Pair your accounts.

Export articles to Mendeley

Get article recommendations from ACS based on references in your Mendeley library.

Pair your accounts.

Export articles to Mendeley

Get article recommendations from ACS based on references in your Mendeley library.

You’ve supercharged your research process with ACS and Mendeley!

STEP 1:
Click to create an ACS ID

Please note: If you switch to a different device, you may be asked to login again with only your ACS ID.

Please note: If you switch to a different device, you may be asked to login again with only your ACS ID.

Please note: If you switch to a different device, you may be asked to login again with only your ACS ID.

MENDELEY PAIRING EXPIRED
Your Mendeley pairing has expired. Please reconnect
ACS Publications. Most Trusted. Most Cited. Most Read
My Activity
CONTENT TYPES

Human Amylin in the Presence of SARS-COV-2 Protein Fragments

Cite this: ACS Omega 2023, 8, 13, 12501–12511
Publication Date (Web):March 23, 2023
https://doi.org/10.1021/acsomega.3c00621

Copyright © 2023 The Authors. Published by American Chemical Society. This publication is licensed under

CC-BY-NC-ND 4.0.
  • Open Access

Article Views

1013

Altmetric

-

Citations

LEARN ABOUT THESE METRICS
PDF (4 MB)
Supporting Info (2)»

Abstract

COVID-19 can lead to the onset of type-II diabetes, which is associated with the aggregation of islet amyloid polypeptides, also called amylin. Using molecular dynamics simulations, we investigate how the equilibrium between amylin monomers in its functional form and fibrils associated with diabetes is altered in the presence of SARS-COV-2 protein fragments. For this purpose, we study the interaction between the fragment SFYVYSRVK of the envelope protein or the fragment FKNIDGYFKI of the spike protein with the monomer and two amylin fibril models. Our results are compared with earlier work studying such interactions for the two different proteins.

This publication is licensed under

CC-BY-NC-ND 4.0.
  • cc licence
  • by licence
  • nc licence
  • nd licence

Introduction

ARTICLE SECTIONS
Jump To

SARS-COV-2 infections affect not only the respiratory system but a multitude of organs in the human body. (1−4) However, the underlying mechanisms for the resulting broad spectrum of symptoms and complications of COVID-19 are not always understood. For instance, diabetes patients have often more severe symptoms of COVID-19, but SARS-COV-2 infections may also trigger the onset of diabetes. (5−8) One possible cause for disease symptoms in individuals with type-II diabetes is the aggregation of islet amyloid polypeptides (IAPP, also known as amylin). Amylin aggregation (or in more general, the onset of type-II diabetes) can be caused by various factors, for instance, inflammation induced by infections. However, one can speculate that amylin aggregation is also triggered directly by SARS-COV-2 protein fragments. In previous work, we have presented evidence for such a mechanism for Serum Amyloid A (9) and α-synuclein. (10) However, amylin may be affected differently by interactions with viral proteins since it is more stable than these two proteins, being approximately 65% helical and stabilized by a disulfide bridge. For this reason, we investigate in this study whether interactions with SARS-COV-2 protein fragments also alter amylin amyloid formation and how this effect differs from the one seen in our previous studies. For this purpose, we use all-atom molecular dynamics simulations to investigate the effect of two virus protein fragments on amylin monomers (Figure 1c) and two fibril models (Figure 1d,e). The first fragment, SK9, is from the envelope protein and allows us to connect our work to the above described previous one, while the spike-protein fragment FI10, unique for SARS-COV-2, has been shown to form amyloids in vitro. (15) Note that the two fragments, shown in Figure 1a,b, are both from viral surface proteins, increasing the chances for interaction with extracellular proteins such as amylin. Unlike in the earlier work on Serum Amyloid A (9) and α-synuclein, (10) we do not see a shift in the ensemble of monomers toward more aggregation-prone conformations but rather protection of the native conformation. However, we observe the stabilization of fibril structures that not only depends on the viral protein fragments but also on the amylin fibril geometry.

Figure 1

Figure 1. Licorice representation of the two SARS-COV-2 protein fragments used in this study: (a) SK9 and (b) FI10. Ribbon representations of the (c) amylin monomer (PDB ID: 2L86) and the two fibril models with (d) 2F4L and (e) 6ZRF are also shown. The N- and C-terminals are colored in blue and red, respectively.

Results and Discussion

ARTICLE SECTIONS
Jump To

Monomer Simulations

We start our investigation into the effect of SARS-COV-2 protein fragments on amylin aggregation by first looking into the changes of the ensemble of monomer conformations induced by viral proteins. In our previous work, we have demonstrated that the presence of the nine-residue-long C-terminal fragment SK9 (SFYVYSRVK) of the SARS-COV-2 envelope protein shifts the ensemble of SAA (9) and αS (10) monomers toward more aggregation-prone conformations, increasing in this way the probability for amyloid formation. The starting point for our present simulations is the amylin monomer model with PDB ID 2L86 (11) in complex with SK9. By comparing two different programs (AutoDock Vina and HADDOCK) to identify potential binding sites, (12−14) we come up with two distinct sets of trajectories. This allows us to probe the effect of the initial binding site on our results. A third set of trajectories was started from a configuration where the Spike protein fragment FI10 (FKNIDGYFKI) binds to the amylin monomer. This fragment, cleaved by the enzyme neutrophil elastase, which is released from neutrophils during acute inflammation, is unique for SARS-COV-2 and has been shown to form amyloids in vitro. (15) Simulations of SK9 or FI10 interacting with amylin are compared with control simulations where the viral protein fragments are absent. PDB files of start and final configurations of all runs are available in the Supporting Information.
In Figure 2a, we show the number of original binding contacts (contacts that exist at start) as a function of time. Initially, SK9 has in one set (binding site derived with Autodock Vina) 37 contacts with the amylin monomer and 35 in the other set (binding site found with HADDOCK), and FI10 forms 38 contacts with the amylin monomer. In order to compare the various trajectories, we normalize the number of contacts for each system to a value of 1 at start. The log–log plot in the inset shows that the number of contacts decreases by a power law with the exponent 0.34(1) for SK9 and of approximately 0.5 for FI10. These exponents indicate a diffusive motion of the viral protein fragments in relation to the amylin monomer, which, in the case of SK9, is slower than in normal diffusion and normal (i.e., random-walk-like) for FI10. Note, however, that in the process of this diffusive motion, many of the disappearing original contacts are replaced by contacts not seen in the start conformation. As a consequence, the total number of contacts between the viral protein fragment and amylin decreases much slower (see Figure 2b), staying, for SK9, close to 30–40% of the initial number. For FI10, the number of contacts stays even at 60–70%. The center-of-mass distance between both molecules is approximately 5–20 Å but occasionally increases up to 50 Å; see Figure S2. Hence, most of the time, the viral protein fragment stays attached to the amylin monomer, but occasionally, it separates transiently. Together with our binding free energy estimates, the above observations indicate stable binding between the viral protein fragment and amylin. For SK9, we find a value of −21(3) kJ/mol for the trajectories starting from the Autodock Vina generated binding site and a value of −22(5) kJ/mol for trajectories with the HADDOCK generated initial binding site. These values are calculated by taking the last 800 ns of the trajectories into account and only decrease to −18.2 (3) and −18(5) kJ/mol when only the last 200 ns are considered. Hence, the binding of SK9 to amylin depends little on the binding site and decreases only slowly. For FI10, we find values of −22(2) kJ/mol, taking the last 800 ns into account. When restricting the calculation to the last 200 ns, we have to rely on only two trajectories as, in the third one, FI10 stayed in contact with amylin over the whole time, that is, it did not allow the estimation of a binding energy with the approach by Bellaiche and Best. (41) Hence, our values of −28(4) kJ/mol for the two trajectories where we could use the approach is an underestimation. Nevertheless, this value indicates already that FI10 is binding more tightly with the amylin monomer than SK9, which is supported by all other quantities in Table 1. We remark that we observe the independence from the binding site and the difference between SK9 and FI10 and also when the binding energies are estimated directly from the start conformation using Prodigy. (16) However, the obtained absolute values of the free energies, namely, −45.6 (0) kJ/mol for SK9 docked with AutoDock Vina and −49.4 (7) kJ/mol for those docked with HADDOCK and −41.6 (7) kJ/mol for FI10, are larger for Prodigy, which uses an approximation.

Figure 2

Figure 2. (a) Averaged number of original contacts between SK9 and amylin as a function of time. (b) Corresponding average number of all contacts (including newly formed ones). Contacts are normalized to 1 for the respective start configurations. Red marks the data from runs where the initial binding site of SK9 was found with AutoDock Vina; green is where the binding site was found with HADDOCK. Blue marks the data from runs where FI10 interacts with amylin.

Table 1. Averages of Various Quantities Measured in Simulations of the Amylin Monomer Alone or in the Presence of SK9 or FI10a
systemtime intervaltotal contactsoriginal contactsSASARghelicityend-to-end distanceΔGbind (kJ/mol)
control800 ns  0.96 (4)1.04 (5)0.6 (1)1.325 (5) 
200 ns  0.98 (7)1.1 (1)0.6 (2)1.27 (8) 
w/ SK9800 ns0.39 (8)0.05 (2)0.96 (3)1.14 (6)0.79 (9)1.4 (3)–21 (4)
200 ns0.33 (8)0.04 (1)0.96 (3)1.07 (6)0.68 (3)1.3 (4)–18 (4)
w/ FI10800 ns0.60 (2)0.054 (7)1.00 (5)1.2 (1)0.79 (5)1.61 (9)–22 (2)
200 ns0.70 (5)0.08 (3)1.03 (6)1.2 (1)0.72 (3)1.7 (3)–28 (4)
a

Averages are over the last 200 or 800 ns and over all respective trajectories. Contacts and binding free energies are between SK9 or FI10 and amylin. With the exception of the binding free energies, all quantities are normalized to 1 at start.

