\hfill \\ \frac{1-{\left(\frac{r_{ij}-5.5}{2.5}\right)}^2}{1-{\left(\frac{r_{ij}-5.5}{2.5}\right)}^8}\hfill & \text{for}{r}_{ij}>5.5\hfill \end{array},$$ (4)

where $r_{ij}$ is the distance in angstroms between the C_α atoms of two residues that form hydrogen bonds (H-bonds) in the α or β conformations, $n_{ij}$, similar to $n\left(RMSD\right)$ in Eq. 2, is a switching function between 0 and 1, and $n_{contact}$ measures the number of C_α contacts of the α or β structures. The $n_{contact}$ for the α structure accounts for all the i to i+4 residual contacts in the segment comprising residues 7–21. For the β structure, $n_{contact}$ measures the C_α contacts between residue pairs (9,34), (10,33), (11,32), (12,31), (13,30), (14,29), (15,28), (16,27), (17,26), (18,25), and (19,24). For the other three hIAPP derivatives, due to lack of structural information, no native contact CVs and only two RMSD CVs were used to sample the peptide conformation.

Note that the main focus of this work is to evaluate the relative free energies of distinct hIAPP monomer conformations. Previous simulation studies relied on parallel tempering or replica-exchange techniques to explore hIAPP monomer conformations (43,44). Such studies revealed that hIAPP can adopt α-helical, β-hairpin, and random coil structures, where the antiparallel β-sheet is the main secondary structure in the β-hairpin conformation. Note that in the hIAPP mature fibril, peptides stack into in-register parallel β-sheets, not antiparallel ones. To further determine whether parallel β-sheets arise in the hIAPP monomer, we performed bias-exchange simulations with additional CVs, including intrapeptide H-bonds, dihedral angles, and C_γ contacts (see Supporting Material). The resulting free-energy profile as seen in Fig. S1 and Table S1 is consistent with that presented later in this work, in terms of only ${\alpha }_{rmsd}$ and anti-${\beta }_{rmsd}$ CVs. This finding provides additional evidence indicating that the hIAPP monomer by itself does not adopt a parallel β-sheet conformation.

BE-Meta simulations

To calculate the peptide conformational free energy with the CVs described above, we performed BE-Meta simulations (46) using the PLUMED plugin for GROMACS (62). The free-energy profiles were constructed using the METAGUI package for VMD (63,64). BE-Meta combines metadynamics and replica-exchange techniques (65–68) to enhance the sampling of various CVs. In metadynamics, a time-dependent bias is added to help the system escape local minima and enhance sampling. The bias added consists of a sum of Gaussian potentials (65):

$$V_G\left(\xi ,t\right)=W\sum _{t^{\prime }\le t}exp\left(-\sum _{i=1}^d\frac{{\left[{\xi }_i-{\xi }_i\left(t^{\prime }\right)\right]}^2}{2{\sigma }_i^2}\right),$$ (5)

where $V_G\left(\xi ,t\right)$ denotes the bias applied to the system at time t and CV values $\xi ={\xi }_1,{\xi }_2,\dots .{\xi }_d$; d is the number of CVs used to guide the simulation; W and σ are the height and width, respectively, of each Gaussian; and ${\xi }_i\left(t^{\prime }\right)$ is the ith CV at time $t^{\prime }$. In this work, W was 0.1 kJ/mol and σ was 0.5 for all CVs. Gaussian potentials were deposited every 1 ps. An exchange of configurations was attempted between two replicas every ${\tau }_{ex}=$ 10 ps using Metropolis criteria (46):

$$p_{ij}=min\left(1,e^{\left[V_G^i\left({\xi }^i,t\right)+V_G^j\left({\xi }^j,t\right)-V_G^i\left({\xi }^j,t\right)-V_G^j\left({\xi }^i,t\right)\right]/k_{B}T}\right),$$ (6)

where $p_{ij}$ is the probability of accepting an exchange between replica i and j, $k_B$ is the Boltzmann constant, T is the temperature, ${\xi }^x$ is the CV value in system x, and $V_G^x\left({\xi }^y,t\right)$ is the bias of system x corresponding to the CV value of system y at time t. In this study, for the case of two CVs, four replicas were employed. No bias was applied to the first replica. In the second and third replicas, Gaussian potentials were added to the first and second CVs, respectively. In the fourth replica, the bias acted simultaneously on both CVs.

