Systems biology approaches to macromolecules: the role of dynamic protein assemblies in information processing

Abstract

Macromolecular protein assemblies govern many cellular processes and are disturbed in many diseases including cancer. Often seen as static molecular machines, protein complexes involved in signal transduction networks exhibit intricate dynamics that are critical for their function. Using the RAS-RAF-MEK-ERK pathway as example we discuss recent progress in our understanding of protein complex dynamics achieved through mathematical modelling, computational simulations and structural studies. The emerging picture highlights that both spatial and temporal dynamics cooperate to enable correct signal processing and the fine tuning of timing, duration and strengths of signalling. These dynamic processes are subverted by oncogenic mutations and contribute to tumorigenesis and drug resistance.

Introduction

The assembly of macromolecular protein complexes is now firmly established as an organising principle of biological processes. Originally viewed as molecular machines that carry out purely material tasks, such as gene transcription, protein translation, protein degradation, or biosynthetic and metabolic processes [1], protein complexes now also being recognised as information processing systems that coordinate and compute cell fate decisions. A distinguishing feature between production type and information processing protein complexes is the dynamic re-organisation of the latter in response to stimuli, which seems an integral part of their compute functions. Interestingly, these dynamic features recur across scales from network assemblies to protein dimers. Now, systematic principles start emerging driven by systems biology approaches that combine omics studies, detailed structural analysis and mathematical modelling. Here, we review these advances presenting some salient underlying theoretical principles and focussing on experimental evidence over the last 5 years that has translated these principles into actionable knowledge. We conclude with examples for the application of this knowledge in drug target discovery and development. We will mainly use the RAS-RAF-MEK-ERK pathway as paradigm (Fig. 1) but will highlight the generalisable principles. This pathway regulates cell proliferation, differentiation, and survival, and is hyperactivated in ~50% of human cancers [2] and several developmental disorders [3].

Figure 1. An illustrative cartoon of the RAS/RAF/MEK/ERK pathway.

Typically, this pathway is activated by receptor tyrosine kinases (RTKs), which upon ligand binding recruit proteins that activate RAS proteins. Activated RAS can bind to downstream effectors including RAF kinases mutually facilitating dimerization. RAF dimers phosphorylate and activate MEK, which in turn phosphorylate and activate ERK. ERK has >650 targets both in the cytosol and nucleus [41].

Some principles of dynamic protein complex organisation

Dynamic (dis)assembly requires that interaction strengths are modest or tuneable. Many receptors activate RAS proteins by facilitating GTP loading of RAS, which causes a conformation change that enables RAS to bind to effectors and activate downstream signalling. GTP hydrolysis reverses this conformational change releasing the effectors and terminating signalling. Interestingly, all effectors compete for the same binding site in RAS making the composition of RAS-effector complexes strongly dependent on the concentration of activated RAS. If activated RAS is low, high affinity effector complexes dominate, while at just threefold higher active RAS concentrations cells become transformed, and complexes of RAS with abundant low affinity effectors prevail [4]. Different types of oncogenic RAS mutations also change the preference of effector binding, potentially explaining tissue specific effects of RAS mutations [5]. Thus, both the type of mutation and the relative abundance of activated RAS and its effectors determine the biological outcome of RAS signalling by dictating the composition of RAS-effector complexes.

This principle of generating specificity in signalling by dynamically changing the composition of protein complexes due to competing interactions seems general. An analysis of competitive protein-protein interactions (PPIs) in the epidermal growth factor receptor (EGFR) network identified many mutually exclusive PPIs, and showed that artificially changing the abundance of interaction partners specifically rewired pathways [6]. One would expect that this rewiring occurs on a sliding scale, where increasing the abundance of one competitor gradually decreases complexes containing the other competitor. That is not always the case. If competing protein interactions are combined with affinity modulation, abrupt transitions between different protein complexes can occur [7]. RAF-1 can bind to either MST2 or MEK, inhibiting pro-apoptotic signalling by MST2 or stimulating proliferation by activating MEK-ERK signalling. Feedback phosphorylation from the MST2 pathway on S259 directs RAF-1 to MST2 binding, whereas pS259 dephosphorylation induced by growth factors enables RAF-1 to bind to RAS and activate MEK [7]. This PPI transition is switch-like and commits the cell to mutually exclusively fates, proliferation or apoptosis. Thus, competing PPIs can serve as quantitative sensors of protein abundances, but also as switches when the affinity of PPIs is dynamically controlled (Fig. 2).

