Characterization

Hydrogel formation

First, we wanted to find out if we could use our part to form a hydrogel. All gels were formed by resuspending the ELPs in MilliQ (MQ) water at different w/v%. The leucine zipper-containing ELPs were expected to instantaneously form a gel upon increasing the temperature above T_t.

We saw that for the constructs containing the leucine zippers, a viscous fluid formed upon dissolving the ELPs in MQ water at 4°C. Subsequently, after warming the solution to room temperature, it became very turbid and a gel started to form. Images of the gels can be seen in figure 5.

The construct containing Z1-A120-Z2 formed a gel at 5% and 10% (w/v%) as described above. Upon cooling the gel down to 4°C again, the gel remained intact. As a control, Z2-A120-Z2<( BBa_K4905008) was also dissolved to form solutions of 5% and 10% (w/v%). This construct too, was able to form a gel instantly upon heating to RT. However, in contrast to the leucine zipper-ELPs, the gel disassembled again when the solution was cooled to 4°C. This indicates that the latter of the two constructs shows reversibility, which is expected when the ELPs are brought below their T_t, while the former construct does not show this reversibility. This indicates that at high concentrations, the interactions are not completely reversible.

Transition temperature determination with UV-Vis

The transition temperatures (T_ts) of our constructs Z1-A_[120]-Z2 and Z2-A_[120]-Z2 were determined with the use of UV-visible spectroscopy. Solutions of 5 and 20 μM were made in MQ water. Absorbances were measured at 350 nm. T_ts were determined to be 23.5°C and 21°C for both constructs, respectively (Figure 6). This is remarkable, since the expectation was that the construct containing Z1 and Z2, two leucine zippers that are be able to dimerize, would have a lower T_t, with the reasoning that the interaction of these domains would cause the ELPs to become insoluble at a lower temperature. We suspect these unexpected results might be due to the fact that because of the zippers, there might be less possibilities for the ELPs to form crosslinks. This could cause the T_t to go up instead of down like we expected.

When Z2-A_[120]-Z2 crosses the T_t at 21°C, aggregates form leading to a sharp increase in absorbance. The absorbance then drops, which might be explained by the non-complementary two zipper domains repelling each other, leading to disassembly of the aggregates. Upon further increasing the temperature, the hydrophobic interactions become stronger, which further increases the absorbance, until finally even the hydrophilic block becomes insoluble above 50°C.

Z1-A_[120]-Z2 shows a different temperature dependent behavior compared to Z2-A_[120]-Z2, in that it shows a smaller increase in absorbance when it crosses T_t of 23.5°C. We think the interaction between Z1 and Z2 stabilizes the aggregates, forming into a branched network of ELPs, even at low concentrations of 5 μM.

Absorbances were also measured upon cooling the solution back down. Once the solution goes below T_t, the absorbances of returned back to base levels, indicating that the ELPs can be reversibly precipitated and resuspended again (Figure 7) at these low concentrations.

Dye release from hydrogel

To determine the relative ability for the diffusion of small molecules through our hydrogels, we made gels at two different weight-to-volume percentages containing the fluorescent dye rhodamine B. Finding out whether small molecules can gradually diffuse out of the hydrogel might be interesting for future applications, a drug could by incorporated for instance.

First, a calibration curve was made to determine at which concentration rhodamine B should be added to the gel. Figure 8 shows a linear relationship up to 26 μM of Rhodamine B. We decided to make the gels with a Rhodamine B concentration of 104 μM, since the amount of dye diffusing out of the gels was expected to be far lower than this amount. A 5% gel as well as a 10% gel were tested. Our hypothesis was that diffusion through the 5% gel would be faster than when a higher weight percentage of the ELPs was used,

The gels were made at a volume of 100 μL. They were then brought to RT and washed with 1.5 mL of warm MQ. Then, 1 mL of warm MQ was pipetted on top of the gels, and they were incubated for up to 8 hours at RT, with samples of 100 μL being taken every 2 hours. After measuring the fluorescence intensity, the values were corrected for the reduction in volume upon taking the samples. From figure 9, it becomes clear that the dye diffuses faster from the 5% gel than from the 10% gel. However, more experiments are needed to truly find the rate of diffusion, it is hard to draw a conclusion now with only little data points.