Note that, except for the terminal residues, we find little difference in the frequencies in that SK9 residues bind to amylin with slightly higher values for the central Y5 and lower values for residues S6–S8 than for residues F2–V4. On average, the binding probability of an SK9 residue is approximately 60(10)% for the last 800 ns and 50(20)% when calculated only over the last 200 ns. Here, the terminal residues are excluded in the calculation of the binding probabilities. Similarly, we find for FI10 a binding probability of approximately 60(10)% when measured over the last 800 ns and 70(20)% when measured only over the last 200 ns. However, residues D5–I10 of FI10 have higher binding propensities than the first three residues (FKN); see Table S1. The frequencies are again independent of the initial binding site. Similarly, no clear pattern is seen in the amylin residues binding with SK9 or FI10 (see Figure S3), reflecting the diffusive movement of the viral protein fragment in relation to the amylin monomer.
The effect of the binding of the viral protein fragment on the ensemble of amylin monomer conformations can be seen by comparing the simulations of amylin interacting with either SK9 or FI10 with the control simulations (where the viral protein fragments are absent). One example is the root mean square deviation (RMSD) to the start conformation, which we show as a function of time in Figure 3. Shown are averages over all trajectories for each system. We have merged all trajectories of SK9 interacting with the amylin monomer as the data for SK9 differ little between the runs starting from binding sites generated with Autodock Vina and the ones generated by HADDOCK. After a rapid initial increase, the RMSD stays in the control simulations, and for amylin interacting with either SK9 or FI10, it is almost constant over the whole length of the trajectory. However, while the RMSD values differ little between the control and amylin in the presence of either SK9 or FI10, visual inspection of the final configurations (Figure 4) shows differences in size and helicity to the control. For instance, the average radii of gyration of the amylin monomers are 11.4(2) Å for the SK9 and 12.9(1) Å for the FI10 simulations, which are higher than in the control 11.2(1) Å. A similar pattern is seen for the end-to-end distance where we find 19(5) Å in the SK9 simulations, 26(5) Å in the presence of FI10, and 20(1) Å for the control; see Table 1. While the average solvent-accessible surface area (SASA) of the amylin monomer is similar in all cases, we find that the ratio of the exposed surface of hydrophilic residues to that of hydrophobic residues is 0.86(2) in the simulations where FI10 interacts with amylin, which is comparable to the 0.85(3) measured in simulations where SK9 interacts with amylin, but both values are smaller than the value (0.875(5)) measured in the control simulations where neither of the viral protein fragments is present.

Figure 3

Figure 3. Average root mean square deviation (RMSD) to the start conformation as a function of time for amylin in the presence of SK9 (red) or FI10 (blue) and in the absence of SK9 and FI10 (black). The RMSD is evaluated over backbone atoms only. Averages are calculated over all trajectories for each system.

Figure 4

Figure 4. Representative snapshots of the final amylin conformations in the (a) absence and presence of (b) SK9 or (c) FI10. The N- and C-terminals are colored in blue and red, respectively.

The reason for the differences in these ratios calculated from averages taken over the last 200 ns can be seen in Figure 5 where we show the root mean square fluctuations (RMSFs) of residues evaluated over the last 200 and 800 ns. This quantity allows us to quantify the flexibility of residues. Hence, the lower values seen in the presence of SK9 or FI10 indicate a stabilization of the monomer. Especially, we see in the control simulations that residues N22–N31 have a higher flexibility than others, but this distinction is not seen in the simulations where the viral protein fragments are present. This is interesting as human amylin differs in the segment N22–N31 from the less aggregation-prone rat amylin: rat amylin contains three proline residues between residues 20 and 29 that disrupt the secondary structure due to structural hindrance and decreased flexibility of the protein backbone. As a result of these residues, which are also found in the amylin fibril-inhibiting Pramlintide, rat amylin is less aggregation-prone and more soluble than human amylin. Hence, interactions with SK9 or FI10 mimic the protective effect of the rat mutations or in the amyloid-inhibiting drug Pramlintide.

Figure 5

Figure 5. Residue-wise root mean square fluctuation of the amylin monomer in the absence (black) or presence of SK9 (red) or FI10 (blue) calculated over the (a) last 800 ns and (b) last 200 ns of the simulations. Shaded regions mark the standard deviations of the shown quantities.

Indeed, we find that the average helicity, also shown in Table 1, stays marginally higher when amylin interacts with either SK9 (68(3)%) or FI10 (72(3)%) than in the control (60(20)%). However, these averages may be misleading. We show in Figure 6 the residue-wise helicity measured over the last 200 ns in simulations with either SK9 or FI10 present but subtracting the corresponding values measured in the control. In the presence of the viral protein fragments, residues F15–S28 consistently have higher average helicity than observed in the control while residues A5–N14 have a lower average helicity. This decrease of helicity for segment A5–N14 is more pronounced for FI10, which at the same time also stabilizes helicity in residues L16–T30 more than SK9. Residues S20–S29 have been shown to form the primary amyloidogenic domain in amylin, (17) while the N-terminal residues (1–8) are not directly involved in fibril formation. (18) Hence, we conclude that both SK9 and FI10 are stabilizing the native monomer conformation of amylin, decreasing the chance of unfolding and seeding amyloids. This is unlike what we observed in our previous work where we found that SK9 shifted the ensemble of SAA (9) and αS (10) monomers toward more aggregation-prone conformations, increasing in this way the probability for amyloid formation.

Figure 6

Figure 6. Residue-wise helicity of the amylin monomer in the presence of SK9 (red) and FI10 (blue) calculated over the last 200 ns. Shown is the difference to the corresponding values in the control simulations where the viral protein fragments are absent.

Table 2. Various Quantities Averaged over the Last 100 ns of Three Trajectories in Simulations of the Amylin 2F4L Fibril Model in the Absence or Presence of SK9
 controlw/ SK9
total SASA (Å2)15,554 (337)15,251 (356)
hydrophobic SASA (Å2)7782 (243)7623 (229)
hydrophilic SASA (Å2)7773 (141)7628 (134)
packing distance (Å)8.2 (8)8.1 (3)
stacking distance (Å)2.9 (2)2.9 (2)
packing contacts90 (11)92 (16)
stacking contacts494 (26)492 (20)
intrachain contacts1226 (6)1242 (3)
packing hydrogen bonds6 (3)7 (3)
stacking hydrogen bonds146 (5)142 (7)
intrachain hydrogen bonds17 (1)16 (1)
packing free energy (kJ/mol)–189.2 (48.3)–184.8 (77.0)
stacking free energy (kJ/mol)–251.1 (163.1)–305.7 (113.7)

Fibril Simulations

According to Le Chatelier’s principle, the equilibrium between amylin in its functional form and aggregated in amyloids can be shifted not only by increasing the propensity of aggregation-prone monomer conformations, raising in this way the probability for amyloid formation, but also by stabilizing existing fibrils. Amyloids with more stable fibrillar structures have been shown to have higher rates of aggregation. (19) In previous work, (9,10) we observed such a stabilizing effect for the SK9 fragment on SAA and αS fibrils. Hence, we have also probed the change in stability of amylin fibrils introduced by interactions with virus protein fragments. Unfortunately, only few models of amylin fibrils have been resolved and published in the PDB. Because of our familiarity with the fibril model put forward by Eisenberg and co-workers, (20) which we have studied in a different context in previous work, (21) we chose again this model, which we called 2F4L. PDB files of start and final configurations are available in the Supporting Information.
At the start of the three trajectories, the viral SK9 peptides form 257 contacts with the 2F4L fibril with approximately 32 contacts per chain. This number decreases rapidly over the first nanosecond and more slowly afterward but stays constant after 100 ns at 109(6) contacts, which is approximately 42% of the original number of contacts, that is, on average 14(1) contacts per chain, leading to a binding free energy of −17(9) kJ/mol between an amylin and a SK9 peptide. Note that not all the observed contacts appear in the start configuration: the SK9 fragment is not tethered to the fibril and new contacts are formed as the old original ones dissolve.
We show, in Figure 7a, the frequency of amylin residues having a contact with SK9 as measured over the last 100 ns. The main binding regions include residues N14–S19 (centered around F15) in the outer layers of the protofibrils and residues F23 and Y37 on the edge of the protofibril layers. Residues N14–S19 make hydrophobic contacts and π–π stacking interactions with SK9 to hold the SK9 close to protofibril layers, stabilizing their stacking. On the other hand, the residues F23 and Y37 form hydrogen bonds and hydrophobic contacts with SK9 that stabilize the packing of the two protofibrils. Hence, the effect of SK9 binding to the amylin chains is a slight stabilization of the fibril. This can be seen in Figure 8a where we compare the residue-wise root mean square fluctuation of the amylin fibril 2F4L bound to SK9 with that of the control (2F4L without SK9 present). This quantity is a measure for the flexibility of a residue, and while the overall form of the RMSF curves are similar, residues in the 2F4L fibril bound to SK9 have consistently lower RMSF values, that is, they are less flexible.

Figure 7

Figure 7. Residue-wise frequency of contacts of (a) the fibril model 2F4L with SK9 and (b) the fibril model 6ZRF with either SK9 (red) or FI10 (blue). Shaded regions mark the standard deviation of the averages.

Figure 8

Figure 8. Residue-wise RMSF for (a) the fibril model 2F4L and (b) the fibril model 6ZRF. The black curve marks data from the simulation in the absence of viral protein fragments, while the red curve shows the results obtained when SK9 is present, and the blue curve data is for when FI10 is interacting with the amylin fibrils. Shaded regions mark the standard deviation of the averages.

Consequently, the final configurations are less disturbed than in the control; see the insets of Figure 9a, which shows the root mean square deviation (RMSD) to the start conformation as a function of time, taken over all backbone atoms of all chains. Note that we consider for the calculation of the RMSD only residues 8–37 as the first seven residues are disordered. While the differences to the control are small, they show the greater stability of 2F4L bound to SK9 than that of the fibril in the absence of SK9. This can also be seen by visual inspection of the final conformations shown in the insets. Consistent with the RMSF plots, this stabilization is mainly due to the reduction of flexibility for the first N-terminal residues, which, in the control, appear to disrupt the stacking of the N-terminal β-sheets.

Figure 9

Figure 9. RMSD to the start configuration as function of time for (a) the fibril model 2F4L and (b) the fibril model 6ZRF. The RMSD is calculated over all backbone atoms in residues 8–37 of all chains in the respective fibril model. Black curves are from the control simulations, while red curves are from data measured in simulations where SK9 is present, and similarly, blue curves are for the case of FI10 interacting with the amylin fibril. Shaded regions mark the standard deviation of the averages. The start configuration and representative final configurations are shown as insets.

