Several proteins involved in DNA and RNA metabolism, including histones and a large number of RBPs, undergo Arg-methylation, which may occur in different flavors (mono, symmetric and asymmetric di-methylation) (Blanc and Richard, 2017). In mammalian cells, Arg (R)-methylation is catalyzed by at least nine Protein-Arginine-Methyl-Transferases (PRMTs), from 1 to 9 (Bedford and Clarke, 2009; Blanc and Richard, 2017). Of these, at least three (PRMT1, 6 and 8) catalyze asymmetrically di-methylation on arginine (ADMA). PRMT1 catalyses the addition of one or two methyl groups to the R residues from the S-adenosylmethionine donor (Yang and Bedford, 2013). Methylation plays a major role in controlling the biophysical properties of RGG/RG motifs and modulates both protein-protein and RNA-protein interactions. Notably, arginine methylation is very relevant for FET proteins (Figure 2B) (Lorton and Shechter, 2019) while not many studies have characterized the impact of arginine methylation in TDP-43 so far, even though some arginine methylated residues had been identified and reported in databases. Indeed, in the phosphosite.org website it is reported that mass spectrometry analyses identified three TDP-43 methylated residues: Arg 42, 275 and 293. The last two modifications are located at the C-terminal domain of TDP-43, thus could in principle influence LLPS, however the functions of these modifications and their regulation have not been investigated yet (Figure 2B). Methylation does not alter the net charge of the protein, but changing its distribution can regulate the capacity of Arg to enter cation-π interactions with aromatic residues (Lorton and Shechter, 2019) and hence the ability of FUS to rapidly and reversibly form liquid droplets and hydrogels (Figure 2A). In fact, LLPS transition of FUS involves hydrogen bonding between an antiparallel β-sheet in the LCD (residues 39–95) and Arg residues in the three RGG-rich regions (Han et al., 2012; Murakami et al., 2015; Patel et al., 2015; Murray et al., 2017). Under normal conditions these Arg residues are heavily mono- or di-methylated (Rappsilber et al., 2003). In contrast, in FTLD they are hypomethylated and FUS is found in neuronal nuclear and cytoplasmic aggregates that frequently contain other members of the FET family. Several lines of evidence indicate that methylation of specific Arg residues (position 216, 259, 407, 472, 473 and 476) has an inhibitory effect on condensate formation (Qamar et al., 2018). Indeed, inhibition of arginine methyltransferase activities with Adenosine dialdehyde (AdOx) produces a significant reduction in asymmetrical di-methylation of FUS at most of these Arg (216, 259, 407, 473, and 476) and promotes LLPS. In contrast, Arg 394 and 481 remain predominantly di-methylated, indicating that a higher methylation turnover occurs only in the case of Arg residues involved in LLPS, allowing dynamically tuning phase separation. Altogether these data suggest that the number of methylated Arg residues can modulate the type of phase separation (liquid-liquid vs liquid-solid), which is driven by multivalent cation-π interactions. FUS is normally soluble and dimethylated in healthy brains while reduced levels of FUS methylation have been detected in insoluble protein inclusions in brains of FTD patients (Suárez-Calvet et al., 2016).

Paradoxically R-methylation can also promote the formation of aggregates (Table 1). Indeed, recent studies demonstrate that R-methylation in the non-canonical NLS domain of FUS influences the subcellular distribution of the protein. This is due to the fact that methylation of Arg residues in the second RGG-rich region of FUS (Figure 1) abrogates the interaction of TRN with the third RGG-rich region thus reducing FUS nuclear import and increasing its cytoplasmic concentration, thus favoring LLPS (Dormann et al., 2012). Indeed, cell treatment with methylation inhibitors or PRMT1 knock down can restore the nuclear localization of the ALS-linked FUS mutant protein Pro⁵²⁵Lys (Dormann et al., 2012). Interestingly, immunohistochemical analysis of FUS-Pro⁵²⁵Lys ALS patients revealed the presence of inclusions with methylated FUS that are undetectable in FTD patients (Dormann et al., 2012). Thus, detection and quantification of methylated forms of FUS can be a valuable biomarker of ALS and not of FTD.

Interestingly, methylation events have been shown to affect the nucleo-cytoplasmic trafficking of other RBPs such as hnRNP-A2 (Nichols et al., 2000) and binding of hnRNP-A1 to single-stranded nucleic acid is significantly reduced after arginine methylation (Rajpurohit et al., 1994).

