Tumorigenic p53 mutants undergo common structural disruptions including conversion to α‐sheet structure

Abstract

The p53 protein is a commonly studied cancer target because of its role in tumor suppression. Unfortunately, it is susceptible to mutation‐associated loss of function; approximately 50% of cancers are associated with mutations to p53, the majority of which are located in the central DNA‐binding domain. Here, we report molecular dynamics simulations of wild‐type (WT) p53 and 20 different mutants, including a stabilized pseudo‐WT mutant. Our findings indicate that p53 mutants tend to exacerbate latent structural‐disruption tendencies, or vulnerabilities, already present in the WT protein, suggesting that it may be possible to develop cancer therapies by targeting a relatively small set of structural‐disruption motifs rather than a multitude of effects specific to each mutant. In addition, α‐sheet secondary structure formed in almost all of the proteins. α‐Sheet has been hypothesized and recently demonstrated to play a role in amyloidogenesis, and its presence in the reported p53 simulations coincides with the recent re‐consideration of cancer as an amyloid disease.

Abbreviations

APBS

Adaptive Poisson–Boltzmann Solver

DSSP

define secondary structure of proteins

FTIR

Fourier transform infrared

MD

molecular dynamics

NMR

nuclear magnetic resonance

NOE

nuclear overhauser effect

PDB

Protein Data Bank

pWT

pseudo‐wild type

RMSD

root‐mean‐squared deviation

RMSF

root‐mean‐squared fluctuation

SASA

solvent accessible surface area

WT

wild type

1. INTRODUCTION

The p53 transcription factor ¹ is one of the most‐studied cancer targets because of its prominent role in tumor suppression. Roughly half of all cancers are associated with mutations to p53 ² with most of those located in the central DNA‐binding domain. Many of these mutations have been studied and characterized, and they commonly display reduced DNA‐binding ability, reduced structural integrity, or both. ³ Extensive experimental characterizations have been performed on p53, leading to a deeper understanding of the mechanisms involved, in some cases proceeding to compounds that are capable of stabilizing the mutants and restoring function, ⁴ , ⁵ even to the point of progressing to clinical trials of potential cancer therapeutics. ⁶ , ⁷ Figure 1 shows the structure of p53, a schematic overview of its secondary structure, and the locations of the 19 destabilizing mutations whose simulations are presented here. The p53 protein belongs to the immunoglobulin‐like β‐sandwich fold represented in Rank 1 of our Consensus Domain Dictionary. ⁸ , ⁹ It consists of two helices H1 and H2, a smaller β‐sheet comprised of Strands S1, S3, S8, and S5 and a larger β‐sheet comprised of Strands S6, S7, S4, S9, S10, S2′, and S2.

FIGURE 1.

(a) p53 crystal structure (Protein Data Bank [PDB]: 2OCJ) labeled with secondary‐structure. Zinc is depicted with a blue ball. (b) Schematic overview of the p53 secondary structures in (a). (c) p53 crystal structure overlaid with the mutations presented here

In the work presented here, we have performed molecular dynamics (MD) simulations ¹⁰ , ¹¹ , ¹² of the DNA‐binding domain of wild‐type (WT) p53, 19 different destabilizing p53 mutants, and a pseudo‐WT (pWT) mutant that is stabilized by 2.65 kcal/mol via four mutations. ¹³ MD is a powerful complement to experimental work because it provides protein structural data with sub‐picosecond, sub‐angstrom resolution. Because zinc is significantly disassociated from p53 at physiological temperatures ¹⁴ and because mutations are associated with zinc loss, ¹⁵ WT and mutant simulations were performed of the apo proteins at 310 K. We also performed additional 298 K holo simulations of WT p53 for comparison to experimental NMR data.

Our purpose in analyzing simulations of p53 was to understand the structural ramifications of known oncogenic mutations. There have been many simulations of the DNA‐binding domain of WT and mutant p53, ¹⁶ , ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ , ²¹ , ²² , ²³ , ²⁴ , ²⁵ , ²⁶ , ²⁷ , ²⁸ but here we present replicate simulations of 20 different mutants to better understand the both the common and specific responses to mutation in this protein. These simulations involve all six “hot spot” residues ²⁹ , ³⁰ and represent eight of the ten most common mutations and 21% of the listings in the IARC TP53 Database ³¹ (version R16). The most common cancer topologies associated with these mutations are the uterus, rectum, colon, liver, brain, stomach, hematopoietic and reticuloendothelial systems, pancreas, bone, ovary, lymph nodes, esophagus, breast, mouth, larynx, prostate, soft tissue, bladder, head and neck, lung, and skin. ³¹ The mutations discussed here are distributed throughout the protein (Figure 1c) and in most cases, experimental data are available. ³ These simulations were performed as part of our Dynameomics project ³² , ³³ (www.dynameomics.org). The Dynameomics project uses MD to characterize the folding and unfolding pathways as well as native‐state dynamics of all known protein folds.

Here, we report our findings, including descriptions of the structural‐disruption motifs that were common to multiple mutants. In addition to large‐scale changes in tertiary structure, one of the most interesting structural‐disruption motifs that we identified was the presence of α‐sheet secondary structure in almost all of the proteins. α‐Sheet structure was first discovered in MD simulations of transthyretin (TTR) at low pH, and it has the same β‐sandwich structure. ³⁴ Later, it was observed in simulations of several other unrelated amyloid proteins under amyloidogenic conditions. ³⁵ , ³⁶ More recently, experimental studies demonstrated that a number of different proteins and peptides form α‐sheet and their aggregation is inhibited by anti‐α‐sheet inhibitors. ³⁷ , ³⁸ , ³⁹ , ⁴⁰ The details of the conformational consequences of introducing disease‐associated mutations into p53, together with a recent recharacterization of cancer as an amyloid disease, ⁴¹ , ⁴² suggest that α‐sheet may play a role in the conformational changes induced by the mutations.