After a certain amount of time $t_{fill}$, most of the free-energy minima of the system in each replica were filled and $V_G$ fluctuated around an average profile. The free energy along the CV, ξ, can be approximated with the time average of $V_G\left(\xi ,t\right)$ after $t_{fill}$ (69):

$$G\left(\xi \right)=\frac{1}{t_{all}-t_{fill}}\underset{t_{fill}}{\overset{t_{all}}{\int }}dtG\left(\xi ,t\right)=-\frac{1}{t_{all}-t_{fill}}\underset{t_{fill}}{\overset{t_{all}}{\int }}dt{V}_G\left(\xi ,t\right),$$ (7)

where $t_{all}$ is the total simulation time, and $G\left(\xi \right)$ is the system free energy as a function of ξ. The convergence of the free-energy profile in each replica was monitored using the protocol proposed by Marinelli et al. (47) and Biarnes et al. (63). As illustrated in Fig. 2, we calculated the average free energies corresponding to the time intervals $\left(t_{fill},t_{all}+t_{fill}/2\right)$ and $\left(t_{all}+t_{fill}/2,t_{all}\right)$, for the replica with the ${\alpha }_{rmsd}$ or anti-${\beta }_{rmsd}$ bias using Eq. 7. For the filling time $t_{all}=200$ ns, and the total simulation time $t_{all}=400$ ns, the two average free-energy profiles are within 1 $k_{B}T$ of each other, indicating that our simulations are converged.

Figure 2.

Convergence of the free-energy profiles for hIAPP, pramlintide, hIAPP[A25P], and hIAPP[S29P,S29P] (from the top to the bottom row, respectively). The replicas with bias are plotted in the left column, and the ones with anti- are in the right column. Following the protocol described by Marinelli et al. (47) and Biarnes et al. (63), the red profile is the time average within the time interval $\left(t_{fill},t_{all}+t_{fill}/2\right)$ using Eq. 7, and the blue one corresponds to the interval $\left(t_{all}+t_{fill}/2,t_{all}\right)$. The black profile is the mean of the two parts.

The information from all replicas is combined to evaluate the free energy along any CVs, ${\xi }^{\prime }$, using the weighted histogram analysis method (WHAM) (47,63,70):

$$G\left({\xi }^{\prime }\right)=-k_{B}Tln\left(\frac{\sum _i^k{n}_{{\xi }^{\prime }}^i}{\sum _j^k{e}^{\left(f^j-V^i\left({\xi }^{\prime }\right)\right)/k_{B}T}}\right),$$ (8)

$$V^i\left({\xi }^{\prime }\right)=\frac{1}{t_{all}-t_{fill}}\underset{t_{fill}}{\overset{t_{all}}{\int }}dt{V}_G^i\left({\xi }^{\prime },t\right),$$ (9)

where k denotes the number of replicas, $n_{{\xi }^{\prime }}^i$ is the number of times the system has the CV value ${\xi }^{\prime }$ in replica i, $V_G^i\left({\xi }^{\prime },t\right)$ is the biasing potential acting on the CV value ${\xi }^{\prime }$ in replica i at time t, and $f^i$ represents constants to be determined via WHAM iterations. This method has been used to study protein folding as well as short amyloid fibril nucleation (47,48). In this study, we applied BE-Meta and WHAM to construct the conformational free energies of hIAPP derivatives along the ${\alpha }_{rmsd}$ and anti-${\beta }_{rmsd}$ CVs.