Very little is still known about the mechanisms by which R-methylation can be erased. Recent data suggest the involvement of R-demethylating enzymes such as KDM3A, KDM4E, KDM5C (Walport et al., 2016) and JMJD6 (Chang et al., 2007), all belonging to the large family of 2-oxoglutarate-dependent dioxygenases. Numerous studies indicate that some RGG motifs are protected from methylation, while other motifs are preferentially recognized by the methylating enzymes. The molecular basis of this difference, however, is still a matter of speculation.

Interestingly, Arg residues in the RGG motifs of FET proteins and hnRNP-A1 can also undergo citrullination, catalyzed by peptidyl arginine deiminase 4 (PAD4), which significantly inhibits protein aggregation and the recruitment of FUS in arsenite-induced stress granules. In agreement with this, a lower PAD4 expression is associated with a higher risk of developing ALS (Tanikawa et al., 2018).

Contrary to methylation, phosphorylation and acetylation change the protein charge with consequent impact on proteins conformation (Hofweber and Dormann, 2019) and pattern of interaction, aspects that we will discuss in the following paragraphs.

A complex interplay between protein phosphorylation and methylation has been recently found to control the dynamics of some RBPs including hnRNP-A1 and TDP-43. For instance, cisplatin treatment (CDDP) induces phosphorylation of protein methyl-transferase PRMT1 by DNA-PK, which redirects PRMT1 activity toward chromatin-associated proteins at the cost of RBP methylation (Musiani et al., 2020). Interestingly, 82% of the down-regulated Arg-methyl sites following PRMT1 phosphorylation by DNA-PK are inside the RGG-containing LCDs of proteins undergoing LLPS. As described above Arg-methylation by PRMT1 on these proteins weakens cation–π interactions between Arg and aromatic (Phe and Tyr) residues, thus reducing LLPS. The effect of DNA-PK on this phenomenon is double. In fact, in addition to phosphorylating and redirecting PRTM1 toward chromatin, DNA-PK phosphorylates the RBPs that are a target of R-methylation in a way that inhibits their interaction with PRTM1. The net effect is that these RBPs accumulate in SG condensates (Giambruno and Bonaldi, 2020). In the case of hnRNP-A2, Tyr-phosphorylation alters the propensity of the protein to undergo LLPS in vitro, prevents partitioning of granule components and hinders aggregation of mutants associated with neurodegenerative disorders. Moreover, different phosphorylation events in the same domain may elicit different effects offering the possibility of tuning protein assemblies (Ryan et al., 2021). C. elegans experiments have identified FYN kinase as a candidate for hnRNP-A2 phosphorylation (Ryan et al., 2021). Indeed, Tyr phosphomimetic mutations, i.e., substitutions with aspartic or glutamic acid that mimic the phosphate negative charge, prevent partitioning in droplets of hnRNP-F and ch-TOG, two molecular partners of hnRNP-A2, while Ser phosphomimetic ones do not (Ryan et al., 2021).

