How do disordered head domains assist in the assembly of intermediate filaments?

Abstract

The dominant structural feature of intermediate filament (IF) proteins is a centrally located α-helix. These long α-helical segments become paired in a parallel orientation to form coiled-coil dimers. Pairs of dimers further coalesce in an antiparallel orientation to form tetramers. These early stages of intermediate filament assembly can be accomplished solely by the central α-helices. By contrast, the assembly of tetramers into mature intermediate filaments is reliant upon an N-terminal head domain. IF head domains measure roughly 100 amino acids in length and have long been understood to exist in a state of structural disorder. Here, we describe experiments favoring the unexpected idea that head domains self-associate to form transient structural order in the form of labile cross-β interactions. We propose that this weak form of protein structure allows for dynamic regulation of IF assembly and disassembly. We further offer that what we have learned from studies of IF head domains may represent a simple, unifying template for understanding how thousands of other intrinsically disordered proteins help to establish dynamic morphological order within eukaryotic cells.

Introduction

The majority of proteins encoded by all life forms fold into stable, three-dimensional shapes. The varied shapes of proteins are specified by the unique linear sequence of amino acids of each protein. The broad universe of chemistry encoded by protein sequences containing most or all of life’s twenty amino acids correspondingly allows for the unique folding of thousands of proteins having intricately varied shapes.

Paradoxically, between 10 and 20% of the proteins encoded by eukaryotic genomes are composed of sequences that evolution has kept in a chemically impoverished state. Instead of using most or all of our 20 amino acids, these protein domains of low sequence complexity tend to be composed of only a few different residue types. These enigmatic protein segments were first discovered as the activation domains of the Gal4 and VP16 transcription factors [1,2].

Two things have happened in the three decades since the initial discovery of low complexity (LC) domains. First, they have been found in virtually all aspects of cell biology. They constitute the permeability barrier of nucleopores, they are universally associated with RNA-binding proteins and transcription factors, they are attached to numerous integral membrane proteins, and they adorn each end of our 75 intermediate filament proteins.

Second, these protein domains of low sequence complexity spawned an entirely new field. Over the past 30 years, thousands of papers have been published by constituents of the “intrinsically disordered proteins” (IDP) field. The field has made numerous contributions, particularly with respect to computational analyses of IDPs [3,4]. This IDP field effectively solidified the correct belief that these proteins are unable to fold into stable, three-dimensional structures.

There exists but one exception to this dogma. Protein domains of low sequence complexity can self-assemble into amyloid-like aggregates. This unanticipated truth was learned from a paradoxical, non-Mendelian form of inheritance in yeast. In an inspired series of studies, Wickner and colleagues found that a specific phenotype of baker’s yeast cells, designated PSI, was transmitted not by DNA but instead by an aggregated form of the Sup35p protein [5,6]. These aggregates impair the normal function of the Sup35p protein in the termination of translation, thus specifying a penetrant, loss-of-function phenotype. Owing to the unusual stability of Sup35 aggregates, they are passed from mother yeast cells to daughters as if constituting a stable genetic state. It was later shown that many yeast proteins of low sequence complexity can, upon over-expression, form amyloid-like aggregates, leading to the speculation that the evolutionary value of these protein domains might correspond to their ability to form aberrantly large and stable aggregates [7]. Although the Wickner discovery has stood the test of time, it is doubtful that evolution has preserved thousands of LC domains encoded by eukaryotic genomes for the purpose of forming prion-like aggregates that inhibit protein function.

As of 10–15 years ago, the IDP field was bracketed by two pylons. On the one side, LC domains were understood to be incapable of folding into any aspect of protein structure. On the other side, LC domains could be forced to fold into stable aggregates, such as Sup35p prions, that eliminate the biological activity of the encoded protein.

Two papers published from the McKnight lab in 2012 offered a middle ground between these two extremes [8,9]. Incubation of the purified LC domain of the fused-in-sarcoma (FUS) RNA-binding protein at a high concentration led to its phase separation into a hydrogel-like state. Experiments aimed at revealing the content of these gels gave evidence of reversible self-association in the form of labile cross-β polymers. The molecular structure of these cross-β polymers was revealed five years later from solid-state NMR spectroscopy studies [10]. Unlike pathogenic, irreversibly assembled prions, the FUS polymers contained no hydrophobic amino acids at the protomer interface. The absence of “hydrophobic glue” at the protomer interface was proposed to account for the labile and reversible nature of FUS polymers.