2. RESULTS

2.1. Comparison of 298 K holo simulations with experiment

As there is a paucity of experimental structural data for apo p53 at 310 K, we validated our protein system by comparing 298 K holo simulations with 298 K holo NMR data (Protein Data Bank ⁴³ accession code [PDB:] 2FEJ ⁴⁴ ). In all three 298 K holo simulations, L1 and the smaller S1/S3/S8/S5 β‐sheet experienced dynamic disruptions. In all cases, however, the protein retained its general β‐sandwich structure and no significant β‐core disruptions or unfolding occurred. With the exception of the loss of S6 β‐structure in one simulation, the larger S6/S7/S4/S9/S10/S2′/S2 β‐sheet was consistently stable. One simulation demonstrated large fluctuations between L1 and H2. The H1 and H2 helices were maintained in all simulations. These results are consistent with those of Fersht and coworkers ⁴⁴ who described the p53 core domain as inherently unstable with highly dynamic tertiary interactions in the hydrophobic core. In addition, they noted conformational changes and alternative conformations of L1, as also observed by MD.

The 298 K holo WT simulations satisfied 89% of the experimental NOEs ⁴⁴ with an average violation distance of 2.0 ± 1.9 Å. Then, 88% of the Nuclear Overhauser Effect (NOE) violations involved side‐chain atoms and most violations involved at least one loop residue. Also 33% of the violations were associated with nine residues, eight of which were in the S1 and L1 regions. These residues had an average violation distance of 3.3 ± 2.1 Å. The remaining residues had an average violation distance of 1.4 ± 1.3 Å. Almost half (45%) of the residue pairs containing a violation also contained a satisfied NOE. The average solvent‐accessible surface area (SASA) of the 298 K holo simulation structures and the 298 K holo NMR structures was correlated, with R = .79. Cα root‐mean‐squared fluctuation (RMSF) behavior was similar between the NMR structures and the 298 K holo simulations (Figure S1a). Note that the results presented below are for apo forms of p53.

2.2. Analysis of 310 K apo WT simulations

High‐resolution structural information is not available for apo p53 lacking zinc. Consequently, our comparisons with experiment are necessarily more indirect. Butler and Loh showed that apo 53 is folded but it is highly dynamic, conformationally altered, and highly aggregation prone. ¹⁴ Per‐residue Cα RMSF analysis of the WT 310 K apo simulations showed general agreement with the WT 298 K holo NMR data [Figure S1(b)]. There were increases in residue fluctuation in the loop region immediately preceding S1, the S3/S4 loop, the S9/S10 loop, and H2. Not surprisingly, the largest increases in fluctuation occurred near the apo zinc‐binding residues C176, H179, C238, and C242. The WT apo simulations had average Cα root‐mean‐squared deviation (RMSD) values of 4.7 ± 0.2 Å, 4.7 ± 0.2 Å, 4.5 ± 0.3 Å; the values for the β‐core were 2.8 ± 0.3, 1.9 ± 0.2 Å, and 2.6 ± 0.2 Å. The majority of the deviation from the starting structure occurred in L1, L2, the S7/S8 loop, L3, and H2 (Figure S2), consistent with ¹⁵ N/¹H HSQC crosspeak shifts in those regions upon removal of zinc. ¹⁴

Secondary‐structure analysis identified sporadic loss of S1 β content, gain of α‐sheet secondary structure between S1 and S3 and between S6 and S7, loss of H1 helical content, and gain of helical content in the H168 region. One WT apo simulation (#2) demonstrated a 20 Å separation between L1 and H2, disrupting the loop–sheet–helix region while another WT simulation (#3) demonstrated a 15 Å separation between L2 and L3.

The per‐residue Cα RMSF and experimental crystallographic B factors were similar to the NMR Cα RMSF discussed above (Figure S3). WT was compared to the crystal structures 2OCJ ⁴⁵ (original simulation structure) and 2XWR ⁴⁶ (WT structure with a five‐residue extended N‐terminus). Experimental crystal structures were found for 10 mutants (PDB codes: 1UOL, ¹³ 4IBQ, ⁴⁷ 4IJT, ⁴⁷ 4IJS, ⁴⁷ 2J1Y, ² 2BIN, ⁴⁸ 3D05, ⁴⁹ 3D06, ⁴⁹ 3D07, ⁴⁹ 2J21, ² 2J1W, ² 4KVP, ⁵⁰ and 2J1X ² ). In some cases, the only mutant structures available for comparison also contained the four additional stabilizing mutations discussed above. ² , ⁴⁸

The Cα RMSF data were generally in agreement between the simulations and the crystallographic B factors, although the crystallographic amplitudes were typically lower, particularly in the loop regions. The largest differences occurred in both termini and the L1, L2, S7/S8 and, to a lesser degree, L3 loop regions. Higher simulation Cα RMSF values can be rationalized due to the lack of zinc, temperature differences (310 vs. 100 K), and the different environments (solvated vs. crystal). In Figure S3, missing B factor data are circled in green and the individual PDB files are labeled with the number of missing residues. Missing B factor data were concentrated in the L1 and L2 loop regions, although the terminal residues were occasionally missing data as well.

2.3. Analysis of quad‐stabilized pWT mutant

The quad‐stabilized pWT p53 protein (PDB: 1UOL) contains four known stabilizing mutations: M133L, V203A, N239Y, and N268D. The mutant has been shown to maintain WT functionality while being stabilized by 2.65 kcal/mol. ⁵¹ The average Cα RMSD values of our simulations were 5.1 ± 0.2, 4.1 ± 0.2, and 3.8 ± 0.4 Å and β‐core Cα RMSD values were 3.1 ± 0.2, 1.9 ± 0.2, and 1.7 ± 0.2 Å. In every simulation, the S7/S8 loop was the region with the highest Cα RMSD values with average deviations from the starting structure of 10.2 ± 1.0, 9.3 ± 1.2, and 6.1 ± 0.7 Å in the three simulations. Most Cα RMSD values were near or below WT ranges, although one simulation had values for S1, L1, and S3 that exceeded WT values by 2‐5 Å and two simulations had values in L2 that exceeded WT values by 5 Å. Cα RMSF values were also near or below WT ranges except for the L1 region, which exceeded WT values by 2 Å and the H1 region, which was lower than WT by approximately 1 Å. Secondary‐structure analysis indicated a loss of H1 helical‐structure in one simulation, the loss of S1 β‐structure in two simulations, and the gain of α‐sheet between S3 and S8 in one simulation and between S1 and S3 in another simulation. L1 and H2 did not separate in any of the three simulations and both L2/L3 separation and L2/S5 separation distance values were less than WT.