However, these visual impressions are difficult to quantify. For instance, the solvent-accessible surface area (SASA) (Table 2) is 15,251(356) Å2 when measured in the simulations with SK9, which is smaller than when measured in the absence of SK9, that is, 15,554(337) Å2. The difference between the two averages is small and within the standard deviations, but the difference becomes more pronounced when excluding the disordered first seven residues from the SASA calculation, leading to 11,445(254) Å2 in the presence of SK9 versus 11,853(227) Å2 in the control. The reduction in the exposed surface area in the complex is slightly higher for hydrophobic residues than for hydrophilic residues, but no clear picture emerges. Little difference is also seen for the average distance between the two folds (8.1(3) Å in the presence of SK9 versus 8.2(8) Å for the control), and the average distance between layers is 2.9(2) Å, which is about the same in the presence and absence of SK9. Similarly, neither our MMPBSA approximations of the packing and stacking free energies nor the measured numbers of stacking or packing contacts or hydrogen bonds in Table 2 give a clear signal explaining the observed weak stabilization.
Correspondingly, we do not see a difference in the frequency of the S29–S29 hydrogen bonds found in our earlier work to be crucial for the packing of the fibril or for the N31–N31 bond to be important for stacking. (21) However, we find the frequency of the F23–Y37 packing hydrogen bonds being raised from 18 to 38%, and the distribution of packing hydrogen bond is tighter in the presence of SK9. While, in the simulations of the control, more than 20 types of hydrogen bonds contribute at least 1% to the measured packing hydrogen bonds, only seven types of hydrogen bonds do this in the simulations where SK9 is present. Hence, while the number of packing hydrogen bonds changes little in the presence of SK9, the bonds are less transient when SK9 interacts with the fibril. The more focused packing can be also seen in the map of packing contacts in Figure S4a–c.
The effect is less obvious for the stacking or intrachain hydrogen bonds. However, we observe an increase in contacts involving the N-terminal residues T4–L16, while such involving the C-terminal residues S20–Y37 decrease or shift more to off-diagonal contacts; see Figure S4d–f. This is consistent with the observed stabilization of the N-termini of the chains. As we find a slight increase in the number of intrachain contacts when SK9 is interacting with the 2F4L amylin fibril, increasing from 153(1) to 155(1), we conjecture that the main reason for the weak stabilization of this fibril geometry by SK9 is due to stabilization of the fold geometry of the chains.
A characteristic of amyloids is the polymorphism of the resulting fibrils, and the virus proteins may change the amylin fibril stability differentially depending on the polymorph. Hence, we also evaluated the effect of SK9 on a second fibril model, which was resolved in 2020 by Gallardo et al. (22) and deposited in the PDB under ID 6ZRF. Simulations of this fibril model were also chosen to compare the effect of SK9 on the fibril stability with that of the FI10 virus protein fragment. PDB files of start and final configurations are available in the Supporting Information.
The SK9 segments form 238 contacts (approximately 30 contacts per chain) initially with amylin chains in the 6ZRF fibril model, which is slightly less than in the 2F4L fibril. Again, these contacts rapidly decrease in the first 10 ns before the number of contacts stabilizes to 97(16) (approximately 40% of the initial number of contacts) over the last 100 ns. The binding free energy between SK9 and the 6ZRF amylin model is −31.4(4.4) kJ/mol, which is larger in magnitude than the 2F4L fibril. However, contacts are formed with similar residues as in the case of the 2F4L fibril; see Figure 7b: residues N14–V17 (centered around F15), N21, F23, L27–T30 (centered around L27), and with the highest probability, Y37. As for the 2F4L fibril, these contacts result mostly from hydrophobic interactions but also from π–π stacking (involving residues F15, F23, L27, S29, and Y37) or transiently formed hydrogen bonds (N14@O-SK9:Y5@HN, L16@O-SK9:Y3@HN, F23@O-SK9:Y3@HN, L27@O-SK9:Y3@HN, and Y37@O-SK9:S1@H3). The net effect is again a stabilization of the amylin fibril: see the reduced flexibility of residues in the RMSF plot of Figure 8b and a reduction in the RMSD compared to the control that is more pronounced than for the 2F4L fibril geometry; see Figure 9b. For residues N14–V17, L27–T30, and Y37, these contacts with SK9 were increased in the frequency of stacking contacts between successive layers when compared to the control; see, for instance, in Figure S6f, residue N14 forming hydrogen bonds between its side-chain carboxamide and other N14 residues in the protofibril. On the other hand, the interaction of SK9 with residues N21 and F23 strengthen the packing interactions between the two protofibrils. Residue F23 interacts with the hydrophobic core formed by residues L27–T30 and Y37 on the opposite protofibril, which in turn brings the protofibrils closer together, resulting in tighter packing. That SK9 stabilizes mainly the packing of the protofibrils, preventing the sliding of the protofibrils sometimes seen in the control simulations, is also confirmed by visual inspection of the final configurations; some of them are shown as insets in Figure 9b.
As a result of this stabilization, the reduction in the solvent-accessible surface area (SASA), resulting from an interaction with the SK9 peptides, is more pronounced for the 6ZRF amylin fibril than for the 2F4L fibril model. This is also observed even when including the first seven residues (see Table 3) but again with little difference between hydrophobic and hydrophilic residues. Unlike for the 2F4L fibril, we also find a pronounced reduction in the packing distance (4.8(0.4) vs 5.9(1.4) Å in the control), which, however, is not seen in the distance between layers (3.0(0.2) vs 2.9(0.2) Å in the control); see Table 3. The differences in SASA values are consistent with a change in the packing free energy of ΔΔG = −44.9 (49.1) kJ/mol between the two protofibrils and ΔΔG = −41.5 (34.4) kJ/mol in the stacking free energy between two layers; see Table 3. The lower binding free energies result from an increase in the number of packing contacts and layer-wise stacking contacts and hydrogen bonds; see Table 3. Note also the additional intrachain hydrogen bond observed in the presence of SK9 that further stabilizes the amylin chain folds. As for the 2F4L fibril model, we find that the presence of SK9 leads to a decrease in the number of different residue pairs that form packing hydrogen bonds: 15 such pairs contribute at least 1% in the control, while, in the presence of SK9, the number is reduced to 5. The F23–A25 bond is dominant in both cases and contributes to approximately 70% of the packing hydrogen bonds. Interestingly, instead of the many kinds of transiently formed packing hydrogen bonds in the control, we find, in the presence of SK9, a strong presence of N21–Y37 pairs. These pairs contribute 19% of the packing hydrogen bonds (instead of 2% in the control), and almost 90% of packing hydrogen bonds are of this type. As a consequence, the map of packing contacts in Figure S5b is more focused on contacts between residues in segment S20–A25 on one fold with residues N35–Y37 on the opposite fold than seen for the control (see Figure S5a with the difference shown in Figure S5c). Similarly, the frequency of stacking contacts, especially such involving residues A8–N14, also increases with the binding of SK9; see Figure S5d,e. These interactions stabilized the N-terminal β-sheets; however, the effect is less pronounced than for the packing and goes with a free energy-raising twisting of these sheets. Hence, the increased frequency of packing and stacking contacts is consistent with the MMGBSA estimates, which indicate that, in the presence of SK9, both elongation and packing are favored more than in the control with the effect slightly more pronounced for the packing of the protofibrils.
Table 3. Various Quantities Averaged over the Last 100 ns of Three Trajectories in Simulations of the Amylin 6ZRF Fibril Model in the Absence or Presence of SK9 or FI10
 controlw/ SK9w/ FI10
total SASA (Å2)12,779 (112)12,040 (267)11,861 (284)
hydrophobic SASA (Å2)6451 (160)6090 (85)6063 (263)
hydrophilic SASA (Å2)6328 (68)5950 (265)5798 (129)
packing distance (Å)6 (1)4.8 (4)4.4 (1)
stacking distance (Å)2.9 (2)3.0 (2)2.9 (2)
packing contacts54 (4)59 (5)62 (2)
stacking contacts503 (10)537 (14)525 (13)
intrachain contacts1254 (2)1255 (22)1256 (13)
packing hydrogen bonds3 (1)2 (1)2 (1)
stacking hydrogen bonds134 (4)139 (5)143 (7)
intrachain hydrogen bonds22 (0)24 (1)25 (2)
packing free energy (kJ/mol)–152.6 (57.2)–197.5 (39.3)–222.6 (24.2)
stacking free energy (kJ/mol)–620.1 (38.5)–661.6 (29.6)–624.3 (50.9)
Comparing the effect of FI10 with SK9 on the 6ZRF fibril, we note that the FI10 segments form 229 contacts (29 contacts per amylin chain) in the start configuration, which is fewer initial contacts than SK9. However, these contacts are more stable, and on average, 139(6) contacts (approximately 61% of the original number) are observed over the last 100 ns, leading to a binding free energy between FI10 and amylin of −29 (3) kJ/mol that is only marginally lower than for SK9. The residue-wise distribution of contacts in Figure 7b is broader than for SK9. Binding sites such as R11-A13 (centered around L12) and H18 are observed more frequently, while F23 and L27 have less pronounced peaks. Residue Y37 has an even higher binding probability to FI10 than to SK9. These contacts are again formed by hydrophobic interactions, π–π stacking, and transiently formed hydrogen bonds (with N14@O-SK9:Y5@HN, L16@O-SK9:Y3@HN, and L27@O-SK9:Y3@HN and L16@O-SK9:Y3@HN replaced by L16@O-FI10:K2@HN, F23@O-SK9:Y3@HN by F23@O-FI10:G6@HN, and Y37@O-SK9:S1@H3 by two new ones: Y37@O-FI10:F1@HN and Y37@HN-FI10:F1@O). The formation of the additional hydrogen bond leads to the larger binding probability for residue Y37. FI10, unlike SK9, has a negatively charged residue that may reduce the flexibility of N-terminal residues by interacting with the positively charged R11. Similarly, interaction of the negatively charged residue on FI10 with H18 could reduce fibril-destabilizing stress on the chain geometry appearing when the histidine is in the charged state.
Overall, we see a stronger interaction between FI10 and the fibril than between SK9 and the fibril. The RMSF plot of Figure 8b indicates that the resulting reduction in residue flexibility, when compared to the control, is slightly larger than for SK9, indicating larger stabilization of the fibril. This observation is supported by Figure 9b where the RMSD is lower than for SK9 binding to the fibril. Visual inspections of the final configuration, shown as inset of the figure, suggests that, similar to SK9, the effect of FI10 on the fibril stability is again mainly by stabilizing packing, and no sliding of the protofibrils was observed.
Consistent with this assumption is that the reduction in the solvent-accessible surface area upon interaction with the viral protein fragment is approximately 180 Å2 larger for FI10 than for SK9 when ignoring the disordered first seven residues. The additional reduction of SASA results mostly from hydrophilic residues; see Table 3. The packing distance (4.4(1) Å) and the inter-layer distances (2.9(2) Å) are likewise smaller than in the case of SK9, indicating that the larger stabilization of the amylin fibril by FI10, when compared to SK9, results from increased packing and stacking interactions. This can be also seen from our MMPBSA approximations of packing and stacking free energies, which show that packing of the protofibrils in the presence of FI10 is more favorable by ΔΔG = −70.0 (43.9) kJ/mol and the stacking of layers by ΔΔG = −3.6(45.2) kJ/mol per chain. The large standard deviation in the stacking free energy difference makes it difficult to compare these results with those of SK9, but it appears that, unlike for SK9, the stabilization of the fibril is mostly due to improving the packing of the protofibrils. This is consistent with the larger number of packing contacts, exceeding that seen in control or when SK9 interacts with fibril. On the other hand, differences between SK9 and FI10 are marginal for the stacking contacts and hydrogen bonds; see Table 3. However, similar to SK9, the presence of FI10 leads to a reduction in the number of packing hydrogen-bond-forming residue pairs from 15 in the control to 4 pairs with the N21–Y37 pairs now contributing 23% of the packing hydrogen bonds and the F23–A25 hydrogen bonds contributing 72%. Hence, 95% of packing hydrogen bonds result from only two types of hydrogen bonds; see also the map of packing contacts in Figure S6a–c. While we also see an increase in stacking contacts involving residues A8–N14 (see Figure S6d,e), the fibril-stabilizing effect is even more than for SK9 due to enhancing the packing of the two protofibrils.
How do our results depend on our simulation set-up? For instance, both virus protein fragments have in our simulations a NH3+ group at the N-terminus and a CONH2 group at the C-terminus. This is due to reduced electrostatic interactions between the opposite charges at the termini. In order to test how our results depend on the choice of end group, we have also simulated a system where FI10 is not capped at the C-terminus, that is, it ends with a COO group. We compare in Figure S7 the RMSD as a function of time for the control and the amylin fibril interacting with the two differently capped FI10 peptides. In both cases, the RMSD is lower than for the control, and therefore the values do not seem to depend on the end group. However, in the RMSF plot of Figure S8, values of the FI10 peptide with the uncapped C-terminus are higher than for the capped one but still lower than of the control. Other quantities such as the SASA (12360(502) Å2) are closer to the values found for the control, (12,779(112) Å2) than for the fibril in the presence of the capped FI10 peptides (11,861(284) Å2). Unlike for the capped peptide, the differences result mostly from hydrophilic residues. Only residues 8–37 were considered in calculating the above SASA values. When including the disordered first seven residues, the values for control (16,848(257) Å2) and the fibril in the presence of the uncapped peptide (16,834(459) Å2) agree within the error bars while still lower for the fibril in the presence of the capped FI10 (16,283(577) Å2). This may be because the uncapped peptide forms more compact configurations as its average end-to-end distance of 20.6(9) Å is smaller than the 23(1) Å of the capped FI10. Therefore, the uncapped peptide binds less tightly to the amylin fibril than the capped one. This can be seen with the number of contacts in the last 100 ns being only approximately 116(6), that is, 51% of the original number of contacts seen for the uncapped FI10 versus 139(6), that is, 61% for the capped peptide. The residue-wise binding frequencies in Figure S9 show a similar but broader distribution for the capped peptide with, for the N-terminal residues K1–H18, less binding propensity. Correspondingly, we find that the difference in binding free energies, that is, ΔΔG = −10.5 (2.3) kJ/mol, resulting from the different end groups, indicates less of an interaction between the uncapped virus protein fragments and the amylin fibril. Hence, we conclude that capping of the termini reduces self-interaction in FI10, leading to more stretched conformations, which in turn interact more strongly with the amylin fibril. This effect, however, does not seem to reduce the overall stabilization of the fibril.

