Comparative Modeling and Analysis of Extremophilic D-Ala-D-Ala Carboxypeptidases

Abstract

Understanding the molecular adaptations of organisms to extreme environments requires a comparative analysis of protein structure, function, and dynamics across species found in different environmental conditions. Computational studies can be particularly useful in this pursuit, allowing exploratory studies of large numbers of proteins under different thermal and chemical conditions that would be infeasible to carry out experimentally. Here, we perform such a study of the MEROPS family S11, S12, and S13 proteases from psychophilic, mesophilic, and thermophilic bacteria. Using a combination of protein structure prediction, atomistic molecular dynamics, and trajectory analysis, we examine both conserved features and trends across thermal groups. Our findings suggest a number of hypotheses for experimental investigation.

Keywords: MEROPS S11; comparative modeling; extremophiles; machine learning; molecular dynamics; serine protease.

Figure 1

Microbial S11 proteases clustered according to protein sequence similarity. The sequences are described by three clusters. The thermophilic and psychrophilic sequences are separated, whereas mesophilic sequences are distributed among the three. Cluster (A) contains thermophilic sequences as well as those from B. subtilis. Clusters (B,C) comprise proteins from psychrophilic organisms, P. fluorescens and E. coli. The highlighted sequence name in each cluster indicates the centroid in sequence space. Panels (D–F) contain sequence logos associated with the three conserved active residue sequence blocks for each protein cluster.

Figure 2

Amino acid compositions normalized by sequence length and by thermal group.

Figure 3

Differences in overall composition by amino acid type. Thermophiles are relatively exiguous in uncharged polar amino acids and enriched in charged residues. On the other hand, psychrophiles are relatively enriched in uncharged polar residues; however, their composition differs considerably by sequence cluster.

Figure 4

Representative structures of bacterial S11 proteases. The structures chosen are for the proteases representing the centroids of the three clusters shown in Figure 1. (A) Q5KXJ0_GEOKA, (B) A0A119D0G3_SHEFR, and (C) A0A106BYZ8_SHEFR. The insets of each panel show the active site residues, including the catalytic SKS triad (labeled in dark cyan, dark blue, and light cyan, respectively), as well as the additional K believed to be important in substrate binding and stabilization of the transition state (light blue).

Figure 5

Rate of exposure (top) and burial (bottom) of residues by amino acid type for all proteases in each sequence cluster (Figure 1), as well as for S12 and S13 sequences (combined). “Exposed” residues had relative SASA values greater than 0.2, with the remaining residues considered to be buried.

Figure 6

Ratio of external to internal surface area as a relative measure of packing, arranged by sequence similarity. In general, a greater fraction of residue surface area is buried in low-temperature proteases; a notable outlier, B9KYG0_THERP, is discussed below.

Figure 7

Ratio of external to internal surface area as a relative measure of packing for 100 ns simulations, arranged by sequence similarity. Trajectory means were determined using parametric bootstrap confidence intervals for accounting for autocorrelation over 5000 frames; 95% confidence intervals are smaller than plotted points and hence not visible. The results are consistent with those of Figure 6.

Figure 8

Dynamic solvation index values (D), by protein; layout based on Figure 7. D value coloring ranges from green (low) to blue (high). High D values are evident in the tail domain, with psychrophiles showing particularly low values of D on select regions opposite the active site.

Figure 9

PSN cohesion by temperature, as assessed by mean degree (left) and mean core number (right). OLS fit shown by central line; the shaded area indicates the 95% confidence bands. For both measures, we see a significant decline in cohesion for proteins with higher observed environmental temperatures.

Figure 10

(left) Mean core number for moieties in exposed versus buried residues. (right) Relationship between mean core numbers and observed environmental temperature, considering only exposed or buried residues, respectively (OLS fits shown by central lines; shaded area indicates the 95% confidence bands). While buried groups are generally in much more cohesive positions, environmental temperature has a similar association with cohesion for exposed and buried groups.

Figure 11

Deep random feature embedding of conformations of the catalytic (SKS) heavy atoms over all trajectories (three highest–variance dimensions). Coloring corresponds to the position on all three dimensions in RGB space (red = first, green = second, blue = third).

Figure 12

Classification tree to separate conformational clusters using interatomic distances. Nodes list (in rows) the dominant cluster for points beneath the node, a fraction of points in each cluster beneath the node, and the fraction of the data set beneath the node. A small number of key distances approximately characterize the cluster states.

Figure 13

Most central conformations for each conformational state; catalytic SKS residues shown in atom/bond representation, with larger spheres highlighting atoms selected by classification tree (Figure 12) to distinguish conformations. Distances noted are in Angstroms and were calculated in VMD.

Figure 14

Posterior transition rate and occupancy estimates from the Markov model. States indicated by row/column. Diagonal panels show posterior occupancy probability by protein; colors indicate the thermal group. $i,j$ off-diagonal panels show 95% posterior intervals for estimated waiting times for transitions from state i to state j (in ns), by protein (Note that uncertainty is high for transitions involving states not observed for a given protein, leading to wide posterior intervals).

Figure 15

Counts of salt bridges by protein. Thermophiles show higher salt bridge counts overall, though two psychrophiles show comparable numbers.

Figure 16

Comparisons between models produced by iTasser and AlphaFold. (A) The iTasser (blue) and AlphaFold (coral) models for A4XKJ5_CALS8 have similar folds for the catalytic domain and the C-terminal “tail” domain. However, the N-terminal signal sequence (yellow) is in the expected helical conformation in the iTasser structure and an unusual extended state in the AlphaFold structure. The insets i. and ii. show these structures overlaid with the crystal structure of a homolog (PDB ID: 4K91 [74]). The iTasser structure does better at reproducing the relative orientation of the two domains. (B) The iTasser (blue) and AlphaFold (coral) models for A0A106C1X3_SHEFR behave similarly to the ones in (A), but here the unfolded N-terminal region extends far beyond the signal sequence. The insets i. and ii. show a compact $\beta$-sheet region overlaid with the experimental structure of DACC_BACSU (see (C)), which is also an S13 protease. In the iTasser structure, this small domain is compactly folded, while the AlphaFold structure predicts an unusual extended loop. (C) The iTasser (blue) and AlphaFold (coral) models for DACC_BACSU are both well-folded and compact. Neither exactly produces the expected $\alpha$-helix for the signal sequence; in the iTasser structure, it is not fully helical, and in the AlphaFold model, the helix is much longer than is typical. Both models (i. and ii.) are in excellent overall agreement with an experimental crystal structure of this protein (green, PDBID: 2J9P [68]). The signal sequence is not present in the construct that was experimentally solved.