Redox-mediated regulation of low complexity domain self-association

Abstract

Eukaryotic cells express thousands of protein domains long believed to function in the absence of molecular order. These intrinsically disordered protein (IDP) domains are typified by gibberish-like repeats of only a limited number of amino acids that we refer to as domains of low sequence complexity. A decade ago, it was observed that these low complexity (LC) domains can undergo phase transition out of aqueous solution to form either liquid-like droplets or hydrogels. The self-associative interactions responsible for phase transition involve the formation of specific cross-β structures that are unusual in being labile to dissociation. Here we give evidence that the LC domains of two RNA binding proteins, ataxin-2 and TDP43, form cross-β interactions that specify biologically relevant redox sensors.

Introduction

This brief review is focused on an unusual class of proteins composed of only a limited subset of the twenty amino acids typically used for proteins to adopt a stable, three-dimensional fold. We designate these proteins as being of “low sequence complexity”, and will refer to protein domains having these unusual features as low complexity (LC) domains. LC domains were first discovered three decades ago in studies of transcriptional activation domains. Activation domains were found as parts of gene-specific transcription factors that, in most cases, were further endowed with prototypic DNA binding domains in the forms of zinc fingers, homeobox domains, leucine zippers and helix-turn-helix domains [1].

The activation domains of transcription factors are unusual in two ways. First, their sequences are composed of gibberish-like repeats of only a few types of amino acids. The activation domain of the SP1 transcription factor discovered by Robert Tjian was composed almost exclusively of glutamine residues [2]. Mark Ptashne’s Gal4 transcriptional activation domain was enriched in acidic amino acids [3,4], as was the herpes simplex virus VP16 protein that my students and I studied back in the 1980’s [5,6]. Studies of these early examples of transcriptional activation domains led to the conclusion that they function in the absence of structural order, a conclusion that currently persists [7,8].

In the intervening decades since the discovery of transcriptional activation domains scientists have discovered LC domains associated with thousands of other cellular proteins. An entire field of research has evolved to study these intrinsically disordered proteins (IDPs), and it is commonly appreciated that these enigmatic proteins constitute upwards of 20% of the proteomes of eukaryotic cells [9].

In 2012 Masato Kato, Tina Han, Shanhai Xie, Xinlin Du and I reported the surprising observation that certain low complexity domains can become phase separated out of aqueous solution [10]. Our studies were initially focused on the LC domain of an RNA-binding protein designated fused in sarcoma (FUS). The domain of interest corresponded to the amino terminal 214 residues of the protein. This region of the FUS protein had long been known to represent a transcriptional activation domain in the context of fusion proteins created by genetic translocation events causative of cancer [11–13]. These translocations affixed the LC domain of FUS, or its two paralogous relatives – EWS and TAF15, onto the DNA-binding domains of any of a number of gene-specific transcription factors, yielding hybrid proteins that were themselves proven to drive significant forms of cancer [14–18].

In the context of these hybrid proteins, the LC domains of FUS, EWS and TAF15 function as potent transcriptional activation domains [14–18]. These observations offered incontrovertible evidence that the LC domain of FUS is capable of significant biological function on its own. In other words, the first 214 residues of FUS can assume the role of activating transcription in the complete absence of the remainder of the FUS RNA binding protein.

Both the intact FUS protein and derivatives corresponding to only its N-terminal 214 amino acids are able to become phase separated. These observations offered the generation of a simple hypothesis. Might it be possible that the molecular forces causative of phase transition are one in the same as those specifying biologic function of the FUS LC domain? If so, the phenomenon of phase separation might allow for reductionist studies of this class of LC domains. Having co-discovered transcriptional activation domains more than three decades ago, yet never having been able to understand how they function in a mechanistic sense, my trainees and I were excited to consider the possibility of approaching LC domain function via the simple assay of phase separation.