From these studies, we hypothesized that the labile self-association of protein domains of low sequence complexity might be of widespread importance to dynamic morphological structures within cells. These observations were picked up by many researchers, thus spawning the trendy “phase separation” field. A particularly useful advance came from the description of an intermediate, liquid-like droplet state that falls between soluble protein monomers and hydrogel polymers. Although it is agreed that some form of protein self-association by purified LC domains must take place to facilitate droplet formation, there is a sharp divide as to the chemical basis for protein:protein interaction. On the one hand, the McKnight laboratory has favored labile cross-β interactions poised at the threshold of thermodynamic equilibrium to define the basis for LC domain self-association [11]. The more prevalent view holds that LC domains establish no form of protein structure when phase-separated into liquid-like droplets [12–18]. The pros and cons of these divergent views have been reviewed elsewhere [19]. As will be articulated herein, studies of intermediate filament proteins have been instrumental in resolving this structure/no structure controversy.

Hitting the jackpot with intermediate filaments

The pharmacological “Rosetta Stone” of the phase separation field is defined by a pair of aliphatic alcohols, 1,6-hexanediol (1,6-HD), and 2,5-hexanediol (2,5-HD). Despite sharing the identical chemical formula, these regioisomeric chemicals have profoundly different effects on living cells. 1,6-HD dissolves the permeability barrier of nuclear pores [20,21], along with a wide variety of nuclear and cytoplasmic puncta not surrounded by investing membranes [22]. The 2,5-HD regioisomer is far less capable of disassembling these various cellular structures.

A nuclear example of these 1,6-HD sensitive intracellular structures was described by Ramon Y Cajal more than a century ago [23]. These Cajal bodies, as well as other morphologically dynamic nuclear and cytoplasmic puncta, are densely packed with RNA and protein. Forward genetic studies investigating germ line specification in flies and worms have revealed the essential molecular components and extensively demystified these RNA-enriched granules [24–26].

As we investigated phase-separated hydrogels and liquid-like droplets, one of our most encouraging observations was that both hydrogels and liquid-like droplets formed from phase-separated LC domains are themselves dissolved by 1,6-HD far more readily than 2,5-HD [22]. It was this observation, perhaps more than any other, that bolstered our belief that test tubes studies on LC domain phase separation might be biologically valid.

Just as we were preparing to publish these observations, we performed what we considered to be no more than an experimental control. Briefly, we asked whether aliphatic alcohols might dissolve cellular structures indiscriminately. After brief exposure to either of the two aliphatic alcohols, we fixed the treated cells and stained them with antibodies to tubulin, actin, and vimentin. Much to our relief, we observed no changes in the organization of either actin filaments or microtubules. By contrast, the active regioisomer (1,6-HD) quickly disassembled vimentin intermediate filaments, which were left intact by the inactive chemical (2,5-HD).

Figure 1a shows images of cells expressing a Green Fluorescent Protein (GFP):vimentin fusion protein in response to the administration of each of the two aliphatic alcohols. The GFP-tagged vimentin IFs are dissolved by 1,6-HD but not by 2,5-HD. Precisely, the same observations were quickly made with vimentin IFs assembled from purified recombinant protein. As viewed by transmission electron microscopy, vimentin IFs are rapidly disassembled upon exposure to 1,6-HD, but not 2,5-HD (Figure 1b). The videos of Supplemental Data Figures S1 and S2 show live cell images of fibroblasts expressing GFP:vimentin following exposure to either 1,6-HD or 2,5-HD [19]. Quite clearly, vimentin intermediate filaments are different from both actin filaments and microtubules with respect to sensitivity to aliphatic alcohols.

Figure 1. 1,6-hexanediol dissembles vimentin intermediate filaments both in vivo and in vitro.

Panel A shows cultured mammalian cells expressing a GFP:vimentin fusion protein [54]. Left panels show cells before exposure to aliphatic alcohols. Right panels show cells 5′ after exposure to 1,6-HD (upper image) or 2,5-HD (lower image). Panel B shows electron micrographs of vimentin intermediate filaments assembled from recombinant protein at different time intervals post-exposure to either 1,6-HD (upper images) or 2,5-HD (lower images). Scale bar = 1 μm.

At the time, we had failed to appreciate that the central coiled-coil α-helices of IF proteins are universally flanked by head and tail domains of low sequence complexity. We quickly cloned the head and tail domains of five different IF proteins and tested each sample for phase separation (vimentin, peripherin, neurofilament light, neurofilament medium, and neurofilament heavy) [22]. In all five cases, the head domains were observed to phase separate. By contrast, none of the five tail domains exhibited this property.