The faster diffusion rate of the 5% gel is in line with our hypothesis that a lower weight percentage gel is able to release small molecules faster than a 10% gel. Since Rhodamine B is a much smaller molecule than a cytokine like IL-10, it would be interesting to see how the diffusion of such a larger molecule would be affected by our gels. The weight percentage of the gel might therefore not only affect the division rate of the bacteria, but it can also play a role in the release rate of the therapeutic.

CCK8

We wanted to find a way that we could measure the amount of cells that we can stop from dividing with our hydrogel and whether they would stay alive. We came across Cell counting kit 8 (CCK8). This is an assay that is commonly used to determine the number of alive cells in a sample. The kit uses a highly water-soluble tetrazolium salt, WST-8, which produces a formazan dye upon reduction in the presence of an electron mediator. The amount of formazan generated by dehydrogenases is directly in proportion to the number of living cells. Yang et al. (2021) have demonstrated that this assay is also suitable for the detection of living E. coli^[4].

We used this assay to test whether our hydrogelated cells would be more resistant to several conditions, including incubation with ampicillin and freeze-drying. Our biggest goal was to find out if cells would stop dividing, but stay functional with our hydrogel. To test this, we came up with the idea of combining the CCK8 assay described above with the antibiotic ampicillin. Ampicillin prevents the synthesis of the bacterium’s cell wall. It does this by binding to enzymes necessary for the formation of the cell wall ^[5]. This means it >only kills bacteria when they divide. Our hypothesis is that if our hydrogel prevents the bacteria from dividing, they should be affected less by the presence of this antibiotic, and therefore show a higher survival rate than bacteria that do not contain the hydrogel since these will still divide and be killed by the antibiotic. A schematic representation of the experiment can be seen below.

For this assay, it was nessecary to measure the OD₄₅₀ with a plate reader. This can directly be related to the OD₆₀₀^[4]. For this relationship, the calibration curve in figure 11 was made.

Ampicillin resistance

Three different conditions were tested over a period of time. Namely; no (protein) expression, ELPs with no cross linkers (A120), and ELPs with Leucine zippers (Z1-A120-Z2). Samples were taken at 12h, 20h, and 36h of protein expression. Before incubation ampicillin, samples were diluted to at a baseline of OD₆₀₀ = 0.5. They were then either incubated with ampicillin (1 mg/mL) or without for 1h at 37°C. The results are summarized in Table 1 and Figure 7.

Ampicillin-treated E. coli that were not induced to express proteins, and which were treated with ampicillin, showed at most a WST-8 conversion of 0.71% ± 0.05% compared to cells which were not treated, as measured by absorbance at 450 nm. In contrast, E. coli where expression of A120 was induced, showed a maximum conversion of 6.80% ± 0.17% after 36h of protein expression. This is likely due to the fact that this construct is also able to form a gel, although less strong compared to the Z1-A120-Z2 construct, leading to an attenuation in cell division. Finally, Z1-A120-Z2 was expressed for the three-time points. An increase was observed in the conversion of substrate, most notably after 20h of expression (20.86% ± 4.14%). We had expected to see a higher percentage after 36h of expression, since at this time point, more bacteria should be gelated and thus unable to divide. The standard deviation of the samples is quite high, with this small number of measurements and large deviation, it is hard to tell whether there is really a difference between these two time points. Overall, a larger sample size could help to reduce the variability of this assay in future tests. Other explanations could be that some of the cells were over-gelated, which might have led to a premature death unrelated to the incubation with ampicillin. Additionally, cell death could also have occurred due to nutrients in the culturing medium running out.