What have we learned by use of this controversial, reductionist approach? First, the LC domain of FUS adopts labile, transient molecular structure as the underlying basis for phase separation [10]. Second, the stability of this structure can be regulated by post-translational modification in the form of phosphorylation [19]. And third, the labile, transient structure adopted by the FUS LC domain binds the C-terminal domain of RNA polymerase II as a conceptual basis for its biological function as a transcriptional activation domain [20].

A smoldering controversy:

The aforementioned studies and interpretations have drawn controversy in several forms. First, it has been claimed that the labile cross-β structures formed by LC domains in hydrogels are not present in liquid-like droplets and have no relevance to biology [21]. This belief has come to rest largely upon solution NMR experiments [22–24]. Solution NMR methods conducted at low, non-physiological pH deliver optimal spectra if deployed for the study of relatively small proteins or protein complexes that tumble readily in solution. If a monomeric protein assembles into larger, polymeric complexes, its NMR spectrum broadens and diminishes in intensity. As such, NMR signals derived from, labile, cross-β assemblies would naturally be obscured by the sharp and intense spectra coming from unstructured LC domain monomers. Whatever the case, solution NMR methods have yet to reveal the basis for LC domain self-association that must occur in order for a protein sample to become phase separated from aqueous solution.

If one cannot observe labile cross-β interactions in liquid-like droplets by solution NMR spectroscopy, are there other ways to probe for their presence or absence? To this end the McKnight lab developed two different methods of molecular “footprinting” that diagnose the self-associated, polymeric state of LC domains. One footprinting method employs N-acetylimidazole (NAI) to alkylate accessible serine, threonine, asparagine, lysine, arginine and tyrosine side chains [25,26]; the other employs hydrogen peroxide (H₂O₂) to oxidize accessible methionine residues to methionine sulfoxide [27]. These footprinting methods are conceptually simple and reminiscent of hydrogen:deuterium exchange methods used to distinguish interior versus surface-exposed regions of folded proteins.

In the denatured state, all residues of a disordered LC domain should be accessible to chemical modification by NAI or H₂O₂. By contrast, in the partially ordered, polymeric state, certain residues buried within the structure might be protected from modification. By comparing the patterns of NAI or H₂O₂ modification of a test protein in the denatured state relative to the polymeric state, a molecular footprint can be established. In the case of NAI, this footprint was characterized for polymeric hydrogel samples of the hnRNPA2 LC domain. For H₂O₂ footprinting, the method of which we describe in greater details later in this review, we characterized a polymeric sample of the TDP43 LC domain. After having established the two ground state footprints, we asked whether the same footprints could be seen in liquid-like droplet samples of the hnRNPA2 and TDP43 LC domains. As reported for the NAI footprint of hnRNPA2 [28], and the H₂O₂ footprint of TDP43 [29], liquid-like droplets displayed the same footprints as hydrogel polymers. These experiments directly reveal the presence of cross-β self-association of hnRNPA2 and TDP43 in liquid-like droplets.

A second, less-direct manner for probing liquid-like droplets for the relevance of cross-β interactions came from correlative mutagenesis experiments. The 25 tyrosine and phenylalanine residues located within the LC domain of hnRNPA2 were individually changed to serine. Each variant was then tested for incorporation into either hydrogels or liquid-like droplets. Certain variants did not interfere with incorporation, certain variants mildly impeded incorporation, and certain variants strongly blocked incorporation into either hydrogels or liquid-like droplets. The patterns of effect of the 25 variants were virtually indistinguishable upon tests of incorporation into either form of the phase separated hnRNPA2 protein [28]. These findings neither confirm nor refute the presence of cross-β interactions of the hnRNPA2 LC domain in liquid-like droplets. They do, however, demonstrate that an indistinguishable distribution of amino acid side chains directs hnRNPA2 to self-associate in a manner allowing its incorporation into either of the two phase separated states.

