The role of ordered cooperative assembly in biomolecular condensates

Abstract

Biomolecular condensates, regardless of whether they exhibit liquid-like properties, are in many cases not fully amorphous, but instead exhibit partial degrees of local structure and order. Here, we discuss how ordered interactions may underlie the cooperative assembly and cellular function of a wide variety of partially ordered macromolecular assemblies.

Diversity of biomolecular condensates

In a biological context, liquid–liquid phase separation (LLPS) is a cooperative assembly process driven by multivalent, transient interactions that results in the spontaneous separation of a homogeneous solution into solute-rich and solute-poor regions (Supplementary Fig. 1a)^1,2. Proteins that undergo LLPS tend to have a high level of intrinsic structural disorder, and phase separation via weak multivalency can result either from their relatively nonspecific, promiscuous interactions, or from defined, structured multivalent interactions. The phenomenon of LLPS has been extended to include a broad range of dynamic, nonstoichiometric assemblies broadly termed biomolecular condensates. Not all condensates are liquid, however, but can be solid or gel (solid-liquid mixture) phases. It is also possible that some biological condensates exhibit properties between a solid and a liquid, similar to but likely more complicated than liquid crystals. The term condensate has also been used to describe biomolecular assemblies that form without a phase separation as such, for example through steric constraints or clustering around high-affinity scaffolds¹. Thus, biomolecular condensates are structurally highly versatile and may emerge via distinct mechanisms.

Some biological condensates can be approximated as being essentially amorphous, with negligible propagation of local structure or order. At the opposite extreme, biomolecular assemblies not generally thought of as condensates, such as viral capsids and cytoskeletal filaments, arise through defined, ordered interactions between individual subunits. Ordered interactions between subunits present an opportunity for cooperative assembly mechanisms³ beyond those allowed by multivalency alone, a process we term ordered cooperative assembly (OCA) (Supplementary Fig. 1b). OCA can occur when the growing assembly yields additional, stabilizing contacts that are essentially absent in the interaction between isolated monomers, for example in the growth of microtubules from tubulin dimers. Higher order interactions are also possible, such that component C can only bind to an interface produced after complex formation by components A and B, for example in the assembly of the ribosome. Allosteric interactions can also drive OCA in cases where assembly itself acts as an allosteric effector promoting further assembly. This is the case in actin filaments, where the initially assembled actin trimer undergoes a conformational switch to provide higher inter-subunit affinity. Importantly, it is cooperativity, rather than local order itself, that allows for such highly nonlinear assembly dynamics.

It is now becoming clear that most cellular macromolecular assemblies lie between the two extremes of completely ordered or disordered. In addition, as illustrated below, macromolecular assemblies may in general have properties that fall outside what is usually meant by a condensate in a physical sense. Thus, following earlier usage, we use the more general term pleomorphic ensemble (PE)⁴ to refer to non-stoichiometric biomolecular assemblies that have no well-defined final shape or size and are dynamic at cellular timescales, and that are not necessarily liquid in their physical properties (Supplementary Fig. 1c). Within this definition, PEs can encompass assemblies with multiple interior organizational assembly mechanisms; importantly, both condensates specifically and PEs generally can exhibit both short-range and long-range structural order, allowing them to benefit from OCA. Via a few examples, we discuss multiple mechanisms that drive the formation of PEs, and how the contributions from ordered and disordered assembly may impact their biological functions.