To analyze the effects of the four stabilizing mutations to the p53 protein, we compared the average contact occupancy of WT contacts with the average contact‐occupancy of pWT contacts. From 1,706 residue:residue contacts, we selected those contacts whose WT and pWT occupancy differed by at least 50%. This resulted in 81 contacts, 41 of which were more stable in pWT and 40 of which were more stable in WT. The location of these contacts and the specific residues involved are detailed in Figure 2, Table S1 and Table S2. Contact changes were primarily located in the loop–sheet–helix region, the DNA‐binding region, and the S5/S6 region. Additional contact changes included a few stabilizations across the S1/S3/S8/S5 β‐sheet and a few destabilizations in the N‐terminus of S10 and in the N‐terminal loop.

FIGURE 2.

Crystal structure of stabilized pseudo‐wild type (WT; Protein Data Bank [PDB]: 1UOL) overlaid with the 81 contacts whose occupancy differed by at least 50% from WT. The four quad‐stabilizing mutation sites (M133L, V203A, N239Y, N268D) are shown as spheres

2.4. Overview of analysis of 19 disease‐associated mutants

Individual analyses of the 19 destabilizing mutants are included in the supplementary materials. In general, mutation‐location was not indicative of specific structural behaviors. Over the time frames simulated, tertiary‐disruptions from the starting structure were common, but significant structural loss of the β‐core was not observed. The Cα RMSD of the β‐core was always lower than the full‐protein, and the Cα RMSD of the larger S6/S7/S4/S9/S10/S2′/S2 β‐sheet was almost always lower than that of the smaller S1/S3/S8/S5 β‐sheet. The largest Cα RMSD values were typically seen outside of the β‐core, particularly in Loops L2 and L3 (as expected without the stabilizing zinc‐contact, and in agreement with experiment ¹⁴ ), the loop–sheet–helix region, the loops in the S5/S6/S7 region, and the S7/S8 loop. In general, the largest Cα RMSF values were in the same regions. Most of the time, the Cα RMSD and Cα RMSF values fell within or very close to WT ranges.

These same regions also showed the greatest departure from crystal secondary‐structures. As discussed below, novel helical structures were observed in the N‐terminal loop, L1, L2 and L3 while helical content was often lost in H1. The β‐content of the larger β‐sheet was usually well maintained, but the β‐content of the smaller sheet was often lost. In several instances, the N‐terminal loop was able to form a novel β‐strand aligned antiparallel to S10. The appearance of α‐sheet structure, discussed in more detail below, was also a common occurrence. Common tertiary‐structure changes involved L2 moving away from L3, which disrupted the zinc‐ and DNA‐binding regions, and L1 moving away from H2, which disrupted the loop–sheet–helix region.

2.5. The region around H168 was able to support helical content in apo p53

The region around H168 demonstrated both helical and loop structures; WT and 12 mutants (15 simulations in total) demonstrated helical content around H168 for at least 50% of the time. This is consistent with published data; PDB Define Secondary Structure of Protein (DSSP) analysis of the H168 region indicates a loop for the 2OCJ WT crystal structure and a 3/10 helix for the 2FEJ WT NMR structure. ⁴⁴

2.6. The smaller β sheet was more prone to structural disruption than the larger β sheet

In general, the smaller S1/S3/S8/S5 β sheet lost more secondary structure than the larger S6/S7/S4/S9/S10/S2′/S2 β sheet. This may have been because the shorter strands in the S1/S3/S8/S5 sheet provided fewer stabilizing contacts. It may also have been due to environmental conditions; the average residue in the shorter sheet had 1.4 ± 0.2 times the solvent exposure of a residue in the larger sheet. The solvent‐accessible surface of the WT crystal structure is shown in Figure 3. This figure shows that the shorter β sheet is considerably more solvent exposed than the larger β sheet, with S1, S3, and S8 all contiguously exposed. Much of the larger β sheet is shielded from solvent by the N‐terminal loop, the N‐terminal half of L2, the S6/S7 loop, and the S9/S10 loop. The primary points of solvent exposure in the larger β sheet are S4, S6, and S10. S4 is exposed at the N‐terminal end near the S3/S4 loop. S6 is exposed for much of its length, but it is exposed on its edge, and S6 itself is the edge of the larger β sheet, so in general the strand:strand hydrogen bonds in the larger β sheet are shielded from competitive hydrogen bonding from solvent. S10 is exposed for most of its 10‐residue length, from the S9/S10 loop, past S1, and then into the loop–sheet–helix region. Two of these regions, the S1 and the loop–sheet–helix regions, demonstrated significant disruption from the original starting structure in most of the simulations. Notably, the point where both S10 and S1 are exposed lies at the N268/L111 contact point, the same point where the N268D mutation in the pWT crystal structure establishes a novel hydrogen bond between S10 and S1. ¹³

FIGURE 3.

Rotated views of the β‐sheet solvent exposure of the 2OCJ crystal structure. β‐Sheets are colored and the remainder of the structures are gray. (Top) Solvent‐accessible surface area. (Bottom) Cartoon view

2.7. L1 separation from H2 was a common loop–sheet–helix structural‐disruption motif

The loop–sheet–helix region of p53 is involved in DNA binding, ²⁴ and disruption of this region could compromise binding. K120 and R280 are DNA‐contact residues in L1 and H2, respectively. The experimentally determined distance between the Cα atoms of these two residues in holo p53 ranges between 4.9–7.1 Å (for crystal structures 2OCJ and 1TSR ²⁴ ) and 5.6–9.1 Å (for solution NMR structures 2FEJ). Simulations of the apo forms of WT, pWT, and several mutants established stable conformations that placed these atoms between 5 and 10 Å of each other. However, the apo WT protein and 13 of the 19 mutants all had at least one simulation where these atoms and, by proxy, L1 and H2, were separated by at least 20 Å. The stabilized pWT did not show such a separation; pWT L1 and H2 separation was stable at 5 Å across all three simulations.