Perhaps the most important question at the heart of the controversy between the “function sans structure” advocates and those few of us who believe that LC domains function by adopting labile and transient structural order is whether LC domains adopt a structured state in living cells. To our knowledge, there are only two examples wherein this question has been experimentally addressed. Both such examples made use of the same footprinting methods that were deployed to search for the presence or absence of cross-β structure in liquid-like droplets. NAI probing of cellular hnRNPA2 gave clear evidence of the same footprint observed in both hydrogel polymers and liquid-like droplets [28]. Likewise, H₂O₂ probing of cellular TDP43 gave clear evidence of the same footprint observed in both hydrogel polymers and liquid-like droplets [29]. Importantly, neither of these studies involved the use of over-expressed hnRNPA2 or TDP43, but instead relied solely upon analysis of endogenous protein expressed at the normal cellular level.

We extend consideration of these controversial differences of opinion with what may represent an oft-overlooked point. The McKnight perspective by no means contends that all LC domains function via adopting transient structural order. We understand no more than what we have reported from studies of the LC domains associated with the FUS RNA binding protein, several hnRNP proteins, TDP43, yeast ataxin-2, and seven intermediate filament (IF) proteins (neurofilament light, neurofilament medium, neurofilament heavy, desmin, vimentin, peripherin and the TM1-I/C intermediate filament of fruit flies).

All intermediate filament proteins contain LC domains located at the N- and C-termini flanking their central coiled-coil domains. For the six vertebrate IF proteins, we observe that each of the N-terminal “head domains” is endowed with the ability to form labile, cross-β polymers. By contrast, none of the C-terminal “tail domains” of these six IF proteins is able to self-associate in this way. Despite being of low sequence complexity, we have observed no experimental evidence for cross-β polymerization by the tail domains of vertebrate IF proteins.

Perhaps coincidentally, or perhaps importantly, the head domains of all six IF proteins are critically required for IF assembly – yet none of the tail domains are functionally required for assembly [30]. This pattern of functional dependency is reversed for the TM1-I/C IF protein of Drosophila melanogaster. Proper IF assembly of TM1-I/C is dependent upon its LC tail domain, but not its head domain [31]. Moreover, the TM1-I/C head domain is unable to self-assemble into the labile, cross-β structural conformation, but the tail domain readily does so. Thus, in seven out of seven cases, the IF head or tail domain required for filament assembly is cross-β competent. Likewise, in seven out of seven cases, the IF head or tail domain that is dispensable for filament assembly is cross-β incompetent. These experiments confirm that only a fraction of LC domains we have studied over the past decade act by self-assembling into labile, cross-β structures.

We conclude this discussion dealing with our controversial differences of opinion relative to most people now working in the popular “phase transition” field with recent studies of the fly TM1-I/C intermediate filament protein. As mentioned earlier, TM1-I/C is critically dependent upon its LC tail domain for filament assembly. We have worked collaboratively with Dylan Murray, a solid state NMR spectroscopist, to obtain spectra of the TM1-I/C tail domain as assembled into hydrogel polymers. A combination of X-ray diffraction and transmission electron microscopy had already confirmed that the TM1-I/C tail domain polymers were of a labile, cross-β nature [31]. Fortuitously, the ¹³C/¹⁵N NMR spectrum of TM1-I/C tail domain-only polymers exhibits unusually sharp and well-dispersed peaks. We proceeded with intein chemistry to segmentally label the tail domain within otherwise intact TM1-I/C protein, assembled intermediate filaments, compacted them by centrifugation, and visualized the ¹³C/¹⁵N spectrum of the tail domain in the context of its native biological assembly by solid state NMR spectroscopy. Much to our delight, we observed virtually identical spectra in the tail domain only polymers and assembled TM1-I/C intermediate filaments [32]. To our knowledge, this represents the sole example wherein the structure of an LC domain has been directly visualized in its native biological setting. Whether this pattern of structural concordance holds for the many other LC domains we and others are studying remains to be determined.

It is for the biomedical research community outside of the “phase transition” field to judge the merits of our structure-based concept of LC domain function versus those who dispute our findings. Despite the unease we feel according to the current state of controversy, we can be confident that a combination of time and further study will ensure that scientific truth prevails.