Conclusions

ARTICLE SECTIONS
Jump To

Using extensive molecular dynamics simulations, we have studied the effect of two virus protein fragments from the envelope protein (SK9) and from the spike protein (FI10) of SARS-COV-2 on the ensemble of amylin monomer conformations and on the stability of amylin fibrils. Such interactions may be correlated with the observed onset of type-II diabetes following COVID-19. (5−8) While both virus protein fragments bind to the amylin monomer structure, we do not find the induced shift toward more aggregation-prone conformations that we saw previously in our studies with Serum Amyloid A9 and α-synuclein. (10) Unlike these proteins, amylin has, even as a monomer, a more defined structure, and the mostly helical fold appears to be stabilized by the presence of SK9 or FI10. Indications for such a protective effect seems to be seen also in experiments by the Huang Lab at HUKST that are currently being analyzed (Jinqing Hiuang, private communication).
However, the presence of amylin aggregates, which are proposed to cause symptoms of type-II diabetes, can also be raised by stabilizing existing fibrils, shifting the equilibrium from functional monomers toward the toxic amyloids. We indeed see such stabilization of amylin fibrils by the viral protein fragments. Similar to what we found in our previous work, this stabilizing effect depends on the fibril model. Hence, as we also observed earlier for α-synuclein, (10) interactions with the viral protein fragments may shift the equilibrium between various fibril polymorphs. Fibril stabilization is observed for both SK9 and FI10 and depends only little on details of our simulation set-up. For instance, the uncapped peptide binds by approximately 10 kJ/mol less tightly to the amylin fibril than the capped one, leading to different values in the measured quantities without changing the picture qualitatively. The stabilization of the fibril geometry is mainly by enhancing the packing of protofibrils and less by encouraging the elongation of fibrils. Interestingly, the effect is stronger for FI10. This segment is unique for SARS-COV-2 and cleaved from the spike protein by the enzyme neutrophil elastase released from neutrophils during acute inflammation. In vitro, the FI10 fragment forms amyloids. (15) While not yet supported by experimental evidence, we speculate that SARS-COV-2 infections, often causing chronic inflammation, increase the enzymatic cleavage of the spike protein, releasing, in some cases, amyloidogenic fragments such as FI10. These fragments could then cross-seed amylin fibrils, which in turn may contribute to an onset of type-II diabetes.
Comparing our results with earlier work, we find as a common theme that small SARS-COV-2 protein fragments (as potentially derived by cleavage during infection-caused inflammation) can seed aggregation of amyloidogenic human proteins but this effect depends both on the viral fragment and the structure of the fibril polymorph.

Materials and Methods

ARTICLE SECTIONS
Jump To

System Preparation

We rely on molecular dynamics (MD) simulations to study the effect of SARS-COV-2 proteins on amylin aggregation. Our study focuses on the change in the conformational ensemble and the stability of the amylin monomer and fibrils induced by the presence of the nine-residue segment SFYVYSRVK (SK9) of the envelope protein or the ten-residue segment FKNIDGYFKI (FI10) of the spike protein from SARS-COV-2. The initial configuration of SK9 is the same as the one used in our previous work (9,10) and was derived from the model of the SARS-COV-2 envelope protein deposited in the MOLSSI COVID-19 hub. (23) The N- and C-terminals of the SK9 segment are capped by a NH3+ and −CONH2 group, respectively. Following Nyström and Hammarström, (15) the initial configuration of the FI10 viral fragment was derived by extracting the residues F194–I203 from the structure of the SARS-COV-2 spike protein (PDB ID: 6VXX) with the N- and C-terminals of the peptide are capped again by a NH3+ and −CONH2 group, respectively. To compare the effect of the terminal group of the small FI10 peptide on our results, we have also considered the case where the N- and C-termini were capped with a NH3+ and −COO group, respectively.
The initial conformation of the amylin monomer model was taken from the Protein Data Bank (PDB). This structure (PDB ID: 2L86) (11) was resolved by solution NMR, and we add at the N- and C-terminals a NH3+ and a COO group, respectively. Unfortunately, only a few models of amylin fibrils have been resolved and published in the PDB. Due to our familiarity with the fibril model put forward by Eisenberg and co-workers, (20) which we have studied in a different context in previous work, (21) we again chose this model but reduced the number of layers from five to four chains in each of the two protofibrils. This is because we expect that, by decreasing the layers in this manner, the stability of the fibril fragment will be lowered and any potential effect resulting from the presence of SK9 will be seen more clearly. Note that, while we call this amylin fibril model 2F4L, this is not a PDB ID but rather our shorthand notation for “two folds, four layers”. In order to be consistent with earlier work, (21) we add again at the N- and C-terminals a NH3+ and a COO group. The initial conformation of the other amylin fibril model was extracted from the Cryo-EM resolved model (22) with PDB ID 6ZRF such that it has the same number of chains as the 2F4L fibril model. Note that we use −COCH3 and −NHCH3 as end groups for this second fibril model, allowing us to compare our control simulations (in the absence of viral protein fragments) with the experiments described in ref (24). The use of different end groups is justified as our main interest is the comparison of simulations with and without SARS-COV-2 protein fragments.
Simulations starting from the amylin monomer and fibril models described above serve as the control in our study. For investigating the effect of SARS-COV-2 protein fragments on amylin, we have performed simulations that start from configurations where either SK9 or FI10 peptides bind to these models. These start configurations were generated using AutoDock Vina (12) and HADDOCK (13,14) by docking the respective segments in a ratio of 1:1 with either the monomer or fibrils. The resulting configurations are shown in Figure S1. Note that the viral protein segments are not restricted to these initial positions on the monomer or fibrils but can move freely throughout the simulations and even detach from the amylin.

