Adaptability and Specificity: How do proteins balance opposing needs to achieve function?

Abstract

Many proteins select from a small repertoire of 3-dimensional folds retained over evolutional timescales and recruited for different functions, with changes in local structure and sequence to enable specificity. Recent studies have revealed the evolutionary constraints on protein dynamics to achieve function. The significance of protein dynamics in simultaneously satisfying conformational flexibility/malleability and stability/precision requirements becomes clear upon dissecting the spectrum of equilibrium motions accessible to fold families. Accessibility to highly conserved global modes of motions shared by family members, to low-to-intermediate-frequency modes that distinguish subfamilies and confer specificity, and to conserved high-frequency modes ensuring chemical precision and core stability underlies functional specialization while exploiting highly versatile folds. These design principles are illustrated for the family of PDZ domains.

Introduction

The functions and interactions of proteins, complexes, or assemblies are predominantly affected by their intrinsic structural dynamics, which is defined by the 3-dimensional (3D) structure. The structure, in turn, is predominantly encoded by the amino acid sequence, which selectively evolves to ensure function, thus closing the loop. Intrinsic dynamics thus play a central role in the sequence → structure → dynamics → function → sequence cyclical dependency (Fig 1). It refers to the conformational flexibility, collective fluctuations, and allosteric domain movements that are intrinsically accessible to, or favored by, the 3D architecture. During functional interactions, most proteins perform movements in line with their intrinsic dynamics, to adapt to specific changes in their environment (such as ligand binding, changes in pH, multimerization, or complexation) while retaining to a large extent their fold. There are, of course, exceptions such as intrinsically disordered proteins/regions, which occupy unfolded states and can adopt different folds in different situations, such as binding to different partners [1]. Adaptability of the fold to different environments and intermolecular interactions is an evolutionary requirement for stability; but specificity is equally required for function. How do proteins fulfill such apparently conflicting requirements of flexibility/adaptability, on the one hand, and precision/specificity, on the other?

Figure 1: Significance of protein dynamics in enabling functional adaptation and specialization within protein families.

Various events, such as gene duplication and random mutagenesis, give rise to ensembles of homologous protein/domain structures, also called fold families (structurally aligned at the bottom left for PDZ domains). Each family has its unique structure-encoded dynamics under equilibrium conditions, represented by a spectrum of normal modes of motion, and a family-specific signature dynamics (middle bottom for PDZ domains). The proteins whose structures and dynamics are well adapted to do biological functions (in the elliptical shell) are evolutionarily retained/selected and vice versa. Hence, evolution leads to an iterative refinement of protein structure and dynamics to enable fold adaptation and functional specialization (curled blue arrow on the left). An ensemble normal mode analysis allows us to identify and compare the different frequency modes encoded by each structure in the ensemble. Notably, motions at two ends of the spectrum, shortly called global/slowest and very fast modes, are conserved (highlighted in red/brick in the elliptical shell) for diametrically opposed reasons: the former underlies the global mechanics/rearrangements required for adaptation to intermolecular interactions, including ligand binding and allosteric responses; the latter reflects key interactions that cannot tolerate sequence/structure perturbations, such as the highly stable packing at the core of the fold, or specific interactions at chemically active (e.g. catalytic) sites that are precisely retained as the fundamental characteristic feature of the family. LTIF modes usually define subfamilies, that tolerate variations within subfamilies, and drive fold adaptation and functional differentiation. They are conserved within, but vary between, subfamilies. Fast/intermediate modes show maximal variance between members within and across subfamilies (green portions of the ellipse).

The answer perhaps lies in two important facts. First, a ‘measured’ conformational flexibility entropically adds up to support stability, as long as the interactions that maintain the fold or assembly are not disrupted. Excessive conformational flexibility, on the other hand, would disrupt the favorable intra- and intermolecular interactions/energetics, thereby adversely affecting stability or functional associations. For example, the existence of a flexible/disordered loop at the binding epitope is usually required by design to optimize the interfacial interactions and ensure stable binding; but complete lack of a structure or scaffold supporting this loop’s rearrangements may easily lead to loss of function (inability to bind). Second, regarding precision or specificity, the flexible (loop) regions actually offer excellent frameworks to accommodate varieties of amino acids, the specific sequence of which would also drive specificity. Not surprisingly, recognition loops such as the complementarity determining regions (CDRs or hypervariable loops) of antibodies enjoy high structural variability or dynamics that goes hand-in-hand with their sequence variability. In a sense, flexibility assists in specificity, by allowing for the evolutionary selection of the amino acids that would ensure specificity.

A closer look at dynamics as a means of reconciling flexibility and specificity

A deeper understanding of the subtle balance between flexibility/adaptability and precision/specificity using both first physical principles as well as evolutionary perspectives is of crucial importance in many research areas, from synthetic biology to protein engineering to pharmacology. Attention has been directed to assessing protein dynamics as a means of reconciling the need for flexibility and specificity [2–4]. A growing field centers around protein dynamics for analyzing protein evolution, both ancestral [2–11] as well as synthetic/directed [2,5,12–16]. Recent advances in cryo-electron microscopy and computational methods are facilitating the study of large and dynamic systems [17,18], including whole viral capsids in different conformational states [19].