Studies of the methionine-rich LC domain of yeast ataxin-2:

Some of our more recent work focused on the structure and function of LC domains has been conducted collaboratively with Ben Tu and his trainees here at UT Southwestern Medical Center. Ben discovered that the yeast ortholog of ataxin-2, also called poly-A binding protein binding protein (Pbp1), is required to couple autophagy to the metabolic state of mitochondria. Cells grown in glucose-rich culture medium do not engage in autophagy. When grown in lactate, yeast cells mobilize mitochondrial respiration, thus facilitating a switch from glycolytic growth to respiratory growth. This conversion of growth state entails activation of autophagy in the absence of supplemented amino acids.

By use of a forward genetic screen in search for proteins required for autophagy under such respiratory conditions, Ben discovered and confirmed reliance upon the yeast ortholog of ataxin-2 [33]. Reductionist analysis of the parts of the yeast ataxin-2 protein required for its function highlighted the importance of a carboxyl terminal region of 150 amino acids. Variants of the protein lacking this region failed to complement function of the intact ataxin-2 protein.

Recognizing that this functionally essential, C-terminal domain of the yeast ataxin-2 protein was of low sequence complexity, Ben initiated collaborative experiments with Masato Kato to determine whether it might function in a manner analogous to LC domains familiar to the McKnight laboratory. These collaborative efforts quickly revealed that the LC domain of the yeast ataxin-2 protein can self-associate in a manner leading to its phase separation out of aqueous solution [34]. Incubation of the purified LC domain at high protein concentration under physiological monovalent salt concentration and neutral pH quickly yielded liquid-like droplets that, with time, matured to a gel-like state. A combination of electron microscopy and X-ray diffraction gave evidence that the gels were composed of uniform, amyloid-like polymers. The ataxin-2 polymers were labile to disassembly as assayed by semi-denaturing agarose gel electrophoresis [34].

Unlike the LC domains of FUS and many other RNA binding proteins studied previously, that of the yeast ataxin-2 protein is methionine-rich. The 150-residue LC domain of yeast ataxin-2 contains 24 methionine residues, qualifying it as one of the three most methionine-rich proteins encoded by the 5. cerevisiae genome. Wondering whether these methionine residues might endow the yeast ataxin-2 protein with the capacity to sense cellular redox state, we exposed liquid-like droplets of the protein to hydrogen peroxide (H₂O₂). Liquid-like droplets of the ataxin-2 LC domain melted upon exposure to an EC₅₀ concentration of 0.33% H₂O₂. Even a ten-fold higher concentration of H₂O₂ failed to melt liquid-like droplets formed from the LC domains of either FUS or hnRNPA2.

SDS-PAGE analysis revealed oxidation-mediated retardation of the electrophoretic mobility of the yeast ataxin-2 protein. This slowed electrophoretic migration was reversed upon exposure of the oxidized protein to recombinant methionine sulfoxide reductase enzymes specific to the two stero-isomeric forms of oxidized methionine. As shown in Figure 1, enzymatic reduction also revived formation of liquid-like droplets.

Figure 1. Melting of ataxin-2 and TDP43 liquid-like droplets by hydrogen peroxide and reformation via enzymatic reduction of methionine sulfoxide adducts.

Purified samples of the isolated ataxin-2 and TDP43 low complexity domains form phase separated liquid-like droplets upon incubation under physiological conditions of monovalent salt and neutral pH (top panels). Exposure to 0.1% hydrogen peroxide rapidly melts both ataxin-2 and TDP43 liquid-like droplets (middle panels). Upon exposure to two methionine sulfoxide reductase enzymes, thioredoxin, thioredoxin reductase and NADPH, the oxidized proteins become reduced, allowing for re-formation of liquid-like droplets (lower panels). A scale bar indicate 25 μm. Images were reproduced from [34] and [29].