Order and disorder in PE assembly

Non-cytoskeletal filamentous structures are increasingly observed in cell biology (Supplementary Fig. 1d)⁵. One such example is the Wnt ‘signalosome’ in Wnt/β-catenin signalling. Wnt binding on the cell surface triggers an assembly of signalling components on the membrane as well as in the cytosol, driven in part by the oligomerization of the scaffolding protein Dishevelled via its DIX domain. This results in the inactivation of another assembly of proteins, the destruction complex, that otherwise tags β-catenin for degradation. Assembly of the destruction complex likewise involves filament formation by the scaffolding protein Axin, which also oligomerizes via a DIX domain. Thus, core components of the signalosome and destruction complexes separately undergo ordered assembly. Importantly, the DIX domains of Disheveled and Axin co-oligomerize, and recruitment of Axin to the signalosome via this coassembly leads to the inactivation of the destruction complex. Intriguingly, a recent study shows that, in vitro, the DIX domain of Dishevelled assembles into an antiparallel, double-helical filament. This could serve as the basis for OCA, for example, via the interaction of monomers across individual protofilaments. Further, the DIX domain of Axin severs and caps DIX filaments of Dishevelled by binding to the filament ends (Supplementary Fig. 1d)⁶. While the function of this anti-cooperative behaviour is uncertain, it may act to set the overall stoichiometry of the signalosome, regardless of the precise length of the Dishevelled filament⁶. This example illustrates how ordered assembly can potentially act to both positively and negatively regulate PE growth.

PE assembly on a template

In many instances, PEs are constrained by or enriched on a template, for example the cell membrane, DNA or scaffolding proteins. PEs assembled on F-actin constitute one such example. Certain F-actin binding proteins⁷ (ABPs) have been shown to cause allosteric, potentially long-range, cooperative changes in the filament, which in turn affect ABP binding at other binding sites. A particularly strong case has been presented for clusters of the F-actin depolymerizing factor cofilin, where assembly of a seed of two cofilins on actin is proposed to tilt four contiguous actin subunits and consequently promote cooperative growth of a cofilin oligomeric array⁷. Thus, assembly itself acts as an allosteric effector promoting further assembly. The actin binding protein drebrin has likewise been shown to alter the F-actin helical pitch, potentially regulating the ability of other proteins to bind to and remodel F-actin in the context of neuronal synapse maturation⁷. How and whether such long-range allosteric interactions influence other template-based PEs, for example in the case of chromatin, represents an intriguing question for future research.

Simultaneous mechanisms of PE assembly

Adhesion complexes constitute another system in which PE assembly is increasingly being explored through the lens of weak, multivalent interactions (Supplementary Fig. 1e)². Interestingly, there is support for the contribution of a relatively unusual allosteric OCA in the assembly of cellular adhesions. For example, integrin binding to the extracellular matrix (ECM) is opposed by the energetic cost of localized membrane bending. However, the energetic cost of recruiting additional integrins to these regions is predicted to be much lower⁸, resulting in cooperative, all-or-none adhesion formation. More broadly, it is likely that adhesion complexes may combine multiple forms of OCA. For example, formation of integrin-based focal adhesions and downstream signalling appear to be coupled to the formation of ordered oligomeric assemblies of focal adhesion kinase⁹. Similarly, in tight junctions, the transmembrane proteins occludin and claudin assemble into filaments that zipper neighbouring cells together, and are in turn linked to the cytoskeleton by ZO-1, a large scaffolding protein that, in vitro, can undergo LLPS driven by weak, multivalent self-interactions¹⁰. We speculate that these examples from cell adhesion biology, where the molecular components and ultrastructures are comparatively well-characterized, may be helpful in revealing the full complexity of PEs in other contexts, for example at the neuronal and immune synapses.

New frontiers

It seems clear that amorphous LLPS is insufficient to describe the general principles underlying the formation and function of the vast repertoire of cellular biomolecular assemblies. In particular, it now seems likely that propagation of order or structure beyond contact sites with the first molecular neighbour is central in assembling PEs. The combination of order and disorder enables spatial heterogeneity, potentially yielding assemblies that feature subcompartments with distinct functions. In addition, the nonlinearities inherent in OCA may allow PEs to transition between distinct functional states with a higher level of control and complexity than can be accessed using solely the material state of a phase-separated assembly. These properties, individually or in combination, are potentially well-matched to the prominent roles of PEs as signal integration and signal processing hubs. Although it is not the focus of this Comment, we note that far-from-equilibrium states, driven for example by nucleotide hydrolysis, will need to be included in a full understanding of PE dynamics and function. A challenge to experimentalists and theorists alike will thus be to understand how the dynamic properties of PEs contribute to their roles in processing and transducing information within larger signalling networks.