General Simulation Protocol

The set-up and simulation of all systems rely on the GROMACS 2018 and GROMACS 2022 package. (25) We use the CHARMM 36m all-atom force-field (26) with TIP3P explicit water (27) as implemented in the GROMACS package to describe the inter- and intramolecular interactions for the monomer and fibrils. This force-field and water model combination has been found to be well-suited for simulations of fibrils and oligomers in previous work performed in our group (21,28) and in the literature. (29−31) Hydrogen atoms are added with the pdb2gmx module of the GROMACS suite. (25) The start configurations for all systems are put in the center of the cubic box with at least 15 Å between the solute and the edge of the box. Periodic boundary conditions are employed. The systems are solvated with water molecules, and counterions are added to neutralize the system with the Na+ and Cl ions at a physiological ion concentration of 150 mM NaCl. Both the number of water molecules and the total number of atoms are listed in Table 4. The energy of each system is minimized by steepest decent for up to 50,000 steps, and afterward, the system is equilibrated at 310 K for 200 ps at a constant volume and an additional 200 ps at a constant pressure (1 atm), constraining the positions of heavy atoms with a force constant of 1000 kJ/(mol nm2).
Table 4. Simulated Systems
system descriptionatomswater moleculesindependent trajectoriessimulation length (ns)total sampling (ns)
monomer (PDB ID: 2L86)
control26,74787203400012,000
w/ SK9 (AD-Vina)26,7678733310003000
w/ SK9 (HADDOCK)25,8568430310003000
w/ FI1030,41610,083310003000
fibril model 2F4L
control118,16937,8843200600
w/ SK9120,14838,0903200600
fibril model 6ZRF
control149,71048,3563200600
w/ SK9268,41187,3613200600
w/ FI10-CONH2268,42387,3733200600
w/ FI10-COO268,43387,4033200600
Starting from the above-generated initial conformations (shown in Figure S1), simulations are performed at a constant temperature (310 K) and pressure (1 atm). The temperature is controlled by a v-rescale thermostat (32) with a coupling constant of 0.1 ps, and the pressure is kept constant using the Parrinello–Rahman barostat (33) with a pressure relaxation time of 2 ps. By using the SETTLE algorithm (34) to keep water molecules rigid and restraining protein bonds involving hydrogen atoms to their equilibrium length with the LINCS algorithm, (35) we are able to use a time step of 2 fs for integrating the equations of motion. Because of the periodic boundary conditions, we compute the long-range electrostatic interactions with the particle-mesh Ewald (PME) technique using a real-space cutoff of 12 Å and a Fourier grid spacing of 1.6 Å. Short-range van der Waal interactions are truncated at 12 Å with smoothing starting at 10.5 Å. For each system, we follow three independent trajectories differing by their initial velocity distributions. The lengths of the various trajectories are also listed in Table 4.

Trajectory Analysis

The molecular dynamics trajectories are analyzed with the GROMACS tools, (25) VMD, (36) and MDTraj software. (37) For visualization of trajectory conformations, we use VMD (36) and UCSF Chimera. (38) The root mean square deviation (RMSD), root mean square fluctuation (RMSF), radius of gyration (RGY), and solvent-accessible surface area (SASA) are calculated using GROMACS tools (for the latter quantity, using a spherical probe of a 1.4 Å radius). The residue-wise contact frequencies are calculated using VMD and MDTraj software, defining contacts by a cutoff of 4.5 Å in the closest distance between heavy atoms in a residue pair. Hydrogen bonds are defined by a distance cutoff of 3.5 Å between the donor and acceptor atoms and an angle cutoff of 30°. The residue-wise secondary structure propensity is calculated in VMD by the STRIDE algorithm. (39)
In order to avoid, in our case, the prohibitive computational costs of free energy perturbation methods, thermodynamic integration or similar approaches are binding free energies of the viral protein fragments to amylin fibrils approximated by the molecular mechanics Poisson–Boltzmann surface area (MM-PBSA) method as implemented in the GROMACS suite. (25,40) This approximation is not without problems with one important drawback being that it neglects entropic contributions of the solute. As MM-PBSA is therefore not suitable for studying the binding of the viral protein fragments SK9 and FI10 to amylin monomers, we use for these systems instead the approach described in the work of Bellaiche and Best. (41) This approach was used earlier to study the inhibition of Aβ amyloid formation by small peptides (42) and relies on calculating a dimensionless association constant from the distribution of bound and unbound conformations. For comparison, we also estimate binding free energies with the protein binding energy predictive model (PRODIGY), which is based on a simple linear regression of the number of interfacial contacts between residues in a protein–protein complex. (16)

Supporting Information

ARTICLE SECTIONS
Jump To

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.3c00621.

  • Residue-wise binding frequencies of SK9 and FI10 to amylin monomers, start configurations with SK9 and FI10 docked, average center of the mass distance of the amylin monomer and SK9 and FI10, residue-wise binding frequencies of the amylin monomer to SK9 and FI10, contact maps of amylin fibril models in the absence and presence of SK9 and FI10, RMSD and RMSF of fibril model 6ZRF in the presence of FI10 capped/uncapped, and residue-wise contact frequencies of fibril model 6ZRF with FI10 capped/uncapped (PDF)

  • Initial and final structures of all trajectories (ZIP)

Terms & Conditions

Most electronic Supporting Information files are available without a subscription to ACS Web Editions. Such files may be downloaded by article for research use (if there is a public use license linked to the relevant article, that license may permit other uses). Permission may be obtained from ACS for other uses through requests via the RightsLink permission system: http://pubs.acs.org/page/copyright/permissions.html.

Author Information

ARTICLE SECTIONS
Jump To

  • Corresponding Authors
  • Author
    • Andrew D. Chesney - Department of Chemistry & Biochemistry, University of Oklahoma, Norman, Oklahoma 73019, United States
  • Notes
    The authors declare no competing financial interest.

Acknowledgments

ARTICLE SECTIONS
Jump To

Our simulations were done using the SCHOONER cluster of the University of Oklahoma, XSEDE resources allocated under grant no. MCB160005 (National Science Foundation), and TACC resources allocated under grant no. MCB20016 (National Science Foundation). We acknowledge financial support from the National Institutes of Health under grant no. GM120634. We thank Asis K. Jana for the help in the early stages of this project, mainly setting up and running the 2F4L fibril simulations and early parts of the trajectories of amylin monomers interacting with SK9.

References

ARTICLE SECTIONS
Jump To

This article references 42 other publications.