As studied in living yeast cells, the ataxin-2 protein was observed to form amorphous, cloud-like structures surrounding mitochondria. Correlative mutagenesis experiments gave evidence that the ability of the ataxin-2 protein to form its cloud-like morphology in cells was concordant with the ability of its LC domain to self-assemble. Moreover, the solubilizing effects of methionine oxidation observed in test tube experiments were found to correlate with the effects of H₂O₂ applied to living yeast cells. For example, in its assembled, self-associated state, ataxin-2 was observed to bind Kog1– the yeast ortholog of the mammalian RAPTOR protein. This interaction is required for the induction of autophagy, and only takes place in respiratory cells wherein the LC domain of yeast ataxin-2 is in the self-assembled state. Oxidation of methionine residues within the LC domain of ataxin-2 prevents self-assembly and thereby blocks autophagy. Replacement of ataxin-2 methionine residues with either tyrosine or phenylalanine residues yields a protein that fully retains the ability to self-associate and undergo phase separation. As compared with the parental, methionine-rich LC domain, variants bearing changes to either tyrosine or phenylalanine are oxidation-resistant, immune to H₂O₂-mediated melting, and increasingly autophagic in living yeast cells.

Studies of the methionine-rich LC domain of TDP43:

TAR binding protein 43 (TDP43) is an RNA binding protein that is frequently found in an aggregated state in the brain tissue of patients suffering from neurodegenerative disease [35]. Human genetic studies have reported scores of missense mutations causative of neurodegenerative disease that map to a C-terminal region of the TDP43 protein [36]. The C-terminal domain of TDP43 is of low sequence complexity and, like the yeast ataxin-2 protein, methionine rich. Purified samples of the TDP43 protein self-associate in a manner leading to phase separation, and we have recently described experiments showing that phase separation can be regulated by methionine oxidation [29].

Observations coming from studies of the methionine-rich LC domain of TDP43 largely comport with studies of the methionine-rich LC domain of the yeast ataxin-2 protein. Phase separated liquid-like droplets formed from both LC domains are melted by the same concentration of H₂O₂ (0.1-0.3%), and melted droplets can be similarly revived upon methionine reduction by methionine sulfoxide reductase enzymes [29,34]. Full side chain reduction in both cases requires thioredoxin and thioredoxin reductase, with the terminal conversion of NADPH to oxidized NADP.

Studies of self-associated TDP43 included a method of H₂O₂ footprinting. These footprinting experiments were initiated using hydrogel samples composed of uniform TDP43 polymers. The assembled polymers were first exposed to relatively mild conditions of oxidation. The samples were then fully denatured with guanidine and exposed to high concentrations of ¹⁸O-labeled H₂O₂. After quenching with sodium sulfite, the protein was fragmented with chymotrypsin and evaluated by mass spectrometry in order to measure the ¹⁸O/¹⁶O ratio of each of the ten methionine residues within the TDP43 LC domain. A low ratio of ¹⁸O to ¹⁶O for a given methionine residue was interpreted to mean that this residue was easily oxidized when the protein was in the polymeric state. By contrast, a high ratio of the ¹⁸O isotope to the ¹⁶O isotope was interpreted to mean that the given methionine residue existed in a protected state in the TDP43 polymers. As shown in Figure 2, methionine residues 322 and 323 of the TDP43 LC domain were substantially protected from H₂O₂-mediated oxidation when probed in TDP43 hydrogel polymers. Precisely the same methods of H₂O₂ footprinting were used to probe conformation of the TDP43 LC domain in both liquid-like droplet samples of the protein as well as the TDP43 protein endogenous to HEK293 cells. Having observed a qualitatively similar footprint in the latter two samples, we have concluded that the LC domain of TDP43 self-assembles into a specific molecular structure that is the same in hydrogel polymers, liquid-like droplets and living cells.

Figure 2. Hydrogen peroxide footprints of the TDP43 low complexity domain as studied in hydrogel polymers, liquid-like droplets and living cells.