In total, there were 17 apo simulations that adopted a conformation that separated L1 and H2 by at least 20 Å. There were 122 contacts involving at least one loop–sheet–helix residue that were common to all of these conformations; to identify which contacts might be contributing to the L1/H2 separation, we compared them against a WT simulation with a stable crystal‐like L1/H2 distance and minimal Cα RMSD in the loop–sheet–helix region (2.4 ± 0.2 Å). Comparing the WT contact occupancy with the average contact occupancy across the 17 simulations, we identified 47 contacts that had at least a 90% occupancy difference with WT. These contacts, common to all L1/H2 separations, are shown in Figure 4 and Table S3. Of these 47 contacts, 46 of them were weakened or lost relative to WT, suggesting that these contacts may play a role in maintaining the structure of the loop–sheet–helix region by holding L1 and H2 in a crystal‐like conformation. Disruption of this region in the apo forms is consistent with both holo and apo NMR studies, ¹⁴ , ⁴⁴ as well as the loss of site‐specific DNA binding activity by WT apo p53. ¹⁴

FIGURE 4.

p53 structures highlighting disrupted loop–sheet–helix region contacts. Cα atoms of K120 and R280 are colored cyan, stabilizing mutation site M133 is colored yellow. (a) p53 crystal structure (2OCJ). K120/R280 distance is 6.7 Å. (b) p53 wild‐type (WT) MD structure at 80 ns. K120/R280 distance is 20 Å. (c) p53 crystal structure overlaid with contacts common to all 17 L1/H2‐separated conformations. Green lines indicate contacts that were present in the separated conformations but not in the crystal‐like WT conformation. Magenta lines indicate contacts that were not present in the separated conformations and were present in the crystal‐like WT conformation. Table S3 specifies which contacts were present in the separated and nonseparated conformations

2.8. L2 separation from S5 was a common zinc‐region structural‐disruption motif

The L2 loop region stretches between residues 164 and 194 and contains the H1 helix between residues 177 and 180. Zinc‐binding residues C176 and H179 are contained within H1. A common structural‐disruption motif in this region was a separation of L2, particularly the region of L2 C‐terminal to H1, from the S5/S6 loop. To quantify the ubiquity of this separation, we measured the distance between the Cα atoms of D186 in L2 and G199 in the S5/S6 loop. Experimental distances for these atoms range between 9.8–10.6 Å (for crystal structures 2OCJ and 1TSR) and 11.0–15.6 Å (for NMR structures 2FEJ); investigation of the simulations showed that 13 of the 19 mutants demonstrated a separation between these atoms of at least 20 Å in at least one simulation. Neither the apo WT nor the stabilized pWT demonstrated this degree of separation.

In total, there were 14 simulations in which L2 and S5 were separated by at least 20 Å. There were 113 contacts involving at least one L2, S5, or S5/S6 residue that were common to all of these conformations. To identify which contacts might be contributing to the L2/S5 separation, we compared them against a WT simulation with a stable crystal‐like L2/S5 distance and minimal Cα RMSD in the L2 region (3.3 ± 0.3 Å). Comparing the WT contact occupancy with the average contact occupancy across the 14 simulations, we identified 49 contacts that had at least a 90% occupancy difference with the WT. These contacts, common to all L2/S5 separations, are shown in Figure 5 and Table S4. Then, 48 of the 49 contacts were weakened or lost relative to WT; only one contact between L3 and S5 was commonly stabilized among the mutants.

FIGURE 5.

p53 structures highlighting disrupted L2 and S5 region contacts. Cα atoms of D186 and G199 are colored cyan, stabilizing mutation site V203 is colored yellow. (a) p53 crystal structure (2OCJ). D186/G199 distance is 10.6 Å. (b) p53 R248Q mutant MD structure at 80 ns. D186/G199 distance is 24.6 Å. (c) p53 crystal structure overlaid with contacts common to all 14 L2/S5‐separated conformations. Green lines indicate contacts that were present in the separated conformations but not in the crystal‐like wild‐type (WT) conformation. Magenta lines indicate contacts that were not present in the separated conformations and were present in the crystal‐like WT conformation. Table S4 specifies which contacts were present in the separated and nonseparated conformations

2.9. L3 helical‐propensity was associated with disruption of the DNA‐binding region

L3 adopted helical conformations in multiple apo simulations, distorting the DNA‐binding region. Residue S241, which is responsible for contacting the DNA backbone, ²⁴ was contained within the novel helical structure; in the R273C mutant (Figure S4), the Cα atom of R241 was displaced from the starting structure by 7 Å and the side chain of that residue was reoriented.

In most instances, a complete helix did not form, but an alternative α‐sheet‐like structure formed instead. α‐Sheet is comprised of residues in alternating α_R/α_L positions on a Ramachandran plot with the backbone carbonyl oxygens aligned uniformly along the strands (see discussion below). In these instances, the residues on either side of the L3 turn aligned in α‐sheet position and the residues in the turn were typically some mix of α_R and α_L, along with other more distributed loop‐like Ramachandran positions. This MD conformation was very similar to the 2FEJ NMR structure (Figure S5). This is notable. In the WT apo protein loose α‐sheet structure was acquired during MD, and it was very similar to that determined for the holo protein by NMR (compare Figure S5b with c.

In total, secondary‐structure analysis indicated that four simulations representing four different mutants adopted L3 helical structure at least 50% of the time; three of these were right‐handed helices, while one (R273H) entered a right‐handed helical turn, reversed direction, and completed a full left‐handed turn. Similar analyses indicated that 34 simulations representing apo WT, pWT, and 18 of the 19 mutants adopted α‐sheet‐like structure in the L3 turn at least 50% of the time, and that structure was also observed in the holo NMR structures (2FEJ). The only mutant not represented was V143A, which fell below the threshold at 21%.

2.10. Most apo proteins adopted some degree of α‐sheet content

The apo forms of WT, pWT, and 18 of the 19 mutants adopted α‐sheet in at least one simulation for at least 10% of the time (the remaining F134L mutant adopted a novel N‐terminal α‐strand for only 1% of the time, discussed below). The α‐sheets were typically short at three residues in length, two strands wide and centered around residues R110 or L111 in S1 and Q144 or L145 in S3 (Figure 6). These two‐strand, three‐residue sheets typically showed two residues in α_R/α_L positioning and one residue in β position on a Ramachandran plot similar to a β‐bulge. Although not as common, some α‐sheets were able to elongate beyond the original β‐strand residues; one G245S α‐sheet extended from residue 109 to residue 113 in S1 and from residue 144 to residue 149 in S3 (Figure 7). α‐Sheets were also able to extend to Strands S8 and S5, although these were less frequent, and α‐sheet residues were able to align while oriented in either direction (left or right with respect to the α‐strand).

FIGURE 6.