A key step for these goals is dissection of the dynamics of the protein of interest. In physical terms, this amounts to decomposing the intrinsic dynamics into a spectrum of modes and evaluating the significance of the modes of different types/frequencies in fulfilling different roles or enabling different types of function, as will be elaborated below. Such studies are facilitated by elastic network models (ENMs) [20–23], especially the Gaussian network model (GNM) [21] and the anisotropic network model (ANM) [20], which provide robust analytical solutions to this problem [15,24,25].

While computational tools and hardware for all-atom molecular dynamics (MD) simulations are constantly improving, evaluation of intrinsic dynamics using ENMs remains orders of magnitude faster, and ENM-predicted global modes show strong correlations with the essential motions observed in sufficiently long MD simulations [26,27]. As such, ENMs have provided efficient frameworks for analyzing biomolecular systems conformational dynamics over the last two decades [20–23]. They have also provided new insights into the role of intrinsic dynamics in achieving the balance between evolutionary adaptability and functional specificity as described next.

Comparative studies of intrinsic dynamics among family members reveal how dynamics evolved to ensure specialization or differentiation

The high level of redundancy in the Protein Data Bank (PDB) [28] has recently enabled a number of comparative studies, showing that just as protein families have conserved sequence and structural motifs, they also have distinct conserved patterns of signature dynamics [2–4,8,29,30]. This is particularly evident when the dynamic fluctuations are decomposed into different modes of motion [2,3,8,29]. SignDy [3] was recently introduced as a pipeline for characterizing the signature dynamics of protein families, using the ProDy [31] framework for sequence, structure, and dynamics analyses. Briefly, SignDy workflow consists of: (1) Alignment of an ensemble of proteins structures that share the same fold; (2) Normal mode calculation for each ensemble member using ENMs; (3) Evaluation of the signature dynamics by determining the shared global modes; and (4) Classification of family members based on the most discriminative modes that drive functional specificity or evolutionary differentiation. The Reuter, Echave and Micheletti groups have also made significant contributions in this area [2,32,33] and developed servers such as WEBnm@ [34] and ALADYN [35]. The Jernigan lab also showed the importance of evolving dynamics for the identification of functional sites [36–38].

Application to CATH superfamilies and specific case studies for transporters, triosephosphate isomerase (TIM) and related α/β barrel-containing enzymes, type-1 periplasmic binding protein clamshell domains, and lipoxygenases [2,3,8,39] have revealed how proteins select from the available repertoire of normal modes to perform various types of activities (Fig. 1), ranging from global cleft opening/closing and allosteric transitions, to high frequency fluctuations reflective of the energy localization at the core of the folds. As illustrated in Figure 1, we can approximately divide the spectrum of modes into four regimes: slowest or global modes, which account for the large, collective rearrangements of the overall structure, low-to-intermediate frequency (LTIF) modes, intermediate-to-fast modes, and very fast modes; the respective mode numbers vary with protein size. The emerging features are: (i) The global modes are conserved across protein fold family members [2–4,8,9,29,30,40]; (ii) subfamilies and specific family members exhibit differentiated motions in the LTIF regime, which influence functional specificity as revealed by recent studies [3,8]; (iii) The fastest modes are also highly conserved [3,8], their conservation being required for chemical precision, reactivity (e.g. catalysis), or structural stability [41]. The first two groups and their variance across family members define the signature dynamics of the family.

A case study of PDZ domains provides insights into mechanisms of adaptation and specialization

PDZ domains are found in a variety of signaling and scaffolding proteins [42–46], especially at cell junctions such as synapses [47–50]. PDZ domains consists of five β-strands and two α-helices (Fig. 2A). They exhibit a wide range of conformational plasticity and specificity [43–46]. Their extensive functional range renders them critical drug targets [42]. The β-sheet core and close juxtaposition of α2, decorated by flexible loops especially the β2-β3 loop (yellow), provide a highly versatile architecture for efficiently recognizing and binding ligands.

Figure 2: An example signature dynamics analysis of the PDZ domain family illustrates the roles of different mode regimes.

(A) The representative PDZ domain from CATH [55]: sorting nexin-27 (SNX27) bound to PTHR PDZ-binding motif (PDB: 4Z8J) [56], which is critical for PTHR signaling and bone development [56,57]. (B) Overlay of the ensemble of PDZ domain family structures, colored by the extent of structural variation with the most variable regions in red, structurally conserved parts in blue, and regions with intermediate variability in white. The ensemble was aligned using mappings from DALI [58] and refined using RMSD filters of 1.0 Å to reduce redundancy and 10.0 Å to exclude outliers. (C-D) SignDy analysis of this ensemble. (C) Conservation of GNM modes (green curve, based on pairwise overlaps between matched modes of family members) shows high conservation of global modes (dark green shaded; 1 ≤ k ≤ 3), less conservation for intermediate/fast modes (light yellow), and conservation again for the very fast modes (dark orange shaded). LTIF modes show a mixed behavior: the slow portion (light green shaded) shows a moderate conservation and intermediate portion (yellow shaded) is less conserved like the fast modes. The red curve shows the degree of collectivity, which shows anticorrelation with the green curve throughout the spectrum except for global and slow modes. (D) Mean-square (ms) fluctuations (MSFs) of residues for different frequency windows (left) and corresponding color-coded structures (right). In the left panels, the dark blue curves represent the average MSF profile of the ensemble, blue shades represent the standard deviation, and the light blue area displays the range. Global and LTIF modes are highly collective and activate the β2-β3 loop, followed by β1-β2 and α2-β5 loops; the highest frequency modes 94–95 point to the core region (at β3 N-terminal end) as a center of energy localization.