For comparison, OD₆₀₀ of all samples was also measured after incubation with and without ampicillin. When the samples were incubated with ampicillin, they all decreased in OD₆₀₀. The samples incubated without ampicillin in which ELPs were expressed show a relatively stable OD₆₀₀ compared to the sample in which no expression was induced. This was the only sample in which a substantial increase in OD₆₀₀ could be observed.

The OD₆₀₀ was also measured without addition of ampicillin over time for each of the three samples (No expression, A120 and Z1-A120-Z2). After 20h of expression, these were diluted to OD = 0.1 and then measured every 30 minutes. From figure 8, it becomes clear that when no expression of ELPs was induced, the bacteria divided more rapidly than when expression was induced. These findings are in line with the results of the ampicillin treatment. It seems that the presence of the ELPs does not altogether inhibit division, but it does slow it down, which seems to be the reason why they are killed slower upon the addition of ampicillin. However, it is hard to draw a conclusion as to whether the attenuation of bacterial division is driven by the formation of a gel inside the cytosol, or simply because of the crowded environment inside the cell, caused by the presence of a high amount of protein. Therefore, additional controls need to be added in future experiments, like samples in which a mono-or diblock ELP is expressed, in order to account for molecular crowding. Furthermore, a sample where a completely different protein is also interesting, to see what the effect of protein expression is on cell division.

These results were compared to the classic colony counting method. From each sample, a diltion of 10⁵ was made and then plated out on an LB-agar plate. The plates were incubated at 37°C overnight. Images of the plates were taken (Figure 14) and the number of colonies was counted using the colony counter tool from ImageJ. The results of which are summarized in Figure 14 As expected, the wild-type E. coli treated with ampicillin did not form any colonies, whereas untreated bacteria formed several hundreds of colonies. Samples containing A120 and Z1-A120-Z2 both showed a significantly smaller amount of colonies.

Microscopy (FRAP)

Characterization of our gelated cells with the use of microscopy required us to design a model protein which could interact with the hydrogel, and which we could visualize under the microscope. We decided to use VPGIG_[60]-mNeonGreen (part BBa_K4905016), since the hydrophobic ELP part of the protein is able to make hydrophobic interactions with the gel. Because of these interactions, we expected that the proteins would al localize in the same places in the cell. We cloned the gene fragment into a pBad vector under the control of an arabinose promotor, to allow for orthogonality between this vector and the pET24a(+) vector encoding the zipper-ELP construct. Success of cloning was verified by a double digestion of the plasmids as well as sequencing.

However, to our surprise, the sequencing results indicated that instead of a VGPIG_[60] fragment, the mNeonGreen was attached to a VPGAG_[3]VPGGG_[2][12] fragment, which is hydrophilic, relative to the VPGIG_[60] fragment. This is likely due to a labeling mistake by a previous user of the plasmid. Due to a lack of time, we decided to continue using this construct in our microscopy experiments as a model protein to monitor the diffusion of proteins within the gelated bacteria.

Figure 15 shows two samples before and after photobleaching. Following bleach of E.coli which co-expressed Z1-A120-Z2 and and [A3G2]₁₂-mNeonGreen, a gradual recovery of fluorescence could be observed up to 30 seconds post-bleach (Fig. 15A, Fig. 14B). The control, containing only [A3G2]₁₂-mNeonGreen, showed a virtually instantaneous recovery, as evidenced by Fig 15B, Fig 16C.

The raw data was corrected for photobleaching during post-bleach image acquisition and normalized to the intensity pre-bleach. The data was then fitted with equation 1, Where I(t) stands for the normalized fluorescence intensity at time t, A is the mobile fraction, t is time and τ is the time constant^[7]. This was done for each of the measurements, resulting in four graphs for the co-expressed sample and two graphs for the control. Using the equations, the recovery half-life (τ_1/2) was calculated, using t_1/2 = ln(2)*τ for each data set which is summarized in Figure 14. This experiment shows that proteins are distributed throughoit the whole cell and that the diffusion rate becomes lower when our ELP constructs are expressed. It is impossible to exactly determine the intracellular concentration of ELPs, and thus to make a comparison with the gels made extracellularly. However, it would be worthwhile doing these FRAP measurements on more samples in the future where, just as was done for the CCK8 assays, protein expression is induced for a series of different timepoints to see the effect on diffusion.