Wild‐type (WT) apo simulation displaying α‐sheet conformation. (a) Strands S1 and S3 displaying β‐sheet conformation at 0 ns. (b) Strands S1 and S3 displaying α‐sheet conformation at 75 ns

FIGURE 7.

Proposed G245S α‐sheet formation model. Arrows indicate new motion and fade over time. (a) 0 and (b) 31 ns. N‐terminal Q104:OE1 orients to F109:O, causing it to rotate and face W146:O. Main chain hydrogens of F109 and R110 now face Q104:OE1. (c) 45 ns. W146 rotates away from F109. (d) 60 ns. L111:O rotates inward and S1 and S3 slide past each other, finding a stable conformation one residue offset from previous conformation. (e) 85 ns. Q144 and V143 rotate and move toward F113. Note that Q104:OE1, F109:H and R110:H, once in position, remain stable for the entire process. (f) Electrostatic surface showing solvent‐accessible charge‐separation across the α‐sheet

In addition to the S1/S3/S8/S5 α‐sheet, α‐strand conformation was observed at the N‐terminus, in L1, in L3 (discussed above), in the S7/S8 loop, and between Strands S6 and S7. This latter case occurred less frequently; eight mutants (including M237I, which did not exhibit S1/S3 α‐sheet) demonstrated S6/S7 α‐sheet, but all less than 35% of the time. One WT simulation, however, exhibited S6/S7 α‐sheet for 91% of the time (Figure 8).

FIGURE 8.

Wild‐type S6/S7 α‐sheet. S6/S7 α‐sheet was present for 91% of the time. Structure shown at 80 ns

To establish whether the residues were correlated in forming α‐sheets and not simply unstable residues moving between structured and unstructured conformations, we analyzed the last 25 ns of the simulations. For each α‐sheet residue pair (e.g., L111:Q144), we correlated the per‐picosecond secondary‐structure assignments to α‐conformation. If two α‐strand residues were acting as a single α‐sheet, they should be in α conformation, or not, at every picosecond; if they were simply unstructured, there should be no correlation. We also calculated the percentage of time that the residues spent in α‐conformation and (with the exception of the F134L mutant) only analyzed those α‐sheets that were present for at least 10% of the time. A summary of the major identified α‐sheets is shown in Table 1, aggregated into inter‐secondary‐structure α‐sheet contacts.

TABLE 1.

α‐Sheet summary

	NT/S10	S1/S3	S3/S8	S8/S5	L1 loop	S6/S7	S7/S8 loop	L3 loop
R a	0.91	0.95 ± 0.11	1.00 ± 0.00	1.00	0.97	1.00 ± 0.00	0.99	0.74 ± 0.22
% Time b	1	71 ± 28	73 ± 28	23	25	37 ± 29	23	84 ± 25
Simulations	1	23	6	1	1	5	1	38
Mutants c	1	16	5	1	1	4	1	15
WT	No	Yes	No	No	No	Yes	No	Yes
Quad	No	Yes	Yes	Yes	No	No	No	Yes
Notes d	F134L			G245S	V157F		L145Q

^aThe average and standard deviation values were calculated using the correlation coefficients from the α‐sheets in each group. Each individual correlation coefficient was calculated at picosecond resolution between 75 and 100 ns (n = 25,000).

^bα‐Sheets were required to be present for at least 10% of the time. Because of its relevance to the α‐sheet analysis and because of its high degree of structural correlation, the F134L NT/S10 α‐sheet is included here despite only being present 1% of the time.

^cThe pseudo‐wild type is called out separately and is not included in this group.

^dMutant is called out if the α‐sheet appeared in only one simulation.

With the exception of L3, all α‐sheets had average structural correlation values R ≥ .90. Most α‐sheets were in the S1/S3/S8 region, accompanied by several less‐frequent occurrences and several single‐mutant α‐sheets. All of them, however, had strong structural correlations between the two α‐strands.

The α‐sheet occurring in the turn in the L3 loop had a lower average correlation coefficient (R = .74 ± .22) than the other regions, despite being present for more of the time and for more of the mutants than any of the other regions. The N‐terminal α‐sheet was only present for 1% of the time in a single mutant, but with a structural correlation coefficient of R = .91. This sheet was the only α‐sheet involving a novel piece of secondary structure and the only sheet to convert a residue in the larger β‐sheet. It is also worth noting that the converted S10 residue G266 neighbors the rescue‐mutation residue N268 (Figure 9).

FIGURE 9.

Novel N‐terminal α‐sheet. F134L mutant at 75.4 ns. α‐Sheet formed for 1% of the time. G105 and G266 α conformations were correlated with R = .91

A cascade of rotations and rearrangements was common during α‐sheet formation (Figure 7). This sequence often began with an initiating residue, such as L111, rotating into an α‐sheet position and creating an unstable situation for a residue on an adjacent strand, such as Q144. If the initiating residue was sufficiently stable in this position, the adjacent residue rotated into a complementary α‐position. In a three‐stranded α‐sheet, such as that adopted by the Y220C mutant, this cascade of rotations continued into the third strand. During this cascade, the strands often rearranged to optimize their relative positions. In the G245S mutant (Figure 7), this was achieved by S1 and S3 sliding past each other almost one full residue length.

2.11. The N‐terminal residues were capable of forming new secondary structures

In most of the simulations, the residues N‐terminal to S1 remained unstructured. As discussed above, however, the N‐terminal loop was capable of forming α‐sheet with S10. In addition to this new α‐content, 11 simulations representing eight different mutants adopted new β‐content in the N‐terminal residues for at least 10% of the time and as much as 83% of the time. These nonnative strands were typically centered on Y103, ran antiparallel to S10, and were three to four residues in length. Analysis of ϕ/ψ angles indicated that T102 and Y103 were usually in the β‐quadrant of the Ramachandran plot while Q104 was less structured as the residues curved around to form S1. An example of a new N‐terminal β strand is shown in Figure S6.

In addition to creating a new β‐strand, the N‐terminal loop was also capable of adopting helical character. Two simulations adopted a helical turn from Residues 104–108 for a significant percentage of time (M237I: 48% of the time, R273C: 86% of the time) (Figure S7). Other mutants demonstrated N‐terminal helical‐content less frequently, typically <5% of the time. Overall, while the N‐terminal loop adopted some dynamic nonnative structure, both α‐helix and β‐strand, it remained largely unstructured independent of sequence.