Not surprisingly, the PDZ domain has been the focus of several studies exploring its dynamics and allostery, as well as thermodynamics vis-à-vis its ligand-binding properties [46,51–53]. We present in Figures 2 and 3 results from SignDy analysis of a PDZ domain structure ensemble. The GNM mode spectra for family members yielded several features consistent with those observed in other fold families: First, the global modes are highly conserved among family members (shaded green in Fig 2C). The level of conservation levels off at about mode 15, after which it remains flat (with fluctuations), except for the highest frequency end of the spectrum (orange), where we observe a sharp peak. This high conservation at the two ends of the spectrum has been noticed previously and attributed to function and stability requirements (see Fig 1). Second, while the degree of collectivity of the modes (red curve in Fig 2C) decreases with increasing mode number for global modes, a heterogeneous behavior is observed in the LTIF regime: the slower portion (4 ≤ k ≤ 10) follows the same pattern as the global modes, but modes 11 ≤ k ≤ 20 show an anticorrelation between collectivity and conservation, i.e. the more localized motions tend to be more conserved.

Figure 3: Cross-correlations between residue movements and their variance, in different frequency regimes, illustrated for the PDZ domain family.

Four panels show the mean (left) and standard deviation (right) values for the cross-correlations between residue movements, corresponding to global, LTIF, intermediate-to fast and very fast regimes, from top to bottom. The left matrices are colored from anti-correlated in dark blue to positively correlated in dark red via uncorrelated in green. The right matrices are colored from less variable in dark blue to more variable in dark red. Note the strong correlations in the global modes (top left) and the strongly coupled variations in cross-correlations in the fastest modes (bottom right), which show the concerted movements in the soft modes and the necessity for co-varying residue motions in the fastest/stiffest modes.

The distributions of residue mean-square fluctuations (MSFs) in different regimes (Fig 2D) show that the global hinge regions (minima in top panel) exhibit minimal variations among family members. The fastest modes (lowest panel and ribbon diagram) are dominated by fluctuations of L76-Q77 at the N-terminal end of β3, a critical hotspot that bridges binding site element β2 with the rest of the structure. This is in striking contrast to the slow and LTIF regimes where the MSFs are distributed across the domain (Fig 2D top three panels). We also notice a relatively greater degree of variability for the LTIF modes (middle two panels) in line with previous studies [3,8]. The modes in this regime are typically conserved within subfamilies but differentiated across subfamilies, thus defining subfamily specificities. Notably, high variances in LTIF modes occurs at the β2-β3 and α2-β5 loops. The β2-β3 loop and the crevice between β2 andα2 have been identified as ligand binding sites [54].

Residue-residue cross-correlations in different regimes can be seen in Fig 3. The left panels display the average cross-correlations in different regimes, and right panels, their variance. As expected, the global modes (top panel) exhibit robust cross-correlations between en bloc movements of pairs of structural elements moving in the same (red) or opposite (blue) directions. Strong cross-correlations can be detected in the LTIF regime as well, although concertedly moving regions become smaller, and the couplings weaker. Again, the largest variations across family members in this regime take place within the β2-β3 loop, and between that loop and the spatially neighboring β4-α2 loop, pointing to the dominant role of this region in functional differentiation consistent with experimental data. The remaining modes (k > 21) lack any shared feature that stands out, but the variations in pairwise correlations in the fastest mode regime (bottom, right panel) are far from random, reflecting the necessity of coupled rearrangements in the densely packed (core) substructures. These constraints are consistent with the decrease in the rate of evolution with increasing packing density, noted by Echave [32].

Conclusions and Future Directions

In this review, we emphasized how proteins select from a repertoire of modes to enable their functional evolution, and reconcile the need for flexibility/malleability, on the one hand, and precision/specificity, on the other. The slowest modes of protein domains, or small proteins, are recruited and conserved by a variety of sequences that share the same fold, despite their low sequence similarity, because these define highly versatile mechanisms (e.g. cleft opening/closing) that can be ‘plugged in’ within the complex machinery of biomolecular systems. On the other hand, the LTIF regime is where much of the specificity in intrinsic dynamics of family members is found. The focus on dynamics brings into consideration a new way of classifying proteins, based on their shared or differentiated dynamics. The latter could be equally (if not more) informative compared to sequence- or structure-based classifications, as it relates to functional mechanisms and could provide a new dimension in protein design and engineering.