Turning to the biochemical assays perfected by the IF field over the past three decades, we expressed three versions of all five proteins. One version corresponded to the intact IF protein, one to the head-deleted variant, and one to the tail-deleted variant. Upon incubation under conditions permissive of IF assembly, all five of the intact proteins assembled into mature filaments, none of the head-deleted variants assembled into filaments, and all of the tail-deleted variants assembled in a manner indistinguishable from the intact, parental proteins. This was a jackpot discovery for us. For the first time, we could study a class of proteins blessed with defined properties of assembly, wherein we could focus on the biological function of phase separation-competent LC domains.

IF head domains function by adopting labile cross-β structures

When incubated at high concentration in isolation, the head domains of all five IF proteins that we have studied phase separate. Three observations confirm that the head domains phase separate via the formation of cross-β structures. First, when examined by electron microscopy, hydrogels formed from all five head domains reveal morphologically uniform polymers. Second, when hydrogels are lyophilized and examined by X-ray diffraction, all five samples revealed prominent diffraction rings at 4.6–4.7Å. This is precisely the distance that separates protomers in cross-β polymers. Third, when evaluated by solid-state NMR spectroscopy, hydrogel polymers yield spectra diagnostic of cross-β structure.

The latter of these observations provided a unique opportunity to put the labile cross-β concept of LC domain function to test. By use of intein-mediated protein ligation, we prepared segmentally labeled forms of the neurofilament light (NFL) chain and desmin IF proteins. ¹³C/¹⁵N-labeled head domains were expressed in bacterial cells, purified, and ligated to unlabeled α-helical rod and tail domains. We then assembled the segmentally labeled NFL and desmin proteins into mature IFs, concentrated the samples by centrifugation, and collected NMR spectra. One of three outcomes was anticipated. First, the spectra of the assembled, segmentally labeled IFs might match that of the isolated head domains present in labile, cross-β polymers. Second, the spectra of the assembled IFs might differ from labile, cross-β polymers, reflective of a different structural state. Third, the assembled, segmentally labeled IFs might yield spectra indicative of no form of structural order. Either of the latter two outcomes would invalidate our idea that IF head domains abet filament assembly via the formation of transient structural order.

Figure 2 compares the solid-state NMR spectra of NFL and desmin head domain-only polymers with spectra derived from assembled, segmentally labeled IFs. All discernible spectral signals can be seen to match for both the labile, cross-β polymers that coalesce upon the incubation of the isolated NFL and desmin head domains (left panels) as compared for spectra observed for IFs assembled from full-length, segmentally labeled proteins (middle panels). These matching spectra do not constitute formal proof that IF assembly is driven by the ability of head domains to form labile cross-β interactions. By contrast, had these experiments yielded evidence of different spectra in the two states, or no spectral evidence of structural ordered in the assembled IFs, our concept would have been experimentally invalidated. Having avoided two distinct opportunities for invalidation, the labile cross-β concept for IF head domain function survived this wave of experimental investigation [27].

Figure 2. Solid-state NMR spectra of NFL and desmin head domains.

Panel A shows solid-state NMR spectra of the NFL head domain as determined for labile cross-β polymers formed from the isolated head domain (left as shown in blue) or from fully assembled NFL IFs composed of segmentally labeled protein (middle as shown in red). Right image shows an overlay of the two spectra. Panel B shows solid-state NMR spectra of the desmin head domain as determined for labile cross-β polymers formed from the isolated head domain (left as shown in blue) or from fully assembled desmin IFs composed of segmentally labeled protein (middle as shown in red). Right image shows an overlay of the two spectra.

As we were in the process of composing this review, a beautiful cryo-EM tomography study of fully assembled vimentin intermediate filaments appeared on the BioRXive server [28]. This study, authored by Ohad Medalia and his colleagues at the Department of Biochemistry of the University of Zurich, revealed assembled vimentin intermediate filaments to be composed of five cylindrically organized bundles of α-helices, each containing eight vimentin polypeptides. These five bundles are held in place via the protrusion of vimentin head domains into a central, amyloid-like filament occupying the lumen of the filament (Figure 3). It is hard to imagine a more satisfying accounting as to how the cross-β forming head domains of IF proteins coalesce to mediate filament assembly.

Figure 3. Cryo-EM tomographic image of vimentin intermediate filaments.