2.12. Aggregate contact analysis of all destabilizing mutant simulations identified three primary regions of disruption

Figure S8a shows a histogram of aggregated contact‐difference data from all 57 (3 × 19) simulations of destabilizing p53 mutants. In general, most contacts retained WT‐like contact occupancies. The two most destabilized contacts (L114:T125 and L114:Y126) overlapped with the pocket discovered by Wassman et al. ¹⁶ that was shown to be capable of stabilizing the R175H tumorigenic mutant. Figure S8b shows the 31 contacts whose occupancy‐difference relative to WT was >0.3 or <−0.3 (the left and right extremes of Figure S8a). Analysis of these data identified three primary disruption‐regions, overlapping with the secondary‐ and tertiary‐structure analyses discussed above: the loop–sheet–helix region, the N‐terminal loop, and the L2/S8/S5/S6/S7 region. Contacts in these regions whose simulations demonstrated at least 50% occupancy loss relative to WT included 100:252 (12 of 19 mutants), 114:125 and 114:126 (16 of 19 mutants), 200:232 (14 of 19 mutants), 171:249 (15 of 19 mutants), and 165:249 (16 of 19 mutants).

3. DISCUSSION

3.1. α‐Sheet may play a role in p53 aggregation

Cancer is in some respects an amyloid disease; p53 mutations have been statistically correlated with p53 accumulation in tumor cells and long‐term patient survival is statistically lower with aggregating mutants. ⁵² Aggregation of p53 is accelerated by increased temperature, ⁵³ increased pressure, ⁵⁴ destabilizing mutations, ⁵² , ⁵³ demetallation, ¹⁴ denaturants such as urea, ⁵³ and inflammatory agents such a formaldehyde ⁵⁵ ; aggregation can also be induced in native WT protein by seeding it with mutant p53 or with WT p53 that has been converted to an amyloidogenic conformation, ⁴¹ which is facilitated upon removal of the zinc. ¹⁴ Accordingly, the dominant‐negative effect long recognized for p53 mutants is hypothesized to be at least partly attributable to aggregation‐prone mutant p53 converting native WT p53 to an aggregating species. ¹⁴ , ⁵² , ⁵⁶

p53 can also induce aggregation in other proteins, namely the p53 analogues p63 and p73, ⁵² suggesting that there is a common aggregation mechanism at work that is not p53‐specific. This is further supported by the report that the A11 antibody binds soluble p53 ⁵⁷ ; the A11 antibody is known to bind multiple amyloidogenic targets, in particular the soluble oligomeric forms, and the inclusion of the p53 protein into that set supports the hypothesis that there is a common aggregation mechanism that generalizes beyond the p53 family. ⁵⁸

Based on MD simulations of amyloidogenic proteins, we have hypothesized that aggregation involves the formation of α‐sheet secondary structure en route to formation of the β‐sheet fibrils and plaques. ³⁴ , ³⁵ , ³⁶ We discovered this nonstandard structure in MD simulations of a variety of amyloidogenic proteins under amyloidogenic conditions but not under nonamyloidogenic conditions. Interestingly, this structure was predicted by Pauling and Corey and termed “polar‐pleated sheet,” ⁵⁹ and in that same publication they, rightly, dismissed the structure in favor of β‐sheet for native proteins.

We have now observed α‐sheet in MD simulations of many amyloid proteins, including TTR (systemic amyloidosis), ³⁴ , ³⁵ , ⁶⁰ , ⁶¹ β2‐microglobulin (a dialysis‐associated amyloidosis), ⁶² prion protein (Creutzfeldt Jacob disease, bovine spongiform encephalopathy, etc.), ³⁵ , ⁶³ , ⁶⁴ lysozyme (familial storage disease), ³⁵ polyglutamine (Huntington's disease), ⁶⁵ β‐amyloid peptide (Alzheimer's disease), ⁶⁶ superoxide dismutase (amyotrophic lateral sclerosis), ⁶⁷ islet amyloid polypeptide (Type 2 diabetes, unpublished), PSMα1 (Staphylococcus aureus), ³⁹ and CsgA (Escherichia coli). (Bleem A, Chen R, Hady T, Li J, Bryers JD, Daggett V. Synthetic a‐sheet peptides disrupt uropathogenic Escherichia coli biofilms by inhibiting curli fibril formation; unpublished). In addition, α‐sheet formation is not unique to our simulation methods or force field. Other investigators have also observed α‐sheet in silico using other simulation packages and other force fields, including TTR (AMBER program, parm94 force field), ⁶⁸ α‐synuclein (GROMACS, CHARMM27), ⁶⁹ , ⁷⁰ β‐amyloid peptide (GROMACS, CHARMM36m), polyglutamine (GROMOS, GROMOS96), ⁷¹ and others. ⁷² , ⁷³ , ⁷⁴ , ⁷⁵ , ⁷⁶ By way of comparison, we point out that we do not observe α‐sheet formation in “normal” proteins. We have a large reference set of 807 different proteins representing 97% of all known autonomous protein folds that we have simulated as a part of our Dynameomics project. ⁹ , ³² , ³³ , ⁷⁷ , ⁷⁸

While α‐sheet has been observed in many computational studies of amyloidogenic proteins, recent experimental studies provide supporting evidence for our prediction of α‐sheet early in aggregation, particularly in the toxic oligomeric structures. ³⁷ , ³⁸ , ³⁹ , ⁴⁰ As an indirect reflection of α‐sheet structure, Hopping et al. designed peptides with α‐sheet structure complementary to the hypothesized oligomer α‐sheet structure, ³⁶ reasoning that the complementary structures should bind and inhibit amyloidosis. Biophysical studies of the designed peptides support the presence of α‐sheet, in addition to an NMR structure of one of the designs, ⁴⁰ and the peptides successfully inhibit aggregation of TTR, β‐amyloid peptide, islet amyloid polypeptide, ³⁷ , ³⁸ , ⁴⁰ and amyloid‐forming bacterial proteins from S. aureus ³⁹ and E. coli, (Bleem et al., unpublished) as well as in the live bacteria ³⁹ , ⁷⁹ (Bleem et al., unpublished) by preferentially binding the toxic species over the nontoxic species of those unrelated peptides and proteins. Furthermore, we have confirmed that these amyloidogenic peptides and proteins form α‐sheet during aggregation by CD, Fourier transform infrared (FTIR), and solution IR ³⁷ , ³⁸ , ³⁹ , ⁴⁰ (Bleem et al., unpublished). Also, independently, the FTIR peaks we have identified as signatures for α‐sheet ³⁷ , ³⁸ , ³⁹ have been observed by others in a fragment of TTR in the region of the structure predicted in our MD simulations. ⁸⁰