  1. 1
    Shi, S.; Qin, M.; Shen, B.; Cai, Y.; Liu, T.; Yang, F.; Gong, W.; Liu, X.; Liang, J.; Zhao, Q.; Huang, H.; Yang, B.; Huang, C. Association of Cardiac Injury With Mortality in Hospitalized Patients With COVID-19 in Wuhan, China. JAMA Cardiol. 2020, 5, 802,  DOI: 10.1001/jamacardio.2020.0950
  2. 2
    Sheraton, M.; Deo, N.; Kashyap, R.; Surani, S. A Review of Neurological Complications of COVID-19. Cureus 2020,  DOI: 10.7759/cureus.8192
  3. 3
    Su, H.; Yang, M.; Wan, C.; Yi, L.-X.; Tang, F.; Zhu, H.-Y.; Yi, F.; Yang, H.-C.; Fogo, A. B.; Nie, X.; Zhang, C. Renal Histopathological Analysis of 26 Postmortem Findings of Patients with COVID-19 in China. Kidney Int. 2020, 98, 219227,  DOI: 10.1016/j.kint.2020.04.003
  4. 4
    Morris, S. B.; Schwartz, N. G.; Patel, P.; Abbo, L.; Beauchamps, L.; Balan, S.; Lee, E. H.; Paneth-Pollak, R.; Geevarughese, A.; Lash, M. K.; Dorsinville, M. S.; Ballen, V.; Eiras, D. P.; Newton-Cheh, C.; Smith, E.; Robinson, S.; Stogsdill, P.; Lim, S.; Fox, S. E.; Richardson, G.; Hand, J.; Oliver, N. T.; Kofman, A.; Bryant, B.; Ende, Z.; Datta, D.; Belay, E.; Godfred-Cato, S. Case Series of Multisystem Inflammatory Syndrome in Adults Associated with SARS-CoV-2 Infection ─ United Kingdom and United States, March–August 2020. MMWR Morb. Mortal. Wkly. Rep. 2020, 69, 14501456,  DOI: 10.15585/mmwr.mm6940e1
  5. 5
    Beliard, K. A.; Yau, M.; Wilkes, M.; Romero, C. J.; Wallach, E.; Rapaport, R. SARS-CoV-2 Infection Related Diabetes Mellitus. J. Endocr. Soc. 2021, 5, A397A397,  DOI: 10.1210/jendso/bvab048.808
  6. 6
    Hollstein, T.; Schulte, D. M.; Schulz, J.; Glück, A.; Ziegler, A. G.; Bonifacio, E.; Wendorff, M.; Franke, A.; Schreiber, S.; Bornstein, S. R.; Laudes, M. Autoantibody-Negative Insulin-Dependent Diabetes Mellitus after SARS-CoV-2 Infection: A Case Report. Nat. Metab. 2020, 2, 10211024,  DOI: 10.1038/s42255-020-00281-8
  7. 7
    Moreno, D. M.; Ramos, R. J. A.; Fernández, L. G.; Montenegro, A. M. R.; González, M. M.; Torrecilla, N. B.; Albarrán, O. G. Clinical/Biochemical Characteristics and Related Outcomes in People with New-onset Diabetes and COVID -19: Experience from a Single Centre. Pract. Diabetes 2022, 39, 2431,  DOI: 10.1002/pdi.2426
  8. 8
    Hayden, M. R. An Immediate and Long-Term Complication of COVID-19 May Be Type 2 Diabetes Mellitus: The Central Role of β-Cell Dysfunction, Apoptosis and Exploration of Possible Mechanisms. Cell 2020, 9, 2475,  DOI: 10.3390/cells9112475
  9. 9
    Jana, A. K.; Greenwood, A. B.; Hansmann, U. H. E. Presence of a SARS-CoV-2 Protein Enhances Amyloid Formation of Serum Amyloid A. J. Phys. Chem. B 2021, 125, 91559167,  DOI: 10.1021/acs.jpcb.1c04871
  10. 10
    Jana, A. K.; Lander, C. W.; Chesney, A. D.; Hansmann, U. H. E. Effect of an Amyloidogenic SARS-COV-2 Protein Fragment on α-Synuclein Monomers and Fibrils. J. Phys. Chem. B 2022, 126, 36483658,  DOI: 10.1021/acs.jpcb.2c01254
  11. 11
    Nanga, R. P. R.; Brender, J. R.; Vivekanandan, S.; Ramamoorthy, A. Structure and Membrane Orientation of IAPP in Its Natively Amidated Form at Physiological PH in a Membrane Environment. Biochim. Biophys. Acta BBA - Biomembr. 2011, 1808, 23372342,  DOI: 10.1016/j.bbamem.2011.06.012
  12. 12
    Trott, O.; Olson, A. J. AutoDock Vina: Improving the Speed and Accuracy of Docking with a New Scoring Function, Efficient Optimization, and Multithreading. J. Comput. Chem. 2010, NANA,  DOI: 10.1002/jcc.21334
  13. 13
    van Zundert, G. C. P.; Rodrigues, J. P. G. L. M.; Trellet, M.; Schmitz, C.; Kastritis, P. L.; Karaca, E.; Melquiond, A. S. J.; van Dijk, M.; de Vries, S. J.; Bonvin, A. M. J. J. The HADDOCK2.2 Web Server: User-Friendly Integrative Modeling of Biomolecular Complexes. J. Mol. Biol. 2016, 428, 720725,  DOI: 10.1016/j.jmb.2015.09.014
  14. 14
    Honorato, R. V.; Koukos, P. I.; Jiménez-García, B.; Tsaregorodtsev, A.; Verlato, M.; Giachetti, A.; Rosato, A.; Bonvin, A. M. J. J. Structural Biology in the Clouds: The WeNMR-EOSC Ecosystem. Front. Mol. Biosci. 2021, 8, 729513,  DOI: 10.3389/fmolb.2021.729513
  15. 15
    Nyström, S.; Hammarström, P. Amyloidogenesis of SARS-CoV-2 Spike Protein; preprint. J. Am. Chem. Soc. 2022, 144, 89458950,  DOI: 10.1021/jacs.2c03925
  16. 16
    Xue, L. C.; Rodrigues, J. P.; Kastritis, P. L.; Bonvin, A. M.; Vangone, A. PRODIGY: A Web Server for Predicting the Binding Affinity of Protein–Protein Complexes. Bioinformatics 2016, 3676,  DOI: 10.1093/bioinformatics/btw514
  17. 17
    Westermark, P.; Engström, U.; Johnson, K. H.; Westermark, G. T.; Betsholtz, C. Islet Amyloid Polypeptide: Pinpointing Amino Acid Residues Linked to Amyloid Fibril Formation. Proc. Natl. Acad. Sci. 1990, 87, 50365040,  DOI: 10.1073/pnas.87.13.5036
  18. 18
    Jaikaran, E. T. A. S.; Higham, C. E.; Serpell, L. C.; Zurdo, J.; Gross, M.; Clark, A.; Fraser, P. E. Identification of a Novel Human Islet Amyloid Polypeptide β-Sheet Domain and Factors Influencing Fibrillogenesis. J. Mol. Biol. 2001, 308, 515525,  DOI: 10.1006/jmbi.2001.4593
  19. 19
    Thu, T. T. M.; Li, M. S. Protein Aggregation Rate Depends on Mechanical Stability of Fibrillar Structure. J. Chem. Phys. 2022, 157, 055101  DOI: 10.1063/5.0088689
  20. 20
    Wiltzius, J. J. W.; Sievers, S. A.; Sawaya, M. R.; Cascio, D.; Popov, D.; Riekel, C.; Eisenberg, D. Atomic Structure of the Cross-β Spine of Islet Amyloid Polypeptide (Amylin). Protein Sci. Publ. Protein Soc. 2008, 17, 14671474,  DOI: 10.1110/ps.036509.108
  21. 21
    Pandey, P.; Nguyen, N.; Hansmann, U. H. E. d -Retro Inverso Amylin and the Stability of Amylin Fibrils. J. Chem. Theory Comput. 2020, 16, 53585368,  DOI: 10.1021/acs.jctc.0c00523
  22. 22
    Gallardo, R.; Iadanza, M. G.; Xu, Y.; Heath, G. R.; Foster, R.; Radford, S. E.; Ranson, N. A. Fibril Structures of Diabetes-Related Amylin Variants Reveal a Basis for Surface-Templated Assembly. Nat. Struct. Mol. Biol. 2020, 27, 10481056,  DOI: 10.1038/s41594-020-0496-3
  23. 23
    Kryshtafovych, A.; Moult, J.; Billings, W. M.; Della Corte, D.; Fidelis, K.; Kwon, S.; Olechnovič, K.; Seok, C.; Venclovas, Č.; Won, J.; CASP-COVID participants Modeling SARS-CoV-2 Proteins in the CASP-commons Experiment. Proteins Struct. Funct. Bioinforma. 2021, 89, 19871996,  DOI: 10.1002/prot.26231
  24. 24
    Mesias, V. S. D.; Zhu, H.; Tang, X.; Dai, X.; Liu, W.; Guo, Y.; Huang, J. Moderate Binding between Two SARS-CoV-2 Protein Segments and α-Synuclein Alters Its Toxic Oligomerization Propensity Differently. J. Phys. Chem. Lett. 2022, 13, 1064210648,  DOI: 10.1021/acs.jpclett.2c02278
  25. 25
    Abraham, M. J.; Murtola, T.; Schulz, R.; Páll, S.; Smith, J. C.; Hess, B.; Lindahl, E. GROMACS: High Performance Molecular Simulations through Multi-Level Parallelism from Laptops to Supercomputers. SoftwareX 2015, 1-2, 1925,  DOI: 10.1016/j.softx.2015.06.001
  26. 26
    Huang, J.; Rauscher, S.; Nawrocki, G.; Ran, T.; Feig, M.; de Groot, B. L.; Grubmüller, H.; MacKerell, A. D. CHARMM36m: An Improved Force Field for Folded and Intrinsically Disordered Proteins. Nat. Methods 2017, 14, 7173,  DOI: 10.1038/nmeth.4067
  27. 27
    Jorgensen, W. L.; Chandrasekhar, J.; Madura, J. D.; Impey, R. W.; Klein, M. L. Comparison of Simple Potential Functions for Simulating Liquid Water. J. Chem. Phys. 1983, 79, 926935,  DOI: 10.1063/1.445869
  28. 28
    Wang, W.; Hansmann, U. H. E. Stability of Human Serum Amyloid A Fibrils. J. Phys. Chem. B 2020, 124, 1070810717,  DOI: 10.1021/acs.jpcb.0c08280
  29. 29
    Siwy, C. M.; Lockhart, C.; Klimov, D. K. Is the Conformational Ensemble of Alzheimer’s Aβ10-40 Peptide Force Field Dependent?. PLoS Comput. Biol. 2017, 13, e1005314  DOI: 10.1371/journal.pcbi.1005314
  30. 30
    Samantray, S.; Yin, F.; Kav, B.; Strodel, B. Different Force Fields Give Rise to Different Amyloid Aggregation Pathways in Molecular Dynamics Simulations. J. Chem. Inf. Model. 2020, 60, 64626475,  DOI: 10.1021/acs.jcim.0c01063
  31. 31
    Man, V. H.; He, X.; Derreumaux, P.; Ji, B.; Xie, X.-Q.; Nguyen, P. H.; Wang, J. Effects of All-Atom Molecular Mechanics Force Fields on Amyloid Peptide Assembly: The Case of Aβ 16–22 Dimer. J. Chem. Theory Comput. 2019, 15, 14401452,  DOI: 10.1021/acs.jctc.8b01107
  32. 32
    Bussi, G.; Donadio, D.; Parrinello, M. Canonical Sampling through Velocity Rescaling. J. Chem. Phys. 2007, 126, 014101  DOI: 10.1063/1.2408420
  33. 33
    Parrinello, M.; Rahman, A. Polymorphic Transitions in Single Crystals: A New Molecular Dynamics Method. J. Appl. Phys. 1981, 52, 71827190,  DOI: 10.1063/1.328693
  34. 34
    Miyamoto, S.; Kollman, P. A. Settle: An Analytical Version of the SHAKE and RATTLE Algorithm for Rigid Water Models. J. Comput. Chem. 1992, 13, 952962,  DOI: 10.1002/jcc.540130805
  35. 35
    Hess, B.; Bekker, H.; Berendsen, H. J. C.; Fraaije, J. G. E. M. LINCS: A Linear Constraint Solver for Molecular Simulations. J. Comput. Chem. 1997, 18, 14631472,  DOI: 10.1002/(SICI)1096-987X(199709)18:12<1463::AID-JCC4>3.0.CO;2-H
  36. 36
    Humphrey, W.; Dalke, A.; Schulten, K. VMD: Visual Molecular Dynamics. J. Mol. Graph. 1996, 14, 3338,  DOI: 10.1016/0263-7855(96)00018-5
  37. 37
    McGibbon, R. T.; Beauchamp, K. A.; Harrigan, M. P.; Klein, C.; Swails, J. M.; Hernández, C. X.; Schwantes, C. R.; Wang, L.-P.; Lane, T. J.; Pande, V. S. MDTraj: A Modern Open Library for the Analysis of Molecular Dynamics Trajectories. Biophys. J. 2015, 109, 15281532,  DOI: 10.1016/j.bpj.2015.08.015
  38. 38
    Pettersen, E. F.; Goddard, T. D.; Huang, C. C.; Couch, G. S.; Greenblatt, D. M.; Meng, E. C.; Ferrin, T. E. UCSF Chimera?A Visualization System for Exploratory Research and Analysis. J. Comput. Chem. 2004, 25, 16051612,  DOI: 10.1002/jcc.20084
  39. 39
    Frishman, D.; Argos, P. Knowledge-Based Protein Secondary Structure Assignment. Proteins: Struct., Funct., Genet. 1995, 23, 566579,  DOI: 10.1002/prot.340230412
  40. 40
    Kumari, R.; Kumar, R.; Open Source Drug Discovery Consortium; Lynn, A. G_mmpbsa ─A GROMACS Tool for High-Throughput MM-PBSA Calculations. J. Chem. Inf. Model. 2014, 54, 19511962,  DOI: 10.1021/ci500020m
  41. 41
    Bellaiche, M. M. J.; Best, R. B. Molecular Determinants of Aβ 42 Adsorption to Amyloid Fibril Surfaces. J. Phys. Chem. Lett. 2018, 9, 64376443,  DOI: 10.1021/acs.jpclett.8b02375
  42. 42
    Leguizamon Herrera, V. L.; Buell, A. K.; Willbold, D.; Barz, B. Interaction of Therapeutic d -Peptides with Aβ42 Monomers, Thermodynamics, and Binding Analysis. ACS Chem. Neurosci. 2022, 13, 16381650,  DOI: 10.1021/acschemneuro.2c00102

Cited By

ARTICLE SECTIONS
Jump To

This article is cited by 1 publications.