Peptide conformation and cluster analysis

By monitoring the peptide conformations within the ${\alpha }_{rmsd}$ and anti-${\beta }_{rmsd}$ space, we find that the peptide adopts primarily the α structure when ${\alpha }_{rmsd}>3$ and anti-${\beta }_{rmsd}<4$, the β structure when ${\alpha }_{rmsd}<3$ and anti-${\beta }_{rmsd}>5$, and the γ conformation when ${\alpha }_{rmsd}<2$ and anti-${\beta }_{rmsd}<2$. The free-energy difference between two conformations is thus given by

$$\Delta G_{AB}=G_B-G_A=-k_{B}Tln\left(\frac{\sum _{\xi }^B{e}^{-G\left(\xi \right)/k_{B}T}}{\sum _{\xi }^A{e}^{-G\left(\xi \right)/k_{B}T}}\right),$$ (10)

where the free energy $G_i$ of conformation i is evaluated by integrating the free energy $G\left(\xi \right)$ of the region i in CV space ξ.

The peptide may adopt different structures in each conformational region. For example, if the peptide has an anti-${\beta }_{rmsd}$ value of 10, the peptide can form one long antiparallel β-sheet or exhibit several short antiparallel β-strands. To characterize this structural multiplicity in each conformational region, we extracted all of the protein conformations within the region and performed principal component analysis (PCA) on the conformational ensemble (71,72). Based on the projection on the first two modes, the peptide conformations were grouped into clusters using the K-means algorithm (73). To characterize the structural similarity within each cluster, we averaged the observed structures within a cluster and calculated the RMSDs from that mean structure. The free-energy difference between two clusters, m and n, can be estimated by

$$\Delta G_{mn}=G_m-G_n=-k_{B}Tln\left(\frac{\sum _{\text{Ω}}^m{e}^{-G\left({\xi }_{\text{Ω}}\right)/k_{B}T}}{\sum _{\text{Ω}}^n{e}^{-G\left({\xi }_{\text{Ω}}\right)/k_{B}T}}\right),$$ (11)

where $\text{Ω}$ is a conformation in cluster m or n and has CV values of ${\xi }_{\text{Ω}}$, and $G\left({\xi }_{\text{Ω}}\right)$ is the corresponding free energy of ${\xi }_{\text{Ω}}$. Combining PCA and Eq. 11, we further characterized different peptide conformations and their corresponding thermodynamic stability.

Results and Discussion

Human amylin

We used four CVs to sample the conformations of hIAPP: ${\alpha }_{rmsd}$, anti-${\beta }_{rmsd}$, and the native C_α contacts in the N-terminal α-helix and the β-hairpin structures proposed by Reddy et al. (43). The resulting free-energy surface along the ${\alpha }_{rmsd}$ and anti-${\beta }_{rmsd}$ is shown in Fig. 3 A. We find that the difference between the free-energy minima of the α and β conformations is −9.96 kJ/mol, which is lower than the −0.6 kJ/mol reported by Reddy et al. (43) for different CVs. Reddy et al. used the distance between two C_α carbons of a H-bonding pair residues in the folded structures as CVs. Under this definition, the unfolded state requires the H-bonding pair residues to exhibit large separation distances, which in turn implies more extended conformations of the molecule. In what follows, any local conformation with low ${\alpha }_{rmsd}$ and anti-${\beta }_{rmsd}$ values is considered to be unstructured. Note that such a definition for an unstructured coil is less restrictive than the one employed by Reddy et al. The anti-${\beta }_{rmsd}$ CV runs over all possible $3+3$ combinations among 37 residues of hIAPP, and thus is able to sample β-sheet conformations other than the one suggested by Reddy et al. To better estimate any free-energy differences, we integrated the free energies within each region and calculated the corresponding free energy using Eq. 10. Our results are listed in Table 1. The γ conformation, due to a large conformational ensemble in the defined CV region, exhibits the lowest free energy, namely $-6.31$ kJ/mol lower than the α conformation. The β conformation has a free energy that is $-4.04$ kJ/mol lower than the α conformation. Several conformations were observed in the β region. These conformations contribute to an overall low free energy of the β region on the free-energy surface shown in Fig. 3 A.