Hopping et al. also showed that α‐sheet has a strong FTIR absorbance band around 1,680 cm⁻¹ and a weaker absorbance band around 1,640 cm⁻¹, making it separable from β‐sheet, α‐helix, and turns. With this in mind, it is possible that previously published work has measured but not identified the presence of α‐sheet in amyloid analysis. For example, Xu et al. published FTIR difference spectra for one nonaggregating p53 mutant and three aggregating mutants. ⁵² Relative to WT, the aggregating mutants have a large absorbance band at 1,683 cm⁻¹ and a smaller absorbance band at 1,615 cm⁻¹ whereas the nonaggregating species shows essentially no difference from WT holo p53. The authors attribute the 1,683 cm⁻¹ absorbance band to an aggregation‐correlated increase in β‐structure; however, this is not consistent with β‐structure and instead this strong high‐frequency and weak low‐frequency absorption pattern is consistent with α‐sheet FTIR spectra, ³⁷ , ³⁸ supporting the hypothesis that α‐sheet plays a role in p53 aggregation.

The findings described here build upon these previous studies. α‐Sheet was observed in the apo and mutant p53 simulations presented here but not in the holo control simulations; notably, both TTR and β2‐microglobulin are β‐sandwich proteins in the same metafold as p53. In fact, alignment of the p53 crystal structure (PDB: 2OCJ) with the TTR crystal structure (PDB: 1TTA ⁸¹ ) (DeepAlign ⁸² ) indicates that the two primary α‐sheet‐prone strands in p53 (Strands S1 and S3) align with the two primary α‐sheet‐prone strands in TTR (Strands G and A, respectively) (Figure 10). α‐Sheets also demonstrated solvent‐exposed charge separation, one of the structural characteristics hypothesized to facilitate aggregation ³⁰ (Figure 7). Additionally, previously published in silico analyses predicted that many of the putative p53 α‐sheet residues presented here are also aggregation prone. ⁸³ Consequently, we hypothesize that α‐sheet may play a role in p53 dysfunction.

FIGURE 10.

p53 and transthyretin (TTR). p53 S1 aligns with TTR Strand G and p53 S3 aligns with TTR Strand A. Aligned α‐strand residues are colored green. (a) p53 crystal (Protein Data Bank [PDB]: 2OCJ). (b) TTR crystal (PDB: 1TTA)

Cancer has only recently begun to be considered an amyloid disease and more work is needed before the role of α‐sheet, if any, becomes clear. However, between existing experimental data and in silico predictions, there is increasing evidence for a common aggregation‐prone p53 species whose conformation differs from the expected β‐sheet structure. Our simulations, in conjunction with the previous work described above, offer an atomic‐level prediction for the structure of this species.

3.2. Common structural disruptions upon removal of zinc and mutation

As reported above, the largest structural disruptions observed in the simulations were common to many, and sometimes most, of the analyzed proteins, including the apo WT protein. These observations suggest that p53 has native disruption propensities that may be exacerbated by the destabilizing mutations. Several pieces of experimental evidence support our finding that both the WT and mutant proteins express the same structural disruptions. The first was reported by Butler and Loh ¹⁴ who found that zinc is 30% disassociated from WT at 310 K. NMR experiments showed that this apo conformer was structurally different from the native holo conformer; not only was it structurally different, it, like many reported mutants, was destabilized by 3 kcal/mol and was capable of inducing aggregation in WT p53. Moreover, the zinc‐region mutations capable of destabilizing the holo protein by 2–4 kcal/mol only destabilized the apo protein by fractional amounts. If demetallation resulted in different structural disruptions than the mutations, one would expect the mutation to have at least a partially additive destabilizing effect. This effect was not seen in the zinc‐region mutants, suggesting that the apo WT and the apo zinc‐region mutations have similar disruption patterns. Additive disruption was, however, seen in the DNA‐region mutant R282Q, suggesting that the zinc‐region disruption found in apo WT and the loop–sheet–helix region disruption found in apo R282Q mutant have at least partially different disruption mechanisms.

A second piece of evidence comes from Ishimaru et al. ⁵⁴ who used high‐pressure, low‐temperature denaturation to show that WT p53 could be converted to a stable, aggregation‐prone conformer whose aggregation and denaturing characteristics were very similar to the R248Q mutant. Chemical‐shift analysis showed that this conformer had altered structure in many of the same regions as the simulations reported here, with the largest shift in the S1/L1 region followed by smaller shifts in the S5/S6/S7 region, the S9 region, and the H2 region. Further support was presented by Bom et al. whose data indicated that p53 WT and the R248Q mutant demonstrate similar molten‐globule states when exposed to low pH ⁸⁴ and by Gogna et al. ⁸⁵ whose experiments showed that WT p53 in hypoxic solid‐tumor environments had similar immunoprecipitation results to mutant p53.

Additional evidence for a common disrupted conformational‐state was reported by Baroni et al. ⁸⁶ who showed that suppressor‐mutations at Codons 235, 239, and 240 were capable of suppressing the effects of 16 of the 30 most common p53 mutations, suggesting that there is a common conformer endemic to many p53 mutants that can be “rescued.” The final piece of evidence was reported by Bom et al. ⁴¹ who showed that WT and mutant aggregates were equally cytotoxic. Taken together, these data suggest that there is a common toxic conformation and both the WT and the mutant proteins are capable of adopting it, even if the pathways to do so vary from protein to protein.