Panel A shows a longitudinal view of a three-dimensional atomic model of assembled vimentin intermediate filaments. Panel B shows a crosssection of an assembled filament revealing five bundles of α-helices surrounding a central lumen. Each α-helical bundle contains eight vimentin polypeptides. The central lumen contains vimentin head domains assembled in a labile cross-β conformation. Images of Panels A and B are copied from the BioRXive preprint describing cryo-EM tomographic studies conducted by Ohad Medalia and colleagues of the University of Zurich [28].

Respecting our elders

As biochemists, we ply our trade by striving to recapitulate narrow aspects of cell biology in test tube reactions. In order to have confidence that our findings are of biological relevance, we must turn to the field of genetics. In other words, biochemists respect geneticists as our elders. As we consider whether we are moving in the right direction in hypothesizing that IF head domains work by forming labile cross-β interactions, can we learn anything from unbiased genetic studies?

Human genetic studies of patients suffering from Charcot-Marie-Tooth (CMT) disease have revealed mutations within the head domain of NFL [29–31]. These mutations are of an idiosyncratic nature in that they recurrently change one of only two residues within the NFL head domain, proline residue 8 (P8) or proline residue 22 (P22). Mendelian inheritance of CMT has been traced in different families that suffer change of P8 to leucine, arginine, or glutamine, or change of P22 to serine, threonine, or arginine. Biochemical assays of filament assembly have shown that all six of these CMT variants of NFL specify proteins that are incapable of forming intermediate filaments. Why is it that genetic alterations of P8 or P22 to any of a number of different amino acid residues impede filament assembly and manifest as CMT disease?

Studies focused on the isolated head domain of NFL have shown that all P8 and P22 mutational alterations significantly enhance the propensity of the head domain to form cross-β polymers [27]. We interpret these observations to explain the dominant nature of these CMT-causing mutations. The P8 and P22 variants appear to constitute gain-of-function mutations that need to occur on only one allele of our two NFL genes. We further hypothesize that these data give evidence that evolution has crafted the NFL head domain to self-interact weakly.

The IF assembly abetting function of the NFL head domain is interpreted to work only if self-association is properly tuned. If the avidity of head domain self-association is aberrantly strengthened via P8 or P22 mutations, IF assembly is reciprocally impeded.

What is so special about proline at the eighth and 20 s positions of the NFL polypeptide? Unlike the other nineteen amino acids that overwhelmingly prefer the trans-conformational state of the adjacent peptide bond, proline allows for cis occupancy. This difference is particularly important during the process of protein folding. It is likewise the case that proline is also the only amino acid that does not allow for hydrogen bonding by the polypeptide backbone because it lacks the NH group shared by the other nineteen amino acids.

To test whether either of these features of proline might explain the unique importance of P8 and P22 for function of the NFL head domain, we used native chemical ligation to stitch N-terminally located synthetic peptides onto the remainder of the NFL protein. These methods allowed for the introduction of non-natural amino acids into the full-length NFL protein. Replacement of proline residue 8 of the NFL head domain with its dimethyl variant, which is known to impair cis–trans isomerization [32], failed to impair the assembly of mature intermediate filaments [33]. We thus proceeded to consider the possibility that the deficit in IF assembly suffered by the P8 and P22 mutational variants might be attributed to the capacity of the variant amino acid residues to add one additional backbone hydrogen bond to the cross-β structure that, under normal circumstances, allows for labile self-association.

Synthetic peptides bearing any of the three mutational variants of P8 (P8L, P8R, or P8Q), or the three variants of P22 (P22S, P22T, or P22R), were ligated onto the remainder of the NFL polypeptide. In all six cases, these semi-synthetic variants of NFL were incapable of assembling into mature intermediate filaments. If, however, the peptide nitrogen atom associated with the variant amino acid residue was capped with a methyl group – thus preventing backbone hydrogen bonding – IF assembly was fully restored [33]. Figure 4 shows electron micrographs of IFs prepared from the native NFL protein, the tangled mess observed upon attempts at filament assembly by the P8L variant, and fully repaired filaments assembled from the methyl-capped P8L variant.

Figure 4. NFL intermediate filaments formed from native protein, a variant changing proline residue 8 to leucine, and a methyl-capped version of the P8L variant.

Left image shows an electron micrograph of IFs formed from the native NFL protein. Middle image shows tangled protein aggregates formed upon the incubation of a genetic variant of NFL bearing a proline-to-leucine substitution of residue 8. Right image shows normal NFL IFs formed from a version of the P8L variant bearing a methyl group on the peptide nitrogen atom of the variant leucine residue.