I(t) = A(1-e^(-t/τ))+c (1)

Measurement	τ1/2 (s)
1	11.9
2	7.1
3	10.9
4	8.4
Average + stdev	9.6 ± 1.9

To further characterize the gelated versus non-gelated bacteria, we used a DNA staining, Hoechst. We saw that a large number of gelated bacteria showed a clustering of DNA in the center of the cell, whereas for the control sample, the DNA was more homogenously spread throughout the cell. This might indicate that the DNA in gelated cells is not as accessible to replication machinery and that therefore division of the cells is stopped or slowed down.

Molecular Dynamics simulation

Introduction

Molecular dynamics (MD) simulations are commonly used to see how proteins or other molecules move over time. This can contribute to the understanding of molecular systems, which cannot be seen with the eye or even with the best microscopes. In our application of forming a hydrogel with Elastin-Like Polypeptides (ELPs), it can be useful to see how the ELPs will behave at different temperatures. Above a certain transition temperature (Tt), there is a disruption in the thermal energy of the water molecules around the ELPs, which causes the water molecules to move away from the ELPs. This makes room for interactions between the ELPs. It has been found that there will be conformational changes in another ELP sequence when the temperature rises above Tt, which makes them more ordered and increases the hydrophobicity^[11].

It is interesting to see how this applies to our own ELP. The MD simulation will give a better understanding of the differences in the structure below and above the T_t and it will help with understanding how the ELP actually looks like. We will analyze different properties of the proteins, like the secondary structure formation, compactness, and the surface area that is accessible for the solvent, to see how the conformation differs at specific temperatures below and above the Tt.

Methods

MD simulations are typically computationally expensive. This made it necessary to shorten our ELP sequence to make the computational time manageable. We chose to shorten the hydrophobic parts to I_[10] instead of I_[60] and to shorten the hydrophilic part to A_[50] instead of A_[120] or A_[100], as shown in Figure 19. The leucine zippers stay the same. In this length reduction, the proportions are not the same as in the original ELP. However, we wanted to make sure that the hydrophobic parts were not shorter than the leucine zippers. This maintains a greater distance between the Leucine zippers and the hydrophilic part, and it prevents unrealistic folding. The hydrophilic part is kept long since this part gives the ELP its length and it might be interesting to see how this part behaves during the folding. The choices made for this might affect the outcome, but the combination of regions is still the same and the difference between the temperatures can still be compared.

Figure 19: ELP construct used for the MD simulation. In this construct, the hydrophobic VPGIG parts are repeated 10 times instead of 60 times and the hydrophilic part in the middle is repeated 10 times instead of 24 or 20 times. At the ends, the two different leucine zippers are located, which remain the same.

A generalized born implicit solvent is used to fold the protein. In an implicit solvent, the solvent (in this case water) molecules are not individually modeled, but with an approximation of the mean force from the solvent on the protein. Different theories can be used for this, and we chose to use the generalized born model. Implicit solvents are faster than an explicit solvent with simulated water molecules and more detailed than a simulation in vacuum[12]. The MD simulation is performed at temperatures 3 °C, 15 °C, 25 °C, and 37 °C, since the transition temperature was found to be 20 °C in the conducted UV-VIS experiments. The simulations are run in triplet to reduce the chance of coincidence in the results. This is necessary, since the initial velocities are chosen by the NAMD software and thus differ every time. Different velocities can cause differences in the final stable conformations, which can affect the results of the analytical methods.