Figure 3.

(A) Free-energy surface of hIAPP along the ${\alpha }_{rmsd}$ and anti- ${\beta }_{rmsd}$ CVs. Contour lines are drawn every 5 kJ/mol. The regions defined for the α-helix, β-sheet, and random coil conformations are marked with purple, orange, and white rectangles. (B) The representative peptide conformation for the α-helical region is labeled with α. The five representative β conformations identified via PCA analysis are labeled as β1–β5. Also shown are the corresponding free energies and RMSDs from the mean structures of each conformation. The α-helical and β-sheet conformation are identified using the STRIDE algorithm, and are colored in dark purple and light yellow, respectively.Residues Ala-25, Ser-28, and Ser-29 are represented by a stick model colored in red, blue, and green, respectively. The images were generated using VMD (64). (C) For hIAPP, β-sheet segments along the peptide sequence in each β conformation are shown.

Table 1.

Free-energy difference of α, β, and γ conformations evaluated via Eq. 10 for hIAPP derivatives

	α	β	γ
hIAPP	$-$	−4.04	−6.31
Pramlintide	$-$	2.91	−1.76
A25P	$-$	−2.93	−3.64
S28P+S29P	$-$	−1.91	−2.03

The free-energy difference (kJ/mol) listed is relative to that of the α conformations.

To characterize the various β-sheet conformations that arise, we performed PCA on the peptide conformations corresponding to the β region. We identified five β-sheet-rich conformations, as illustrated in Fig. 3 B, all having average RMSDs ≤ 5.27 Å. Fig. 3 C shows the β-strands along the peptide sequence for each β conformation. Three of the structures (β1, β2, and β3) exhibit long β-hairpin conformations. The β1 hairpin is the same as that identified by Reddy et al. (43), where the turn region consists of residues $20-23$. The β2 conformation is similar to β1 but is shifted by one residue, with the turn region located at residues $19-22$. In the β3 conformation, the turn region is shifted toward the C-terminus and located at residues $24-27$, whereas the N-terminus generally folds into a short helix. The β4 structure is a collapsed form of β3 and consists of four short β-sheets with turn regions at residues $18-21$, $24-27$, and $30-33$. The β5 conformation is a four-fold β-sheet structure with three turn regions located at residues $10-13$, $18-21$, and $28-31$, respectively.

One of the most widely cited structures for the hIAPP mature fibril is the one proposed by Luca et al. (32) on the basis of ssNMR, which was further supported by EPR, 2DIR, and x-ray diffraction data (33–35). The β-strands in the U-shape fibril model are located at residues $8-17$ and $28-37$, which are consistent with the β-strands observed in the β1 and β2 conformations reported here. Kajava et al. (31) proposed a three-fold β-strand amyloid model of hIAPP with three β-strands assigned to residues $12-17$, $22-27$, and $31-37$, a structure that is similar to the β5 conformation identified in this work.

As for the α region, the dominant conformation observed in our simulations is shown in Fig. 3 B: the N-terminus of the peptide (residues 1–18) forms an α-helix, and the rest of the molecule adopts an unstructured coil conformation. Due to the random conformation in the C-terminus, the RMSD for the α-helical conformation is 7.19 Å, which is larger than that of the β conformations. The free energy of each conformation can be estimated using Eq. 11. Every β conformation has a free energy that is lower than that of the α conformation. The β1 conformation is the most stable, consistent with the results reported by Reddy et al. (43). All five conformations contribute to an overall lower free energy in the β region than that of the α region. In the literature, the α-helical conformation of hIAPP is considered important for receptor binding, whereas the β-hairpin conformation is viewed as the misfolded state (26,27,43–45). The free-energy data presented here for the hIAPP monomer exhibit multiple folded β states, suggesting that various segments in hIAPP have the ability to form a β-sheet. This observation is consistent with experimental data on fibril formation with different hIAPP fragments (36,38–40).