Figure 2. Competing protein-protein interactions result in a switch-like behaviour of RAF and MST2/Hippo signalling.

(A) Interactions between RAF and MST2/Hippo signalling pathways. (B) Schematic representation of responses of RAF and MST2 activities to the activation of RAS for low and high levels of AKT activity.

The idea of using affinity changes to steer the specific and dynamic formation of signalosomes is also inherent to receptor tyrosine kinase (RTK) and T-cell and B-cell receptor (TCR/BCR) signalling. Activation of these receptors induces tyrosine phosphorylation of their cytoplasmic domains and of specific adaptor proteins which serve as docking platforms for the assembly of signalling complexes that induce the activation of RAS and other essential pathways. These assemblies feature both temporal and spatial mechanisms that enhance receptor specificity. For instance, the TCR can discriminate foreign from self- antigens based on binding half-lives differing only by seconds [8]. This exquisite specificity is achieved by kinetic proofreading (KPR), where a sequence of biochemical reactions act as timer that only triggers a response when all reactions are finished. In the case of the TCR completion of the sequential phosphorylation events is required for the full assembly of the critical signalling complexes. The speed of the timer is determined by the number of reactions but also by kinetic features. The TCR KPR chain contains a kinetic bottleneck reaction, where a critical tyrosine on the LAT adaptor protein deviates from the phosphorylation consensus sequence and is phosphorylated very slowly [9]. This phospho-tyrosine recruits phospholipase-Cγ, which mediates RAS activation. This essential step for T-cell activation only can happen when KPR is completed. The spatial component is co-localisation. Confining proteins to a two-dimensional space by recruiting them to the membrane increases diffusion limited reaction rates as well as the average lifetime of PPIs [10]. In fact, artificially dimerising the EGFR triggers autophosphorylation and signalosome assembly but cannot activate RAS. Downstream signalling only occurs when the EGF ligand induces receptor oligomerisation [11]. We discuss these spatial mechanisms in the next section.

Protein clusters and condensates

Ligand induced clustering is a general phenomenon in receptor signalling. Upon stimulation with a ligand or treatment with inhibitor, HER2 and HER3 receptors can form clusters consisting of hundreds of receptor molecules [12]. The TCR and BCR even form specialised structures called immunological synapses (ISs), where the TCR/BCR is at the centre surrounded by a spatially orchestrated arrangement of adaptor proteins and signal transducers [13]. The IS corals signal transducers in a small space while physically excluding phosphatases, which antagonise the assembly of signalling complexes by dephosphorylating docking sites. A mathematical model supported by experiments shows that this topological constraint is sufficient to explain ligand discrimination by the TCR [14]. Reconstituting TCR signalling to actin remodelling in model membranes showed that clustering increased the local protein concentration to produce liquid-liquid phase separation (LLPS), where kinases were enriched and phosphatases excluded. This biophysical separation enhanced signalling and actin polymerisation [15]. This phase transition not only brings molecules together but also can affect the activation kinetics of individual proteins. Efficient RAS activation in T-cells requires the guanine-nucleotide-exchange-factor SOS, which undergoes an activating conformational switch when recruited to the membrane. The delay (~50 seconds) between recruitment and activation is another checkpoint in KPR that is modulated by phase transitions. LLPS is a biophysical phenomenon where locally high protein concentrations condense to form membrane-less liquid or gel-like organelles. Originally described as stress response (stress granules) LLPS occurs in chromatin regulatory, transcriptional and signalling complexes [16]. It is a versatile way to form large supramolecular protein assemblies that need to transiently coalesce for carrying out a specific function. However, the participating proteins can produce self-perpetuating assemblies, which may be involved in forming a ‘biochemical memory’ [17]. This mechanism could facilitate rapid adaptations to unexpected challenges, e.g. stress and environmental changes, where genetic selection would be too slow to enable survival. The motif of clustering repeats itself at the level of downstream signal transducers. RAS proteins form transient nanoclusters upon activation, which can determine both the strength and specificity of downstream signalling [18]. RAS nanoclusters activate effectors - such as RAF - and downstream signalling by breaking up autoinhibitory conformations and increasing the lifetime and density of activable RAF kinase domains [19]. This is a subtle but important re-interpretation of pathway activation cascades that are ubiquitous in signalling. It suggests that pathway activation is a linked chain of priming and activation events where protein conformations and PPIs change to create a ‘readiness’ level at each step that needs a second event to produce activation. This view may enhance filtering out of background noise and enhance signalling specificity. A stimulus may induce many primed complexes, but only some of them may proceed to activation depending on the signal strength and duration or a complementary signal coming in.