Similarly to R-methylation, phosphorylation can either enhance or suppress LLPS of RBPs in vitro (Table 1), as clearly demonstrated for both FUS and TDP-43. In response to DNA damage the two apical DDR kinases DNA-PK and ATM catalyze the phosphorylation of different sites (Ser-26/Ser-30, Ser-30/Ser-42, Thr-109/Ser-115, and Ser-115/Ser-117 within the ATM and DNA-PK consensus Ser/Thr-Gln) localized in the LCD of FUS, a modification that has been shown to prevent liquid to-solid-state transition and the formation of fibril-like structures (Deng et al., 2014; Monahan et al., 2017; Rhoads et al., 2018a). Although the details of FUS phosphorylation in vivo are still under investigation, the involvement of two apical DDR kinases seems to suggest that protein aggregation and DDR activation might be mechanistically linked in causing neurodegeneration for a subset of ALS and FTD cases. Nevertheless, the consequence of phosphorylation on LLPS is protein- and residue specific. Indeed, Ser/Thr phosphorylation within FUS LCD reduces its aggregation in vitro and in vivo (Monahan et al., 2017), while it has the opposite effect on other proteins such as fragile-X linked protein FMRP (Tsang et al., 2019). Regarding TDP-43, phosphomimetic substitution with glutamic acid of serine 48 (Ser⁴⁸Glu), a highly phosphorylated residue in the N-terminal domain (Rigbolt et al., 2011; Hornbeck et al., 2012; Hornbeck et al., 2015; Wang A. et al., 2018), also reduces LLPS, suggesting that phosphorylation of this residue interferes with the transient and weak intermolecular interactions necessary for phase transition, possibly promoting a more rigid and structured protein conformation (Wang A. et al., 2018). In line with this, different un-phosphorylatable mutations of Ser to Ala weakly induce phase transition. On the other hand, it is also known that phosphorylated TDP-43 represents one of the predominant components of protein aggregates in ALS and FTD (Hasegawa et al., 2008; Guedes et al., 2017), in which TDP-43 is found phosphorylated in its LCD, mainly on Ser 403/404 and 409/410 (Neumann et al., 2020). Nevertheless, whether this phosphorylation occurs before or after aggregation and its possible causative role in stabilizing protein aggregation need to be yet clarified. A possible explanation for these conflicting observations could be that phosphorylation can occur after TDP-43 aggregation, as an attempt to counteract detrimental interactions throughout electrostatic repulsions (Brady et al., 2011). Conversely, another theory that has been put forward proposes that phosphorylation prevents the clearance of the aggregates, thus causing their accumulation. In support of this, Zhang et al. have shown that phosphorylated fragments are more difficult to degrade than the non-phosphorylated ones (Zhang et al., 2010). Moreover, in vitro experiment casein kinases 1 (CK) increases TDP-43 phosphorylation and aggregation (Hasegawa et al., 2008).

Among all the factors that can influence both protein phosphorylation and aggregation, ATP concentration is strictly regulated within cells. On the one hand ATP plays an indirect role in controlling the assembly of condensates via protein phosphorylation; on the other hand, high ATP concentrations (>6 mM) alone can dissolve in vitro-generated liquid condensates of several RBPs, including FUS. The effect of ATP on LLPS directly stems from its hydrotropic nature, achieved due to the presence of the aromatic ring, capable of binding the hydrophobic patches in FUS (RGG and RRM domains), and the triphosphate chain that interacts with water molecules, thus leading to dissolution of protein aggregates (Patel et al., 2017; Rice and Rosen, 2017; Kang et al., 2019). It is worth noting that the cellular ATP concentration is usually in the millimolar range (up to 10 mM), while ADP and AMP are 50 and 10 µM respectively. The high ATP consumption of neurons may reduce its cellular concentration and might contribute to why these cell types are more prone than others to fibrillar degeneration of FUS condensates.

While in other contexts of neurodegeneration protein acetylation has been widely associated with reduced protein aggregation (Saito et al., 2019), FUS and TDP-43 Lys acetylation leads to the formation of cytoplasmic protein aggregates (Table 1; Figures 2A,B).

Recently, three new acetylated Lys residues localized in different domains of FUS have been identified by mass spectrometric approaches (Arenas et al., 2020). In particular, acetylation at Lys315/Lys316 within the RRM domain strongly affects the ability of FUS of binding RNA, while acetylation of Lys510 in the NLS stimulates the formation of FUS-containing cytoplasmic aggregates (Arenas et al., 2020). Moreover, the application of a specific antibody directed against acetylated Lys510 (K510Ac) reveals a significant increase of acetylated FUS in ALS patients-derived dermic fibroblasts, suggesting the involvement of this PTM in FUS pathogenicity (Arenas et al., 2020). In this study, treatments with specific inhibitors proved that Lys510 acetylation is catalyzed by the CREB binding protein (CBP)/p300 (Figure 2B), while de-acetylation is carried out by both histone deacetylases (HDACs) and sirtuins (SIRTs) (Arenas et al., 2020). Although acetylation of FUS and TDP-43 seems to act preferably as a driving force for protein aggregates (Table 1), when occurring in specific positions it can work in the opposite direction.

Proteomics approaches have always identified the N-terminus of FUS as a preferential target of acetylation (Catherman et al., 2013; Rhoads et al., 2018b). Indeed, a recent study revealed a new FUS acetylation by the N-terminal acetyltransferases (NatA-NatF), confirmed by co-expression of recombinant FUS with the Nat A complex, which stimulates LCD LLPS without increasing the formation of aggregates (Bock et al., 2021). N-terminal acetylation is the addition of an acetyl group to the N-terminal amine group through an amide bond thus impeding protonation of the terminal amine reducing the propensity of the nearly uncharged FUS LCD domain to form aggregates.