Just like in López Barreiro et al., first energy minimalization is done for 20 000 time steps. Then, Langevin dynamics is used with Generalized Born implicit solvent. The simulation step was 2 fs with a simulation time of 50 ns. The cutoff distance is 18 Å, the switch distance 16 Å^[13].

The results are analyzed with Root Mean Square Distance (RMSD), Radius of Gyration and relative shape anisotropy, Ratio of Secondary Structures, and Solvent Accessible Surface Area (SASA) to show the different behaviors of the ELP at different temperatures. They are calculated with the MDTraj Python library. For the analysis, the analysis methods are calculated for the 1000 frames of the last 5 ns, which has a stable conformation at all temperatures. Even when the conformation is stable, there are still some small differences. These differences are taken into account with this distribution of the results of the last 5 ns.

Results

In Figure 20, a video is shown from one of the simulations at 25°C. Not all videos are included, since the process is comparable for the other simulations, especially when comparing them by eye. Some frames of those simulations can be found in the file below. When comparing the simulations at the four temperatures, they all start with folding at the ends of the ELP. This was expected since those parts are the most hydrophobic. Some of the simulations also show the folding of the ELP in the middle at the more hydrophilic part. However, there is no clear relationship between temperature and this folding behavior, so it is probably caused by luck in the simulation.

Figure 20: Video with one of the MD simulations at 25°C. The red residues represent the Leucine zipper, the blue residues are the hydrophobic part, and the light-blue residues are the hydrophilic part.

RMSD plots are made for each of the three simulations at the four temperatures. They are made for the complete 50 ns, but also for the last 5 ns to show how stable the conformation is related to frame 9000 (45 ns). This is important since there can still be a lot of changes in the structure while the moved distance of the atoms stays the same. Since there are a lot of figures made, they are not visualized here. However, they can be found on the wiki of the TU Eindhoven 2023 team.

For all of the measurements, the equilibrium seems to be reached. When looking at the plots of the last 5 ns, most RMSD plots have a variability of a maximum of 1 nm. Only the third simulation at 37 °C has a slightly larger difference. An RMSD of 0 describes a stable conformation, but there are always some small deviations. So it can be said that an RMSD of 0.25 nm has a high stability, but also 0.2-0.5 nm will give a high stability^[14]. With this information and when looking at the course of the plots, it can still be concluded that the conformations are relatively stable since the RMSD is still very small.

In Figures 21a, b, and c, the boxplots for the secondary structure analysis of the individual simulations are shown. This shows some variability between the simulations of a single temperature, but also overlapping deviations. This happens especially for the α-helix and β-sheet plots. The simulation results are taken together for each temperature to make them easier to analyze, which is shown in Figure 21d, e, and f.

Figure 21: Secondary structure analysis with a, b, and c for respectively the coils, α-helices, and β-sheets three individual simulations for each of the four temperatures and d, e, and f the overall results with the different simulations per temperature taken together.

Overall, almost no secondary structures seem to be formed. The rate of the coils is much higher than the rates for alpha-helices and beta-sheets. It is possible that more secondary structures make the ELP more stable, but that it is stuck in a local minimum. This can be proved with more simulations and some other simulation techniques.

When comparing the boxplots of the different temperatures, there does not seem to be a convincing difference in the secondary structure formation. There are some small differences in the medians and distributions, but there is no clear relation between those differences. We can thus conclude that there is no difference in the secondary structure between the temperatures.

The radius of gyration results of the individual simulations are shown in Figure 22a. The individual simulation results of 3 and 37°C show large differences in the results. This can be caused by the different velocities that are chosen for the single simulations. Since the individual simulations of 3 and 37°C differ a lot, the total boxplot in Figure 22b shows a large distribution of the results.

Figure 22: Radius of gyration results with a three individual simulations for each of the four temperatures and b the overall results with the different simulations per temperature taken together.