TDP43 protein existing in hydrogel polymers (upper left panel), liquid-like droplets (upper right panel) or living cells (lower left panel) were initially exposed to mild conditions of hydrogen peroxide-mediated oxidation. Following mild oxidation, the samples were denatured and exposed to high concentrations of ¹⁸O-labeled hydrogen peroxide. Following chymotrypsin digestion and mass spectrometry, the ratio of ¹⁸O to ¹⁶O was quantitated for each of 10 evolutionarily invariant methionine residues. In all cases, methionine residues 322 and 323 were observed to be protected from initial, ¹⁶O oxidation, giving evidence that these residues might be protected in the structurally assembled state of the TDP43 low complexity domain. Diagram shown in lower right panel corresponds to the molecular structure of polymers formed by the most evolutionarily conserved region of the TDP43 low complexity domain as resolved by cryo-electron microscopy [37]. Note that hydrogen peroxide-resistant methionine residues 322 and 323 are localized within the structured cross-β core of the assembled polymers, and that partially oxidation resistant methionine residues 336, 337 and 339 are localized on the immediate edge of the polymer core. Graphs were reproduced from [29].

The location of the H₂O₂ footprint of self-associated TDP43 coincides with a region of 22 amino acids that have been perfectly conserved through evolution over the 500M years separating fish from humans. The footprint also coincides precisely in location with a cross-β structure defined by the laboratory of David Eisenberg [37]. Intriguingly, the two amino acid side chains most protected from H₂O₂-mediated oxidation when TDP43 is allowed to assume its self-associated state, methionine residues 322 and 323, are the only methionine residues housed directly within the Eisenberg molecular structure (Figure 2). Whereas further experimentation will be required to rigorously test the validity of this hypothesis, we tentatively conclude that the Eisenberg structure may represent the molecular basis for self-association of the TDP43 LC domain.

For the yeast ataxin-2 protein, genetic experiments have led us to understand why its LC domain might be expected to function as a redox sensor. By contrast, we can only speculate as to why the TDP43 LC domain might be endowed with this feature. Knowing that TDP43 has been found as a constituent of neuronal granules that help facilitate synapse-localized translation of certain mRNAs, we have offered the suggestion that oxidation-mediated disassembly of such granules might be at play in this unexpected example of biological regulation. As shown in Figure 3, we speculate that upon exposure to low concentrations of H₂O₂, the assembled TDP43 LC domain begins to unfold via oxidation of C-terminal methionine residues. Upon exposure to even higher levels of H₂O₂, the entire complex dissolves. Like the ataxin-2 protein, we further hypothesize that disassembling of the TDP43 LC domain complex may take place in proximity to mitochondria [38].

Figure 3. Schematic diagram of hydrogen peroxide-mediated disassembly of TDP43 cross-β polymeric structure.

Three molecules of the TDP43 low complexity domain are displayed in the fully assembled cross-β structural state (left image). Dashed blue lines correspond to regions of molecular disorder, solid, continuous lines correspond to regions of structural order. Region displayed in red corresponds to sequences of perfect evolutionary conservation in the 500 million years separating humans from teleost fish. Exposure to low concentrations of hydrogen peroxide disassembles the C-terminal region of the structure via oxidation of C-terminal methionine residues (middle image). Exposure to high concentrations of hydrogen peroxide leads to extensive methionine oxidation and full disassembly of the cross-β structural state (right image).

We close by pointing out that the human ataxin-2 and TDP43 proteins are prominently implicated in neurodegenerative disease. The LC domains of both of these proteins appear to be subject to redox regulation via methionine oxidation. If biologically valid, these observations may be offering useful hints as to the importance of the methionine sulfoxide reductase enzymes in age-related neurodegenerative disease.

Highlights.

• The low complexity (LC) domains of ataxin-2 and TDP43 are methionine rich.

• Methionine oxidation blocks self-association and phase separation by both LC domains.

• Reduction of oxidized methionines by Msr enzymes restores phase separation.

• Complexes formed by the LC domains represent evolutionarily crafted redox sensors.