Analogously, TDP-43 aggregation state is modulated by acetylation. It has been shown that upon sodium arsenite treatment, the CREB binding protein (CBP) acetylates TDP-43 on Lys145-Lys149. This modification impairs TDP-43 RNA binding ability and produces the accumulation of amyloid-like inclusions containing hyper-phosphorylated TDP-43 (Cohen et al., 2015). Interestingly, TDP-43 mutants bearing acetylation-mimic Lys to Gln substitution form cytosolic aggregates and exhibit other hallmarks of TDP-43 pathology (Wang et al., 2017).

As for arginine methylation in FUS, also TDP-43 tendency to acetylation appears different in ALS and FTD contexts, since the acetylated form of TDP-43 is detectable only in ALS spinal specimens and not in brain specimens from FTD-TDP-43 patients (Cohen et al., 2015), suggesting that this PTM might be a valuable specific biomarker to distinguish between these two pathologies.

One of the cellular mechanisms involved in the clearance of misfolded protein is the ubiquitin-proteasomal system that functions as a “quality control” mechanism. In ALS, a remarkable fraction of ubiquitin (Ub) is sequestered into different types of inclusions (Leigh et al., 1991), thus reducing the pool of Ub available for physiological ubiquitination of different substrates during the execution of many cellular functions including transcription, DNA repair and signal transduction (Hershko and Ciechanover, 1998; Chen and Sun, 2009).

Some evidence suggests that FUS is recruited into ubiquitin-positive cytoplasmic inclusions. However, the ubiquitin-ligase responsible for FUS ubiquitination has not yet been described (Neumann et al., 2009; Deng et al., 2010; Seelaar et al., 2010; Farrawell et al., 2015). It has been proposed that, in a neuronal context, autophagy represents the preferred mechanism for the clearance of misfolded proteins (Nijholt et al., 2011). Recent studies have directly linked autophagy and protein aggregates and the new term “aggrephagy” has been coined to define the mechanism of aggregate clearance by autophagy (Øverbye et al., 2007). Nowadays the contribution of aggrephagy is widely investigated in the context of proteinopathies. Intriguingly, aggrephagy and autophagy compete for limited amounts of intermediate structures (e.g., phagophores) and this could cause reduced autophagy efficiency in resolving the aberrant aggregation of cytoskeleton proteins upon toxic induction (Larsen et al., 2002). On the other hand, it is plausible that the presence of ubiquitin in FUS-containing inclusions indicates an initial attempt to resolve protein aggregates via the proteasome degradation pathway (Farrawell et al., 2015).

Ubiquitination also has an important role in controlling the formation of TDP-43 condensates (Figure 2). Indeed, Lys48 and Lys63 ubiquitination by E3 ubiquitin ligase Parkin leads to cytosolic accumulation of TDP-43 and the formation of cytosolic condensates (Hebron et al., 2013). Interestingly, this re-localization reduces the Parkin mRNA level, which is in turn controlled by TDP-43. Overexpression experiments have proven the formation of a multi-protein complex comprising Parkin, ubiquitinated TDP-43 and HDAC6 that facilitates cytosolic accumulation of TDP-43. Although this cytosolic complex is likely to have a physiological function, a failure of the proteasome function in neurodegenerative diseases leads to the appearance of cytosolic TDP-43 condensates (Hebron et al., 2013). Notably, it has been suggested that the Parkin-mediated ubiquitination may contribute to TDP-43 aggregation (Hebron et al., 2013). It is worth noting that TDP-43 can be ubiquitinated by several other enzymes. An example is the UBE2E class of ubiquitin-conjugating enzymes that ubiquitylate TDP-43 at Lys263 (Hans et al., 2014). More recently, it has been reported that cytoplasmic inclusions resulted from the expression of two fALS mutants: TDP-43-Met³³⁷Val and FUS-Arg⁴⁹⁵x. These mutants co-localized with polymeric Ub^K63, which is associated with the autophagy-related clearance mechanism. Intriguingly, the expression of FUS-Arg⁴⁹⁵x causes the reduction of monomeric ubiquitin levels that can disrupt ubiquitin homeostasis (Farrawell et al., 2020).