There is also evidence supporting the notion of a common rescue‐mechanism. It has been shown that specific mutations can suppress the disruptive effects of multiple p53 mutations ⁸⁷ and, as discussed above, that some suppressor‐mutation locations hold more multi‐mutant stabilization‐potential than others. ⁸⁶ Indeed, Baroni et al. ⁸⁶ speculated that there was a common rescue‐mechanism because the suppressor mutations at Codons 235, 239, and 240 that were capable of rescuing more than half of the tested mutants were dominated by only one or two amino acid changes. In addition to mutations, small molecules, similar in scope to suppressor‐mutant side chains, have been shown to be effective at stabilizing and rescuing p53 mutations. ⁴ , ⁵ , ¹⁶ There are also examples of multi‐mutant p53 rescue‐compounds, such as PRIMA‐1, ⁶ which is converted to compounds that form covalent adducts with the p53 core domain ⁸⁸ (proceeded to clinical trials ⁷ ), and p53R3, ⁸⁹ as well as aggregation inhibitors capable of addressing multiple proteins, as discussed above. ⁸³ In addition to rescuing mutant p53, it has been shown that WT p53, having undergone mutant‐like conformational changes in hypoxic solid tumor environments, can reassert native WT conformations by reoxygenation or by chaperone‐stabilization by native WT p53. ⁸⁵ Taken together, these data support our findings that p53 mutants share common structural‐disruption motifs both among themselves and with WT.

4. CONCLUSIONS

Our goal in performing these simulations was to offer specific, atomic‐resolution data complementary to experimental data. To this end, we have performed MD simulations of WT p53 and 20 mutants and provided dynamic, atomic‐level details not accessible by experiment. Based on similar disruption tendencies observed for the p53 WT and mutants, we propose that p53 mutants can adopt one or more common tumorigenic conformations and that at least one of those conformations—α‐sheet—plays a role in p53 aggregation.

5. MATERIALS AND METHODS

5.1. MD simulations

A 2.05 Å resolution x‐ray crystal structure of the DNA‐binding domain of human p53 (PDB code: 2OCJ, Chain A ³⁵ ) was used as the WT starting structure. The same structure was used to create the 19 destabilized mutant structures by in silico point mutation to the appropriate residue. A stabilized WT‐like quad‐mutant with mutations M133L, V203A, N239Y, and N268D was also simulated using Chain A of the 1.9 Å PDB crystal structure 1UOL. ¹³ As discussed above, WT and mutant simulations were performed apo at 310 K. Additional WT simulations were performed holo at 298 K for comparison to experimental NMR data.

MD simulations were performed using our in‐house modeling package in lucem molecular mechanics (ilmm) with the Levitt et al. force field and established methods. ¹⁰ , ¹¹ , ¹² , ⁹⁰ Validation and detailed comparison of this force field and simulation package with experiment and with other common programs and force fields was recently reported. ⁹¹ The different programs and force fields produced similar results for native protein, but improved conformational sampling was obtained with ilmm method and reasons for that are described. ⁹¹ Protein structures were first minimized with steepest descent minimization for 1,000 steps (1 step = 2 fs). Structures were then solvated using the F3C water model ¹² in a periodic box with walls no closer than 10 Å from any protein atom. The solvent density was set to 0.993 or 0.997 g/ml, the experimental densities for water at 310 and 298 K, respectively. ⁹² Water was minimized for 1,000 steps followed by water‐only dynamics at 298 or 310 K for 500 steps. The solvent was again minimized for 500 steps and then both the solvent and protein were minimized for 500 steps.

Production simulations were performed in triplicate for 100 ns at 298 K (holo, WT) or 310 K (apo, WT and all mutants) for a total simulation time of 6.6 μs. Multiple short replicates provide better sampling than a single simulation of equivalent time. A Maxwellian distribution at low temperature was used to assign initial atomic velocities after which the temperature was increased to 298 or 310 K. A 10 Å nonbonded cutoff was used and the interaction list was updated every two steps. The NVE microcanonical ensemble was used with constant number of particles, energy, and volume. Simulations were performed at neutral pH (neutral His, positive Arg and Lys, and negative Asp and Glu). Structures were saved every 1 ps.

5.2. Simulation analysis

SASA, Cα RMSD, Cα RMSF, secondary structure (using the DSSP secondary‐structure assignment algorithm ⁹³ ), atomic‐contact analyses, and (NOE analysis were performed using ilmm ⁹⁰ analysis modules. SASA values for NMR structures were also calculated using ilmm. Contact analysis was performed using heavy atoms with a 5.4 Å cutoff for carbon–carbon contacts and 4.6 Å cutoff for all other heavy‐atom contacts. Only one atomic contact was necessary for residues to be considered in contact; contact‐occupancy was calculated as the percentage of time two residues were in contact. Contacts between adjacent residues were skipped. The Contact Walker analysis tool was used for mutant/WT contact comparison. ⁹⁴ Cα RMSD and Cα RMSF were performed using the β‐strand residues as reference (for analysis purposes, the β‐strand residues were considered to constitute the core of the protein). The equation RMSF = √3β/(8\π ²) was used to compare Cα RMSF values against crystallographic B factors. Aggregate analyses such as contact analyses, NOE analyses, Cα RMSF, per‐residue averages, and percent time in secondary structure were performed over the last 25 ns (75–100 ns) of the simulations to ensure equilibration. Mutant secondary‐structure contacts were deemed to have changed relative to WT if the mutant and WT total contact‐occupancy avg ± SD ranges did not overlap. Calculations involving DNA‐contacting residues used Residues 120, 241, 248, 273, 276, 277, 280, and 283. ³⁰ The L1/H2 distance was measured between the Cα atoms of K120 and R280. The L2/L3 distance was measured between the Cα atoms of C176 and C242. The L2/S5 distance was measured between the Cα atoms of D186 and G199.

5.3. Additional analyses

Molecular images were created in Pymol, ⁹⁵ UCSF Chimera, ⁹⁶ and VMD. ⁹⁷ Electrostatic surface visualizations were created using Adaptive Poisson–Boltzmann Solver software ⁹⁸ hosted at the National Biomedical Computation Resource (NBCR) (http://nbcr.ucsd.edu) and visualized using UCSF Chimera. Additional interactive and aggregated data analyses were performed using the DIVE visual analytics platform. ⁹⁹ , ¹⁰⁰

AUTHOR CONTRIBUTIONS

Dennis Bromley: Formal analysis; investigation; methodology; software; validation; visualization; writing‐original draft; writing‐review and editing.

Supporting information

Appendix S1: Supporting information

Bromley D, Daggett V. Tumorigenic p53 mutants undergo common structural disruptions including conversion to α‐sheet structure. Protein Science. 2020;29:1983–1999. 10.1002/pro.3921