There is a clear difference between the radius of gyration at 15 °C and 25 °C, which are just below and just above the transition temperature. When looking at the median at 37 °C, also these results show a lower radius of gyration than 15 °C. At 3 °C, the radius of gyration is for two of the simulations very low compared to the other temperatures. When looking at the final conformations of the simulations, this can be caused by the shape of the stable configurations. When the shape is more cylindrical instead of spherical, which seems especially the case for the simulations at 15°C, the radius of gyration will increase. As mentioned in the methods section, we also computed the relative shape anisotropy.

Figure 23: Relative shape anisotropy results with a three individual simulations for each of the four temperatures and b the overall results with the different simulations per temperature taken together.

In Figure 23a, the relative shape anisotropy is shown for the individual simulations. There is clearly a difference in the shape between the different simulations, also in those with the same temperature. For example 25°C, which has a clear variability between the simulations. Some of the differences were already expected after the analysis of the radius of gyration. The second simulation at 3°C is clearly less spherical than the other two simulations at that temperature. Overall, when looking at Figure 23b, the shape of the ELP at 15°C is less spherical than at 25°C. Because of the large variabilities between the measurements at 3 and 37°C, it cannot be concluded how this is affected at lower and higher temperatures.

This analysis suggests that there is a difference in the compactness of the ELP below and above the transition temperature. Below the transition temperature, the ELP is less compact. At a temperature above the transition temperature, the ELP will become more hydrophobic with fewer solvent molecules around it. This makes the folded ELP more compact, but also more likely to induce interactions with other ELPs. However, for harder conclusions with the role of solvent molecules included, explicit simulations are needed.

In Figure 24a, the individual results show some deviations between the simulations at specific temperatures, especially at 3 and 37 °C. When taking the individual simulations together for each temperature in Figure 24b, some small deviations are shown. SASA at 25 °C seems to be lower than SASA at 15 °C, while the median SASA at 3 °C is also lower than the rest of the temperatures, but with a large deviation. The results for 37 °C also have a large deviation, with a median higher than the median of 15 °C. When comparing the results with the relative shape anisotropy, there is a relationship between more spherical simulation results and a lower SASA. There seems to be a smaller difference than the difference found in the radius of gyration and the anisotropy, which might be because ends that stick out probably have more effect on SASA than the relative shape anisotropy.

Figure 24: SASA results with a three individual simulations for each of the four temperatures and b the overall results with the different simulations per temperature taken together.

The conclusion will be that there is a small but significant difference in SASA at 15 and 25 °C, but that SASA at 37°C becomes higher again. So there is a small effect of the transition temperature on SASA, but not as clear as the radius of gyration. It was expected that SASA would become much lower when passing the transition temperature^[11], but in our ELPs it seems more subtle. This might thus be caused by the ends of the ELP that stick out of the spherical shape of the folded ELP.

Conclusion

In the analysis of the MD simulation of a shortened version of our ELP, different analysis methods were applied. These bring us to the conclusion that the shape of the folded ELP will be altered when increasing the temperature. At a higher temperature, the results suggest a more compact ELP with a more spherical shape. When the temperature increases even more to 37°C, the difference compared to 15°C will decrease. A temperature of 3°C generally results in more compact conformations but with a large variability between the simulations. The explanation for the more compact ELP at higher temperatures is that the residues become more hydrophobic and thus have more interactions with each other. When multiple ELPs are brought together, it is expected that this property results in increased interactions between the ELPs, especially between the hydrophobic ends and between the Leucine zippers.

The radius of gyration and relative shape anisotropy shows the most convincing difference between temperatures below and above the transition temperature, while the secondary structure analysis did not find a clear difference. For SASA, there seems to be a difference, but it is not as clear as the radius of gyration and relative shape anisotropy. For more convincing results, it would probably be better to use explicit solvent simulations and to perform more simulations at more temperatures. However, this and other more extensive experiments on the ELP interactions will be done in further research.