Pramlintide

Pramlintide, or symlin, differs from hIAPP by three proline mutations at residues 25, 28, and 29, which are known to attenuate the propensity for fibrillization (15,20). We conducted BE-Meta using ${\alpha }_{rmsd}$ and anti-${\beta }_{rmsd}$ as CVs to sample its conformational space. The resulting free-energy profile as a function of ${\alpha }_{rmsd}$ and anti-${\beta }_{rmsd}$ is displayed in Fig. 4 A. Compared with hIAPP, pramlintide exhibits lower free energies in the α region and higher free energies in the β region. By simply comparing the free-energy minima, we find that the β conformation is 0.27 kJ/mol higher than the α conformation. After integrating the free energy of each region, as listed in Table 1, we find that the free-energy difference between the α-helix and the β-sheet conformation is $-2.91$ kJ/mol. This suggests that the three proline mutations do inhibit the formation of conformations rich in β-sheet content. In addition, according to the free-energy surface, pramlintide can adopt states with ${\alpha }_{rmsd}>9$, indicating that the proline residues also stabilize the α-helix at the N-terminus.

Figure 4.

(A) Free-energy surface of pramlintide along the ${\alpha }_{rmsd}$ and anti-${\beta }_{rmsd}$ CVs. Contour lines are drawn every 5 kJ/mol. The regions defined for the α-helix, β-sheet, and random coil conformations are marked with purple, orange, and white rectangles. (B) The representative peptide conformation for the α-helical region and the three representative β conformations identified via PCA analysis are labeled as α and β1–β3, respectively, along with the corresponding free energies and RMSDs from the mean structures. Residues Pro-25, Pro-28, and Pro-29 are represented by a stick model colored in red, blue, and green, respectively. The structure color code is the same as in Fig. 3. (C) For pramlintide, β-sheet segments along the peptide sequence in each β conformation are shown.

Similar to the case with hIAPP, we performed PCA to characterize the pramlintide conformations in the β region. Three β conformations (β1, β2, and β3) were identified, as shown in Fig. 4 B, with RMSDs smaller than 6.14 Å. The β1 conformation has six short β strands, with the proline residues located at the turn regions. In the β2 conformation, the peptide forms a long antiparallel β-sheet, with segments containing no proline residues. The β3 structure consists of four adjacent antiparallel β strands, similar to the Greek key motif. As for the α region, one major conformation is observed in which the segment comprising residues $4-17$ forms an α-helix and the rest of the peptide remains unstructured, similar to the hIAPP α conformation. Among the three β conformations, the most stable one, β1, has a free energy that is 2.55 kJ/mol higher than that of the α conformation, indicating that the helical conformation is the most stable folded structure of pramlintide. This again confirms that the three proline residues inhibit the formation of β-sheets and stabilize the α-helix. The β2 conformation has two β-strands consisting of segments $9-17$ and $31-36$, which contain no proline residues. This observation also agrees with experimental data showing that hIAPP fragments of residues $10-19$ and $30-37$ exhibit amyloidogenic properties (38–40). Indeed, Green et al. (15) showed that although the three proline substitutions greatly attenuate fibril formation, pramlintide can still form small amounts of fibrils (and certainly more than rIAPP). These results are consistent with the existence of β-rich metastable conformations in pramlintide inferred from our simulations. Green et al. also showed that the three nonproline, rat-specific residues, i.e., Arg-18, Leu-23, and Val-26, modulate the fibril-forming kinetics, highlighting the importance of these residues in amyloid formation.

hIAPP[A25P] and hIAPP[S28P,S29P]