  1. Asis K. Jana, Recep Keskin, Fatih Yaşar. Molecular Insight into the Effect of HIV-TAT Protein on Amyloid-β Peptides. ACS Omega 2024, Article ASAP.
  • Abstract

    Figure 1

    Figure 1. Licorice representation of the two SARS-COV-2 protein fragments used in this study: (a) SK9 and (b) FI10. Ribbon representations of the (c) amylin monomer (PDB ID: 2L86) and the two fibril models with (d) 2F4L and (e) 6ZRF are also shown. The N- and C-terminals are colored in blue and red, respectively.

    Figure 2

    Figure 2. (a) Averaged number of original contacts between SK9 and amylin as a function of time. (b) Corresponding average number of all contacts (including newly formed ones). Contacts are normalized to 1 for the respective start configurations. Red marks the data from runs where the initial binding site of SK9 was found with AutoDock Vina; green is where the binding site was found with HADDOCK. Blue marks the data from runs where FI10 interacts with amylin.

    Figure 3

    Figure 3. Average root mean square deviation (RMSD) to the start conformation as a function of time for amylin in the presence of SK9 (red) or FI10 (blue) and in the absence of SK9 and FI10 (black). The RMSD is evaluated over backbone atoms only. Averages are calculated over all trajectories for each system.

    Figure 4

    Figure 4. Representative snapshots of the final amylin conformations in the (a) absence and presence of (b) SK9 or (c) FI10. The N- and C-terminals are colored in blue and red, respectively.

    Figure 5

    Figure 5. Residue-wise root mean square fluctuation of the amylin monomer in the absence (black) or presence of SK9 (red) or FI10 (blue) calculated over the (a) last 800 ns and (b) last 200 ns of the simulations. Shaded regions mark the standard deviations of the shown quantities.

    Figure 6

    Figure 6. Residue-wise helicity of the amylin monomer in the presence of SK9 (red) and FI10 (blue) calculated over the last 200 ns. Shown is the difference to the corresponding values in the control simulations where the viral protein fragments are absent.

    Figure 7

    Figure 7. Residue-wise frequency of contacts of (a) the fibril model 2F4L with SK9 and (b) the fibril model 6ZRF with either SK9 (red) or FI10 (blue). Shaded regions mark the standard deviation of the averages.

    Figure 8

    Figure 8. Residue-wise RMSF for (a) the fibril model 2F4L and (b) the fibril model 6ZRF. The black curve marks data from the simulation in the absence of viral protein fragments, while the red curve shows the results obtained when SK9 is present, and the blue curve data is for when FI10 is interacting with the amylin fibrils. Shaded regions mark the standard deviation of the averages.

    Figure 9

    Figure 9. RMSD to the start configuration as function of time for (a) the fibril model 2F4L and (b) the fibril model 6ZRF. The RMSD is calculated over all backbone atoms in residues 8–37 of all chains in the respective fibril model. Black curves are from the control simulations, while red curves are from data measured in simulations where SK9 is present, and similarly, blue curves are for the case of FI10 interacting with the amylin fibril. Shaded regions mark the standard deviation of the averages. The start configuration and representative final configurations are shown as insets.

  • References

    ARTICLE SECTIONS
    Jump To

    This article references 42 other publications.