Dynamic allosteric regulation of RAF kinases

PPIs often also include a component of allostery as binding can change or constrain protein conformations. An intricate example is RAF dimerization, which is extensively investigated as it is one of the main causes of drug resistance in melanoma [20]. BRAF and CRAF are kinases which phosphorylate and activate MEK and subsequently the ERK pathway. Their regulation is complex involving RAS binding and membrane recruitment, dephosphorylation of inhibitory residues and phosphorylation of activating, residues, homo- and heterodimerisation inducing allosteric activation [21]. RAS and RAF binding cooperate in mutually enhancing dimerization and signalling output [22]. Dimerization is critical for oncogenic transformation. Interestingly, mixed dimers between wild-type and oncogenic mutant KRAS, i.e. dimers where only the mutant KRAS protomer binds RAF, fail to induce RAF dimers and RAF activation, suppressing oncogenic transformation but conferring higher resistance towards MEK inhibitors [23]. The mechanism is unknown, but a plausible hypothesis is that MEK inhibition breaks the negative feedback loop whereby ERK subdues the activity of the RAS activator SOS [24]. This would enhance activation of the wildtype KRAS resulting in the formation of a KRAS dimer with two active protomers and subsequent RAF kinase dimerization and enhanced resistance to MEK inhibition. BRAF-CRAF heterodimers have the highest catalytic activity followed by BRAF homodimers [25]. Interestingly, within the dimer one protomer can allosterically activate the other protomer without the need for possessing kinase activity [26]. Thus, the activator protomer can be replaced by a catalytically inactive RAF or the scaffold KSR1 if they are in the activated conformation, i.e. the R-spine aligned, either through appropriate mutations or phosphorylation of activating residues [26]. This explains the conundrum that some BRAF mutations found in cancer actually impair or abolish kinase activity. These mutants signal through heterodimerisation with CRAF [27]. This also means that a protomer bound to a RAF inhibitor still can allosterically activate the other protomer (Fig. 3A).

Figure 3. Allosteric interactions and transactivation of RAF dimers.

(A) Different ways of activation of activation of RAF monomers and transactivation of RAF dimers. From left to right: RAF monomers are activated following the binding to RAS but have relatively low kinase activity; Hetero-dimers of active CRAF-BRAF have high activity; Hetero-dimer of CRAF and mutant kinase-dead BRAF where transactivated CRAF protomer has high kinase activity; BRAF-KSR heterodimer, in which KSR transactivates BRAF; Partially inhibited CRAF-BRAF heterodimer where one protomer has bound an inhibitor but the other protomer is highly active. (B) Synergistic inhibition of RAF dimer with two inhibitors with conformation specificity. Because of intra-molecular thermal motions, the αC-helix and the DFG-motif flip between an active (IN) and inactive (OUT) conformations in an inhibitor-free RAF dimer. Due to spatial constraints, two αC-helices cannot simultaneously assume OUT conformations in a dimer. Type I½ (e.g. Vemurafenib) RAF inhibitors selectively bind to a DFG-IN, αC-OUT conformation of a protomer. This shifts the αC-helix of the other protomer into an αC-IN conformation, and the second molecule of Type I½ RAF inhibitor cannot bind the other protomer in a dimer. However, such protomer acquired increased affinity for Type II RAF inhibitor (e.g. Sorafenib or TAK-632). Thus, Type I½ and Type II RAF inhibitors can synergistically inhibit RAF dimer, whereas neither inhibitor is effective when applied separately.

Interestingly, experimental observations suggest that if one protomer in the dimer is bound to a RAF inhibitor, the other protomer is likely inhibitor free and activatable by the inhibited molecule [28]. As most RAF inhibitors induce dimerization, this mechanism results in paradoxical pathway activation and drug resistance. The puzzling phenomenon that a kinase dimer cannot bind drugs that have low nanomolar affinities and are present in high micromolar concentrations was explained by a biophysical model that considered thermodynamic factors [29]. This model shows that the thermodynamically favoured protein complex is a RAF dimer where one protomer binds drug and the other is drug free. Structurally, this translates into an asymmetry of the dimer and that the affinity for binding another RAF inhibitor molecule is dictated by the structures of the free protomer and inhibitory drug. This insight led to the counterintuitive prediction that two structurally different RAF inhibitors carefully chosen to fit the structural and thermodynamic requirements could pharmacologically synergise to inhibit the RAF dimer. This prediction was borne out through a refined model that considers activating and inhibitory phosphorylations, thermodynamic and structural constraints, network context and mutational background [30]. The model identified different types RAF inhibitor combinations that synergises to suppress paradoxical ERK pathway activation and the proliferation of cancer cells. These results emphasize the importance of the molecular context.