Our interpretation of these experiments is simple. The native head domain of NFL has evolved to be capable of forming labile and transient cross-β interactions. These interactions are poised at the threshold of thermodynamic equilibrium and are assembled principally via a hydrogen bond network formed by the polypeptide backbone as taught to us by Linus Pauling over 70 years ago [34]. We offer that the avidity of these weak and labile cross-β interactions is enhanced substantially if either P8 or P22 of the NFL head domain is replaced by any of the other nineteen amino acids. We further hypothesize that the basis for the enhanced avidity of head domain self-association by the P8 and P22 variants can be attributed to the addition of a single extra hydrogen bond specified by the amino acid that replaces proline, thus enhancing the lattice of backbone hydrogen bonds that facilitate intermolecular self-association.

Related studies on a disease-causing mutation in the head domain of the keratin-8 IF protein have recently been published [35]. A missense mutation changing glycine residue 62 to cysteine in the keratin-8 head domain is known to form aggregated amyloid bodies believed to be causative of alcoholic steatohepatitis of the liver. Focused studies comparing the native head domain of keratin-8 with the G62C variant have given clear evidence of mutation-induced enhancement of head domain self-association. Thus, dominant gain-of-function mutations in both the NFL and keratin-8 intermediate filament proteins have been traced to aberrantly enhanced head domain self-association.

Towards an understanding of filament regulation

It has long been recognized that intermediate filaments disassemble during mitosis [36,37]. One mechanism that may account for filament disassembly is head domain phosphorylation, which is known to take place during cytokinesis [38–43].

Having observed that protein kinase-A (PKA) phosphorylation causes the disassembly of both NFL and desmin intermediate filaments as assayed in vitro, we wondered whether this might result from the inhibition of head domain self-association. Phase-separated hydrogels were first formed from mCherry fusions to either the desmin or NFL head domain. GFP fusion proteins were then made to the same head domain fragments and tested for binding to the two mCherry hydrogels. The GFP:NFL fusion protein was observed to bind hydrogels formed from the mCherry:NFL fusion protein, but not to gels formed from mCherry:desmin. It was likewise observed that the GFP:desmin fusion protein was bound by hydrogels formed from the mCherry:desmin fusion protein, but not mCherry:NFL. These observations confirm the specificity of head domain self-interaction.

In order to test the effects of PKA-mediated phosphorylation, we first bound the GFP:desmin fusion protein to its cognate mCherry hydrogel, as well as the GFP:NFL fusion to its cognate hydrogel. Samples were then incubated with ATP alone, PKA alone, or a combination of ATP and PKA. GFP signal was monitored by confocal microscopy, yielding evidence in both cases of protein release only from hydrogel samples exposed to both ATP and PKA (Figure 5). We simplistically offer that head domain phosphorylation causes some form of inhibition of head domain self-association, perhaps reflecting ionic repulsion from negatively charged phosphate groups.

Figure 5. Phosphorylation-mediated reversal of NFL and desmin head domain self-association.

Phase-separated hydrogels were formed from mCherry fusions to the NFL head domain (left images) or desmin head domain (right images). Hydrogels were bound by GFP fusions to NFL and desmin head domains prior to incubation with ATP alone (top images), protein kinase A alone (middle images), or both ATP and protein kinase A (lower images). GFP fusion proteins were released from both NFL and desmin hydrogels only upon co-incubation with both ATP and protein kinase A.

Landing pads for RNA granules

Three decades ago Mary Lou King published evidence that the germ granules of frog eggs associate tightly with a vimentin-rich cytoskeletal structure that could be isolated from oocytes via biochemical fractionation [44]. Parallel findings have been reported from studies of germ granules in fly eggs. An atypical IF protein designated TM1-I/C co-localizes with the germinal polar granules located at the posterior termini of Drosophila melanogaster oocytes [45]. Intriguingly, studies of the TM1-I/C IF protein evolved from unbiased forward genetic studies in flies showing that this protein is functionally required for polar granule formation and establishment of the germ cell lineage [46]. Knowing that localized deposition of germ granules in frog and fly eggs is of vital importance for appropriately positioned specification of the eventual germ lineage, these independent, orthogonal studies give evidence that RNA granules might associate with IFs in a causal manner.