To examine the effects of proline mutations, we calculated the conformational free energies of hIAPP peptide with one mutation at Ala-25 (hIAPP[A25P]), as well as two mutations at Ser-28 and Ser-29 (hIAPP[S28P,S29P]). The free-energy surface along ${\alpha }_{rmsd}$ and anti-${\beta }_{rmsd}$ for hIAPP[A25P] is shown in Fig. 5 A. The free-energy differences among α, β, and γ conformations estimated via integration over distinct regions are listed in Table 1. The β conformation for A25P has a free energy that is −2.93 kJ/mol lower than the corresponding α conformation. Although the difference is smaller than that in hIAPP, the β conformation is still slightly more favorable. Fig. 5 B displays the six peptide conformations in the β region of the free-energy map, identified via PCA and cluster analysis. The most favorable β conformation, β1, is similar to the β3 conformation of hIAPP, with the turn region located at residues $24-27$. Its free energy is −4.06 kJ/mol lower than that of the α conformation. The other conformations, β2 and β3, consist of four to five short β-stands. The β4 conformation is a three-fold β-strand structure, whereas β5 and β6 exhibit four antiparallel β-strands. For conformations β1–β5, the Pro-25 residue is found at or near the turn region. Also, among the six β conformations, only β1 has a long hairpin. When contrasted with the three hairpin conformations observed in hIAPP, this finding suggests that proline indeed serves as a β-sheet breaker and attenuates long-β-sheet formation.

Figure 5.

(A) Free-energy surface of hIAPP[A25P] along the ${\alpha }_{rmsd}$ and anti- ${\beta }_{rmsd}$ CVs. Contour lines are drawn every 5 kJ/mol. The regions defined for α-helix, β-sheet, and random coil conformations are marked with purple, orange, and white rectangles. (B) The representative peptide conformation for the α-helical region and the six representative β conformations identified via PCA analysis are labeled as α and β1–β6, respectively, along with the corresponding free energies and RMSDs from the mean structures. Residues Pro-25, Ser-28, and Ser-29 are represented by a stick model colored in red, blue, and green, respectively. The structure color code is the same as in Fig. 3. (C) For hIAPP[A25P], β-sheet segments along the peptide sequence in each β conformation are shown.

The free-energy surface of hIAPP[S28P,S29P] as a function of ${\alpha }_{rmsd}$ and anti${\beta }_{rmsd}$ is shown in Fig. 6 A. As shown in Table 1, the β-sheet region has an overall free energy of −1.91 kJ/mol relative to the α-helical region. The reduction of the free-energy difference is more than that observed for hIAPP[A25P], suggesting that sequential mutations at Ser-28 and Ser-29 have a larger inhibiting effect on β-sheet formation for the hIAPP monomer. PCA and cluster analysis reveal two major β conformations for hIAPP[S28P,S29P], as shown in Fig. 6 B. The β1 conformation is a three-strand, antiparallel β-sheet conformation, similar to the β4 conformation of hIAPP[A25P]. The β2 conformation is the most stable for hIAPP[S28P,S29P], with a free energy of −4.08 kJ/mol relative to the α conformation. The β2 conformation is similar to the β3 structure of hIAPP and the β1 structure of hIAPP[A25P] in which the two proline mutations break the long β-hairpin into two pairs of antiparallel β-sheets. Although Pro-28 and Pro-29 suppress long-β-sheet formation, the two β conformations still have favorable free energies compared with the α-conformation. For both hIAPP[A25P] and hIAPP[S28P,S29P], the α-helical regions have one dominant α-conformation, in which residues $4-17$ form an α-helix. Because the helical segment does not include the mutation sites investigated in this study, the α conformations for all mutated hIAPP monomers are similar to those found in wild-type hIAPP.

Figure 6.

(A) Free-energy surface of hIAPP[S28P,S29P] along the ${\alpha }_{rmsd}$ and anti- ${\beta }_{rmsd}$ CVs. Contour lines are drawn every 5 kJ/mol. The regions defined for α-helix, β-sheet, and random coil conformations are marked with purple, orange, and white rectangles. (B) The representative peptide conformation for the α-helical region and the two representative β conformations identified via PCA analysis are labeled as α, β1, and β2, respectively, along with the corresponding free energies and RMSDs from the mean structures. Residues Ala-25, Pro-28, and Pro-29 are represented by a stick model colored in red, blue, and green, respectively. The structure color code is the same as in Fig. 3. (C) For hIAPP[S28P,S29P], β-sheet segments along the peptide sequence in each β conformation are shown.