    1. 1
      Shi, S.; Qin, M.; Shen, B.; Cai, Y.; Liu, T.; Yang, F.; Gong, W.; Liu, X.; Liang, J.; Zhao, Q.; Huang, H.; Yang, B.; Huang, C. Association of Cardiac Injury With Mortality in Hospitalized Patients With COVID-19 in Wuhan, China. JAMA Cardiol. 2020, 5, 802,  DOI: 10.1001/jamacardio.2020.0950
    2. 2
      Sheraton, M.; Deo, N.; Kashyap, R.; Surani, S. A Review of Neurological Complications of COVID-19. Cureus 2020,  DOI: 10.7759/cureus.8192
    3. 3
      Su, H.; Yang, M.; Wan, C.; Yi, L.-X.; Tang, F.; Zhu, H.-Y.; Yi, F.; Yang, H.-C.; Fogo, A. B.; Nie, X.; Zhang, C. Renal Histopathological Analysis of 26 Postmortem Findings of Patients with COVID-19 in China. Kidney Int. 2020, 98, 219227,  DOI: 10.1016/j.kint.2020.04.003
    4. 4
      Morris, S. B.; Schwartz, N. G.; Patel, P.; Abbo, L.; Beauchamps, L.; Balan, S.; Lee, E. H.; Paneth-Pollak, R.; Geevarughese, A.; Lash, M. K.; Dorsinville, M. S.; Ballen, V.; Eiras, D. P.; Newton-Cheh, C.; Smith, E.; Robinson, S.; Stogsdill, P.; Lim, S.; Fox, S. E.; Richardson, G.; Hand, J.; Oliver, N. T.; Kofman, A.; Bryant, B.; Ende, Z.; Datta, D.; Belay, E.; Godfred-Cato, S. Case Series of Multisystem Inflammatory Syndrome in Adults Associated with SARS-CoV-2 Infection ─ United Kingdom and United States, March–August 2020. MMWR Morb. Mortal. Wkly. Rep. 2020, 69, 14501456,  DOI: 10.15585/mmwr.mm6940e1
    5. 5
      Beliard, K. A.; Yau, M.; Wilkes, M.; Romero, C. J.; Wallach, E.; Rapaport, R. SARS-CoV-2 Infection Related Diabetes Mellitus. J. Endocr. Soc. 2021, 5, A397A397,  DOI: 10.1210/jendso/bvab048.808
    6. 6
      Hollstein, T.; Schulte, D. M.; Schulz, J.; Glück, A.; Ziegler, A. G.; Bonifacio, E.; Wendorff, M.; Franke, A.; Schreiber, S.; Bornstein, S. R.; Laudes, M. Autoantibody-Negative Insulin-Dependent Diabetes Mellitus after SARS-CoV-2 Infection: A Case Report. Nat. Metab. 2020, 2, 10211024,  DOI: 10.1038/s42255-020-00281-8
    7. 7
      Moreno, D. M.; Ramos, R. J. A.; Fernández, L. G.; Montenegro, A. M. R.; González, M. M.; Torrecilla, N. B.; Albarrán, O. G. Clinical/Biochemical Characteristics and Related Outcomes in People with New-onset Diabetes and COVID -19: Experience from a Single Centre. Pract. Diabetes 2022, 39, 2431,  DOI: 10.1002/pdi.2426
    8. 8
      Hayden, M. R. An Immediate and Long-Term Complication of COVID-19 May Be Type 2 Diabetes Mellitus: The Central Role of β-Cell Dysfunction, Apoptosis and Exploration of Possible Mechanisms. Cell 2020, 9, 2475,  DOI: 10.3390/cells9112475
    9. 9
      Jana, A. K.; Greenwood, A. B.; Hansmann, U. H. E. Presence of a SARS-CoV-2 Protein Enhances Amyloid Formation of Serum Amyloid A. J. Phys. Chem. B 2021, 125, 91559167,  DOI: 10.1021/acs.jpcb.1c04871
    10. 10
      Jana, A. K.; Lander, C. W.; Chesney, A. D.; Hansmann, U. H. E. Effect of an Amyloidogenic SARS-COV-2 Protein Fragment on α-Synuclein Monomers and Fibrils. J. Phys. Chem. B 2022, 126, 36483658,  DOI: 10.1021/acs.jpcb.2c01254
    11. 11
      Nanga, R. P. R.; Brender, J. R.; Vivekanandan, S.; Ramamoorthy, A. Structure and Membrane Orientation of IAPP in Its Natively Amidated Form at Physiological PH in a Membrane Environment. Biochim. Biophys. Acta BBA - Biomembr. 2011, 1808, 23372342,  DOI: 10.1016/j.bbamem.2011.06.012
    12. 12
      Trott, O.; Olson, A. J. AutoDock Vina: Improving the Speed and Accuracy of Docking with a New Scoring Function, Efficient Optimization, and Multithreading. J. Comput. Chem. 2010, NANA,  DOI: 10.1002/jcc.21334
    13. 13
      van Zundert, G. C. P.; Rodrigues, J. P. G. L. M.; Trellet, M.; Schmitz, C.; Kastritis, P. L.; Karaca, E.; Melquiond, A. S. J.; van Dijk, M.; de Vries, S. J.; Bonvin, A. M. J. J. The HADDOCK2.2 Web Server: User-Friendly Integrative Modeling of Biomolecular Complexes. J. Mol. Biol. 2016, 428, 720725,  DOI: 10.1016/j.jmb.2015.09.014
    14. 14
      Honorato, R. V.; Koukos, P. I.; Jiménez-García, B.; Tsaregorodtsev, A.; Verlato, M.; Giachetti, A.; Rosato, A.; Bonvin, A. M. J. J. Structural Biology in the Clouds: The WeNMR-EOSC Ecosystem. Front. Mol. Biosci. 2021, 8, 729513,  DOI: 10.3389/fmolb.2021.729513
    15. 15
      Nyström, S.; Hammarström, P. Amyloidogenesis of SARS-CoV-2 Spike Protein; preprint. J. Am. Chem. Soc. 2022, 144, 89458950,  DOI: 10.1021/jacs.2c03925
    16. 16
      Xue, L. C.; Rodrigues, J. P.; Kastritis, P. L.; Bonvin, A. M.; Vangone, A. PRODIGY: A Web Server for Predicting the Binding Affinity of Protein–Protein Complexes. Bioinformatics 2016, 3676,  DOI: 10.1093/bioinformatics/btw514
    17. 17
      Westermark, P.; Engström, U.; Johnson, K. H.; Westermark, G. T.; Betsholtz, C. Islet Amyloid Polypeptide: Pinpointing Amino Acid Residues Linked to Amyloid Fibril Formation. Proc. Natl. Acad. Sci. 1990, 87, 50365040,  DOI: 10.1073/pnas.87.13.5036
    18. 18
      Jaikaran, E. T. A. S.; Higham, C. E.; Serpell, L. C.; Zurdo, J.; Gross, M.; Clark, A.; Fraser, P. E. Identification of a Novel Human Islet Amyloid Polypeptide β-Sheet Domain and Factors Influencing Fibrillogenesis. J. Mol. Biol. 2001, 308, 515525,  DOI: 10.1006/jmbi.2001.4593
    19. 19
      Thu, T. T. M.; Li, M. S. Protein Aggregation Rate Depends on Mechanical Stability of Fibrillar Structure. J. Chem. Phys. 2022, 157, 055101  DOI: 10.1063/5.0088689
    20. 20
      Wiltzius, J. J. W.; Sievers, S. A.; Sawaya, M. R.; Cascio, D.; Popov, D.; Riekel, C.; Eisenberg, D. Atomic Structure of the Cross-β Spine of Islet Amyloid Polypeptide (Amylin). Protein Sci. Publ. Protein Soc. 2008, 17, 14671474,  DOI: 10.1110/ps.036509.108
    21. 21
      Pandey, P.; Nguyen, N.; Hansmann, U. H. E. d -Retro Inverso Amylin and the Stability of Amylin Fibrils. J. Chem. Theory Comput. 2020, 16, 53585368,  DOI: 10.1021/acs.jctc.0c00523
    22. 22
      Gallardo, R.; Iadanza, M. G.; Xu, Y.; Heath, G. R.; Foster, R.; Radford, S. E.; Ranson, N. A. Fibril Structures of Diabetes-Related Amylin Variants Reveal a Basis for Surface-Templated Assembly. Nat. Struct. Mol. Biol. 2020, 27, 10481056,  DOI: 10.1038/s41594-020-0496-3
    23. 23
      Kryshtafovych, A.; Moult, J.; Billings, W. M.; Della Corte, D.; Fidelis, K.; Kwon, S.; Olechnovič, K.; Seok, C.; Venclovas, Č.; Won, J.; CASP-COVID participants Modeling SARS-CoV-2 Proteins in the CASP-commons Experiment. Proteins Struct. Funct. Bioinforma. 2021, 89, 19871996,  DOI: 10.1002/prot.26231
    24. 24
      Mesias, V. S. D.; Zhu, H.; Tang, X.; Dai, X.; Liu, W.; Guo, Y.; Huang, J. Moderate Binding between Two SARS-CoV-2 Protein Segments and α-Synuclein Alters Its Toxic Oligomerization Propensity Differently. J. Phys. Chem. Lett. 2022, 13, 1064210648,  DOI: 10.1021/acs.jpclett.2c02278
    25. 25
      Abraham, M. J.; Murtola, T.; Schulz, R.; Páll, S.; Smith, J. C.; Hess, B.; Lindahl, E. GROMACS: High Performance Molecular Simulations through Multi-Level Parallelism from Laptops to Supercomputers. SoftwareX 2015, 1-2, 1925,  DOI: 10.1016/j.softx.2015.06.001
    26. 26
      Huang, J.; Rauscher, S.; Nawrocki, G.; Ran, T.; Feig, M.; de Groot, B. L.; Grubmüller, H.; MacKerell, A. D. CHARMM36m: An Improved Force Field for Folded and Intrinsically Disordered Proteins. Nat. Methods 2017, 14, 7173,  DOI: 10.1038/nmeth.4067
    27. 27
      Jorgensen, W. L.; Chandrasekhar, J.; Madura, J. D.; Impey, R. W.; Klein, M. L. Comparison of Simple Potential Functions for Simulating Liquid Water. J. Chem. Phys. 1983, 79, 926935,  DOI: 10.1063/1.445869
    28. 28
      Wang, W.; Hansmann, U. H. E. Stability of Human Serum Amyloid A Fibrils. J. Phys. Chem. B 2020, 124, 1070810717,  DOI: 10.1021/acs.jpcb.0c08280
    29. 29
      Siwy, C. M.; Lockhart, C.; Klimov, D. K. Is the Conformational Ensemble of Alzheimer’s Aβ10-40 Peptide Force Field Dependent?. PLoS Comput. Biol. 2017, 13, e1005314  DOI: 10.1371/journal.pcbi.1005314
    30. 30
      Samantray, S.; Yin, F.; Kav, B.; Strodel, B. Different Force Fields Give Rise to Different Amyloid Aggregation Pathways in Molecular Dynamics Simulations. J. Chem. Inf. Model. 2020, 60, 64626475,  DOI: 10.1021/acs.jcim.0c01063
    31. 31
      Man, V. H.; He, X.; Derreumaux, P.; Ji, B.; Xie, X.-Q.; Nguyen, P. H.; Wang, J. Effects of All-Atom Molecular Mechanics Force Fields on Amyloid Peptide Assembly: The Case of Aβ 16–22 Dimer. J. Chem. Theory Comput. 2019, 15, 14401452,  DOI: 10.1021/acs.jctc.8b01107
    32. 32
      Bussi, G.; Donadio, D.; Parrinello, M. Canonical Sampling through Velocity Rescaling. J. Chem. Phys. 2007, 126, 014101  DOI: 10.1063/1.2408420
    33. 33
      Parrinello, M.; Rahman, A. Polymorphic Transitions in Single Crystals: A New Molecular Dynamics Method. J. Appl. Phys. 1981, 52, 71827190,  DOI: 10.1063/1.328693
    34. 34
      Miyamoto, S.; Kollman, P. A. Settle: An Analytical Version of the SHAKE and RATTLE Algorithm for Rigid Water Models. J. Comput. Chem. 1992, 13, 952962,  DOI: 10.1002/jcc.540130805
    35. 35
      Hess, B.; Bekker, H.; Berendsen, H. J. C.; Fraaije, J. G. E. M. LINCS: A Linear Constraint Solver for Molecular Simulations. J. Comput. Chem. 1997, 18, 14631472,  DOI: 10.1002/(SICI)1096-987X(199709)18:12<1463::AID-JCC4>3.0.CO;2-H
    36. 36
      Humphrey, W.; Dalke, A.; Schulten, K. VMD: Visual Molecular Dynamics. J. Mol. Graph. 1996, 14, 3338,  DOI: 10.1016/0263-7855(96)00018-5
    37. 37
      McGibbon, R. T.; Beauchamp, K. A.; Harrigan, M. P.; Klein, C.; Swails, J. M.; Hernández, C. X.; Schwantes, C. R.; Wang, L.-P.; Lane, T. J.; Pande, V. S. MDTraj: A Modern Open Library for the Analysis of Molecular Dynamics Trajectories. Biophys. J. 2015, 109, 15281532,  DOI: 10.1016/j.bpj.2015.08.015
    38. 38
      Pettersen, E. F.; Goddard, T. D.; Huang, C. C.; Couch, G. S.; Greenblatt, D. M.; Meng, E. C.; Ferrin, T. E. UCSF Chimera?A Visualization System for Exploratory Research and Analysis. J. Comput. Chem. 2004, 25, 16051612,  DOI: 10.1002/jcc.20084
    39. 39
      Frishman, D.; Argos, P. Knowledge-Based Protein Secondary Structure Assignment. Proteins: Struct., Funct., Genet. 1995, 23, 566579,  DOI: 10.1002/prot.340230412
    40. 40
      Kumari, R.; Kumar, R.; Open Source Drug Discovery Consortium; Lynn, A. G_mmpbsa ─A GROMACS Tool for High-Throughput MM-PBSA Calculations. J. Chem. Inf. Model. 2014, 54, 19511962,  DOI: 10.1021/ci500020m
    41. 41
      Bellaiche, M. M. J.; Best, R. B. Molecular Determinants of Aβ 42 Adsorption to Amyloid Fibril Surfaces. J. Phys. Chem. Lett. 2018, 9, 64376443,  DOI: 10.1021/acs.jpclett.8b02375
    42. 42
      Leguizamon Herrera, V. L.; Buell, A. K.; Willbold, D.; Barz, B. Interaction of Therapeutic d -Peptides with Aβ42 Monomers, Thermodynamics, and Binding Analysis. ACS Chem. Neurosci. 2022, 13, 16381650,  DOI: 10.1021/acschemneuro.2c00102
  • Supporting Information

    Supporting Information

    ARTICLE SECTIONS
    Jump To

    The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.3c00621.

    • Residue-wise binding frequencies of SK9 and FI10 to amylin monomers, start configurations with SK9 and FI10 docked, average center of the mass distance of the amylin monomer and SK9 and FI10, residue-wise binding frequencies of the amylin monomer to SK9 and FI10, contact maps of amylin fibril models in the absence and presence of SK9 and FI10, RMSD and RMSF of fibril model 6ZRF in the presence of FI10 capped/uncapped, and residue-wise contact frequencies of fibril model 6ZRF with FI10 capped/uncapped (PDF)

    • Initial and final structures of all trajectories (ZIP)


    Terms & Conditions

    Most electronic Supporting Information files are available without a subscription to ACS Web Editions. Such files may be downloaded by article for research use (if there is a public use license linked to the relevant article, that license may permit other uses). Permission may be obtained from ACS for other uses through requests via the RightsLink permission system: http://pubs.acs.org/page/copyright/permissions.html.