How might RNA granules come to bind intermediate filaments? Assembled IFs are collared in a repetitively periodic manner by coincidentally localized head and tail domains. For conventional IFs that contain long, centrally located coiled-coil domains, these collars are displayed at periodic intervals separated by approximately 47 nm [47,48]. Wondering whether these repetitively periodic collars might constitute Velcro-like landing pads for interaction with protein domains of low sequence complexity derived from RNA granules, we incubated assembled vimentin IFs with a fusion protein linking GFP to the LC domain of the FUS RNA-binding protein. As shown in Figure 6, the GFP:FUS protein forms spherical puncta along the axial length of vimentin IFs.

Figure 6. Periodic binding of a chimeric GFP:FUS protein to vimentin intermediate filaments.

Panel A shows electron micrographs of vimentin intermediate filaments formed from recombinant protein in either the absence (left) or presence (right) of a fusion protein linking GFP to the low complexity domain of the fused-in-sarcoma (FUS) RNA-binding protein (scale bar = 500 nm). Panel B shows higher magnification view of vimentin IFs exposed to GFP:FUS protein (scale bar = 50 nm). Panel C shows histogram of puncta-to-puncta distance along axial length of vimentin intermediate filaments incubated with GFP:FUS protein.

Having observed repetitive puncta separated by roughly 47 nm along vimentin IFs, we hypothesize that the GFP:FUS protein may interact with repetitively spaced head or tail domain collars. Since vimentin IFs cannot be assembled in the absence of the head domain, we were unable to test for head domain dependency of GFP:FUS binding. By contrast, it was possible to assemble vimentin IFs lacking the disordered tail domain, leading to the observation that tail-less vimentin intermediate filaments are unable to bind GFP:FUS [22].

The experiments as shown in Figure 6 are obviously contrived. We have no reason to believe that the FUS protein – even as a known constituent of RNA granules [49–51] – has been evolutionarily crafted to bind vimentin intermediate filaments. The pattern of filament binding revealed in Figure 6 does, however, hint to a biochemical mechanism by which RNA granules might adhere to intermediate filaments. Furthermore, we are encouraged by the beautiful work recently published from the laboratory of Daniel Kaganovich showing an intimate interaction between vimentin intermediate filaments and RNA granules in differentiating stem cells [52].

The studies described herein clearly distinguish functional differences between the head and tail domains of IF proteins. Although both head and tail domains are of low sequence complexity, we have never observed IF tail domains to be capable of self-association and phase separation. By contrast, every single head domain we have studied to date is readily capable of forming hydrogels composed of labile cross-β polymers [22]. We have likewise observed that five different IF proteins are incapable of assembling into mature filaments if the head domain is removed. Reciprocally, all five of the tail-deleted IF proteins we have studied achieved filament assembly in a manner indistinguishable from the intact proteins. As yet, we have no means of computational analysis capable of revealing the underlying sequences of an LC domain responsible for its ability to self-associate and phase separate. An important goal of future research on protein domains of low sequence complexity will be the generation of AI-enabled computational methods predictive of the compact cross-β cores responsible for facilitating biologically relevant self-association.

Conclusions

The reductionist experiments described herein offer a mechanistic concept as to how the head domains of IF proteins might assist in filament assembly. If correct, the interpretations outlined herein offer a reference point useful for considering the function of LC domains in many other aspects of cell biology. We have published a number of studies on other LC domains with particular focus on RNA-binding proteins. In all such cases, we have discovered LC domain self-association to be mediated by labile cross-β structures, yet none of our previous studies have included functional data comparable with what we have learned from studies of IF head domains. If we have been led in the correct direction by our studies on IF head domains, our progress rests squarely on the shoulders of rigorous biochemical work carried out over the past four decades by the IF field.

We close with consideration of anticipated challenges that, if effectively negotiated, might lead to significant advancement in this field of study. First, if we are able to effectively map the locations and sequences of hundreds of cross-β cores within LC domains, it may be possible to develop machine-learning-based methods useful for the in silico identification of the cross-β cores present within our thousands of protein domains of low sequence complexity. Second, when a given cross-β core becomes sell-associated, it transiently generates a unique molecular structure. These transient protein structures, we propose, will allow for heterotypic interaction with other proteins. An example of this type of heterotypic interaction can be found in the binding of the repetitive C-terminal domain of RNA polymerase II to cross-β polymers formed from the LC domain of both the FUS and TAF-15 RNA-binding proteins [53]. When viewed from this perspective, it becomes clear that we are only at the earliest stage of our understanding of this new form of dynamic and transient biological order within eukaryotic cells.

Data availability

Data will be made available on request.