Recent structural studies of BRAF macromolecular complexes by cryo-electron microscopy have refined this concept (Fig. 4). Kondo et al. resolved the structure of a phosphorylated BRAF kinase domain dimer complexed to the dimeric scaffold protein 14-3-3 [31]. It is an asymmetric dimer where both protomers are in an ON state ready for catalysis, but where the C-terminal tail of one protomer obstructs the active site of the other protomer (Fig. 4A). This configuration is stabilized by binding of the phosphorylated C-terminal tail to 14-3-3 proteins. 14-3-3 is an obligatory dimeric protein that binds to phospho-serines/threonines and enhances RAF dimerization [25]. Park et al. resolved the structure of full-length BRAF complexed with MEK1 and 14-3-3 dimer [32]. They found an inactive BRAF-MEK1 complex restrained by a clamp formed by 14-3-3 binding to N- and C-terminal residues in BRAF (Fig. 4B). This constellation prevents membrane recruitment and BRAF dimerization but allows MEK binding. Activation reconfigures 14-3-3 binding to the C-terminal sites now crosslinking the two BRAF protomers (Fig. 4C) promoting dimerization and activation [32]. Interestingly, this structure also suggested that ATP binding keeps RAF in an inactive monomeric conformation, which was confirmed in an independent study. 14-3-3 binding and most oncogenic BRAF mutation overcome this inhibition by rotating the dimerization interface towards each other and enforcing dimerization [33].

Figure 4. 14-3-3 proteins promote both monomer inactive and dimer active configurations of RAF kinases.

(A) Structure of a BRAF dimer stabilized by 14-3-3 proteins (PDB code: 6UAN), in which the C-terminal tail of one BRAF protomers (active BRAF, blue) inactivates the other BRAF protomer (inactive BRAF, pink). (B) Structure of an autoinhibited BRAF-MEK1-14-3-3 complex (PDB code: 6NYB). A 14-3-3 dimer binds pS365 and pS729 residues of BRAF on the opposite sites of the molecule, stabilizing the autoinhibited inactive monomer conformation of BRAF. (C) Structure of active, dimeric BRAF bound to two molecules of MEK1 (PDB code: 6Q0J). The 14-3-3 dimer binds to the pS729 residues of the two BRAF protomers in a dimer stabilizing the active dimeric conformation of BRAF.

These structural insights confirm previous biochemical results suggesting that 14-3-3 could stabilize both active and inactive conformations of RAF [34]. The N-terminal 14-3-3 binding site needs to be dephosphorylated to allow RAS binding and activation [35], but 14-3-3 binding to the C-terminal site is essential for allowing ATP binding and overcoming inhibition [36]. Thus, RAS binding is accompanied by 14-3-3 switching from intramolecular to intermolecular binding that promotes RAF dimerization and subsequent MEK phosphorylation.

Conclusions

Macromolecular protein complexes not only organise signalling by bringing the required partners together, but also provide intricate dynamic regulation through posttranslational mechanisms, oligomerization, LLPS, and allosteric mechanisms. Even allosteric regulators are allosterically regulated: MEK binding to KSR1 allosterically enhances KSR1’s ability to allosterically activate BRAF [37]. These intricacies argue the case for context dependent drug design rather than the traditional approach of targeting molecules in isolation. For instance, the difficulties of blocking RAS proteins directly can be surmounted by targeting its dimerization and nanoclustering [38]. Mathematical models and molecular dynamic simulations are leading the way to understand the information processing logic of dynamic properties and harness them as drug targets [39,40].

Highlights.

• Cellular information processing involves dynamic changes in protein assemblies

• Protein assembly kinetics are controlled by posttranslational modifications, localization, and liquid-liquid phase separation

• Competing protein-protein interactions combined with affinity changes can function as signalling switches

• Asymmetric dimerization of kinases can cause resistance to kinase inhibitors

• 14-3-3 proteins can stabilize both inactive and active conformation of RAF kinases