References

[1] Gradišar, H., & Jerala, R. (2010). De novodesign of orthogonal peptide pairs forming parallel coiled-coil heterodimers. Journal of Peptide Science, 17(2), 100–106. https://doi.org/10.1002/psc.1331

[2] Fernández‐Colino, A., Arias, F. J., Alonso, M., & Rodríguez-Cabello, J. C. (2015). Amphiphilic Elastin-Like Block Co-Recombinamers Containing Leucine Zippers: Cooperative Interplay between Both Domains Results in Injectable and Stable Hydrogels. Biomacromolecules, 16(10), 3389–3398. https://doi.org/10.1021/acs.biomac.5b01103

[3] L. E. Contreras-Llano et al., “Engineering Cyborg Bacteria Through Intracellular Hydrogelation,” Adv. Sci., Mar. 2023, doi: 10.1002/ADVS.202204175.

[4] Alber, T. (1992). Structure of the leucine zipper. Current Opinion in Genetics and Development, 2, 205–210

[5] Christensen, T., Hassouneh, W., Trabbic-Carlson, K., & Chilkoti, A. (2023). Predicting Transition Temperatures of Elastin-Like Polypeptide Fusion Proteins. https://doi.org/10.1021/bm400167h

[6] Yang, X., Zhong, Y., Dong, W., & Lu, Z. (2021). A simple colorimetric method for viable bacteria detection based on cell counting Kit-8. Analytical Methods, 13(43), 5211–5215. https://doi.org/10.1039/d1ay01624e

[7] Bereda, G. (2022) Clinical Pharmacology of Ampicillin. Journal of Pharmaceutical Research & Reports Volume 3(3): 3-3

[8] Israeli, E., Shaffer, B. T., & Lighthart, B. (1993). Protection of Freeze-Dried Escherichia coli by trehalose upon exposure to environmental conditions. Cryobiology, 30(5), 519–523. https://doi.org/10.1006/cryo.1993.1052

[9] Eils, R., & Kappel, C. (2004). Fluorescence Recovery after Photobleaching with the Leica TCS SP2. Biology. https://www.leica-microsystems.com/science-lab/fluorescence-recovery-after-photobleaching-with-the-leica-tcs-sp2/

[10] Pille, J., Van Lith, S. A. M., Van Hest, J. C. M., & Leenders, W. P. J. (2017). Self-Assembling VHH-Elastin-Like peptides for photodynamic nanomedicine. Biomacromolecules, 18(4), 1302–1310. https://doi.org/10.1021/acs.biomac.7b00064

[11] N. K. Li, F. G. Quiroz, C. K. Hall, A. Chilkoti, and Y. G. Yingling, “Molecular description of the lcst behavior of an elastin-like polypeptide,” Biomacromolecules, vol. 15, no. 10, pp. 3522–3530, Oct. 2014, doi: 10.1021/BM500658W/SUPPL_FILE/BM500658W_SI_001.PDF.

[12] J. Kleinjung and F. Fraternali, “Design and application of implicit solvent models in biomolecular simulations,” Curr Opin Struct Biol, vol. 25, no. 100, p. 126, 2014, doi: 10.1016/J.SBI.2014.04.003.

[13] D. López Barreiro, A. Folch-Fortuny, I. Muntz, J. C. Thies, C. M. J. Sagt, and G. H. Koenderink, “Sequence Control of the Self-Assembly of Elastin-Like Polypeptides into Hydrogels with Bespoke Viscoelastic and Structural Properties,” Biomacromolecules, vol. 24, no. 1, pp. 489–501, Jan. 2023, doi: 10.1021/ACS.BIOMAC.2C01405/ASSET/IMAGES/LARGE/BM2C01405_0003.JPEG.

[14] M. Arnittali, A. N. Rissanou, and V. Harmandaris, “ScienceDirect Structure Of Biomolecules Through Molecular Dynamics Simulations,” 2019, doi: 10.1016/j.procs.2019.08.181.