Several experiments have suggested that amylin remains amyloidogenic upon single-residue proline mutations at residues 25, 28, or 29 (16,17). In addition, the amylin peptides corresponding to species having one or two prolines, such as degu (Pro-28 and Pro-29), cat (Pro-29), and hamster (Pro-25 and Pro-29; see Fig. 1), are known to develop islet amyloid deposits (11–14). Figs. 3–6 C shows the β-sheet segments along the peptide sequence in each β conformation for the hIAPP derivative. Due to their cyclic structural nature, proline residues are found in the turn regions or at the end of β-strands, acting as β-sheet breakers and inhibiting long-β-sheet formation. When different β-strand sequences are compared, the β4 and β5 conformations of hIAPP[A25P] and the β1 conformation of hIAPP[S28P,S29P] exhibit strong similarities to the three-strand amyloid model proposed by Kajava et al. (31). This observation suggests that hIAPP[A25P] and hIAPP[S28P,S29P] are able to retain some amyloidogenic characteristics. According to our free-energy results (Table 1), the β conformation becomes less favorable as proline mutations are introduced. The consecutive mutations at Ser-28 and Ser-29 have a greater effect than a single mutation at Ala-25 on β-sheet inhibition. However, the β conformations are still more favorable than the α conformations for both hIAPP derivatives. All three proline mutations in pramlintide are necessary to make the α conformation more thermodynamically stable than the β conformations. This observation agrees with experimental results on the amyloidogenic nature of variants of IAPP corresponding to different species. Our results show that the proline substitutions modulate the conformation of IAPP monomers because proline reduces the amyloidogenic tendency by suppressing β-sheet formation.

Conclusions

BE-Meta simulations were used to study the conformational space and the corresponding free energy of several mutants of IAPP, namely hIAPP, hIAPP[A25P], hIAPP[S28P, S29P], and pramlintide (hIAPP[A25P, S29P, S29P]). All four hIAPP derivatives were found to adopt multiple β-sheet conformations, unstructured coils, and one major α-helical conformation in which the N-terminus adopts an α-helical state. Several of the β-rich conformations observed here are consistent with previous experimental reports.

In one of the possible aggregation mechanisms of hIAPP, monomers can interconvert between native folded states, e.g., between the α and γ conformations and the β-rich conformations (43–45). The β-conformers can act as the amyloidogenic precursor by templating the folding of additional monomers, and promote self-assembly into small aggregates and subsequent fibril formation. The results presented here provide insights into the thermodynamic stability of distinct folded states, which could serve as precursors to fibril formation, and may be useful for developing therapeutic strategies aimed at inhibiting early-stage aggregation.

Our simulated free energies help explain the tendency for aggregation (or lack thereof) of the amylin peptides found in different species. For hIAPP, hIAPP[A25P], and hIAPP[S28P,S29P], the β-rich states exhibit lower free energies than the corresponding α-rich conformations. The [S28P,S29P] mutations are more effective than the A25P mutation at raising the free energy of the β-rich conformations. For pramlintide, where all three proline mutations are introduced, the α-helical conformation becomes more thermodynamically favorable than the β-rich conformations. Our combined results indicate that although [S28P,S29P] mutations are effective at reducing β-sheet formation, one really needs three mutations, namely [A25P, S29P, S29P], to eliminate hIAPP’s ability to aggregate.

Another possible aggregation mechanism of hIAPP involves the association of two hIAPP α-conformers into small nuclei (3,74). In future work, the methodology developed here will be applied to probe the effect of proline mutations on the dimerization of different hIAPP conformers and to explore the thermodynamics of nuclei formation.