The level of ubiquitinated proteins within the cell is tightly regulated by deubiquitinating enzymes (DUBs), which counterbalance ubiquitin ligase activity and comprise a large family of enzymes with different specificities and catalytic activities. Due to their role in removing ubiquitin signaling, DUBs are implicated in a wide range of cellular processes and differentially accumulate in distinct functional compartments, based on their primary role (Clague et al., 2019). For instance, some cytosolic DUBs are coupled with the proteasome activity, and therefore may potentially be of major relevance in the modulation of protein aggregation state in the context of neurodegenerative disorders. Among these, Ubiquitin-specific protease 14 (USP14) is catalytically active only when bound to the 26S proteasome and contributes to the cleavage of ubiquitin from substrates before their degradation (Borodovsky et al., 2001; Hu et al., 2005). It has been shown that proteasome-associated USP14 deubiquitinates TDP-43 and that USP14 inhibition accelerates TDP-43 turnover. In particular, overexpression of WT USP14 in mouse embryonic fibroblasts leads to increased TDP-43 levels. This effect is abolished by both the expression of USP14 catalytically inactive form and USP14 small molecule inhibition (Lee et al., 2010), prompting the idea that ubiquitin chain trimming by USP14 might act as an antagonist of proteasome function.

It is worth noting, that DUBs have also been implicated in autophagy mechanisms. A genetic screen in Drosophila larval fat body identified Ubiquitin Iso-peptidase Y (UBPY), also called USP8, as a key player in the autophagy flux, whose RNAi-mediated silencing led to lysosomal defects and accumulation of malfunctioning autophagosomes (Jacomin et al., 2015). Notably, TDP-43 Lys²⁶³Glu mutant undergoes pathological hyper-ubiquitination and aggregation and if the ubiquitin-conjugating enzyme UBE2E3 actively ubiquitinates it, UBPY is able to counteract this PTM. The silencing of UBPY in fact enhances neurodegeneration in the retina of TDP-43 Drosophila ALS model system with accumulation of ubiquitinated and insoluble TDP-43. In this way, UBPY participates in the regulation of TDP-43 Lys²⁶³Glu solubility and exerts a neuroprotective function in Drosophila melanogaster (Hans et al., 2014). Moreover, recent results demonstrated that the Ubiquitin Specific protease 10 (USP10) positively regulates the stability of the autophagic protein LC3B, counterbalancing LC3B degradation and thus enhancing clearance of protein aggregates under stress conditions (Jia and Bonifacino, 2021).

Intriguingly, not only can DUBs modulate the protein aggregation state but they are also are recruited to membrane-less organelles or even have the ability to phase separate themselves. One example is the human USP42, which drives nuclear speckle phase separation dependent on its de-ubiquitinating activity, thus governing mRNA splicing events (Liu et al., 2021). In addition, it has been proven that other two human DUBs, namely USP5 and USP13, are recruited to stress granules shell and dictate their stabilization or disassembly through their activity of removing ubiquitin chains (Xie et al., 2018). Human USP5 also seems to target and regulate the expression of hnRNP-A1 (Vashistha et al., 2020), whereas the fly DUB Otu possess an LC domain that drives the formation of amyloid-like granules, which resemble FUS and hnRNP-A1 structures and are indispensable for Otu enzymatic activity (Ji et al., 2019).

Overall, while ubiquitinated RBPs seem to be generally associated with aggregates formation, the clearance of this form by means of DUBs could be useful in reversing or preventing these toxic events. Nevertheless, intricate networks, which are far from being fully understood, controls the cellular balance of protein ubiquitination and ubiquitin mediated signaling, which could contribute to fostering proteinopathies in neurodegenerative disorders.

Poly ADP-ribosylation (PARylation) is a reversible PTM in which ADP-ribose (ADPr) units are added to the Glu, Asp, Lys, Arg or Ser residues by poly (ADP-ribose) (PAR) polymerases (PARPs). This process is reversed by PAR glycohydrolase (PARG) (Slade et al., 2011). It has been shown that PARylation regulates the dynamics of the SGs and that PARP1 activation upon cell treatment with H₂O₂ markedly increases the level of PARylated hnRNP-A1 (Duan et al., 2019). Indeed, Lys298 (K298), immediately C-terminal of the M9 motif at the CTD of the protein (Figure 1) is a PARylation site. Interestingly, hnRNP-A1 also contains a PAR-binding motif, located between the two RRM domains (position 92–113, Figure 1). It mediates the interaction with PARylated proteins and its mutation abrogates the recruitment of hnRNP-A1 to stress granules. Decreased PARylation levels suppress the formation of SGs and the recruitment of hnRNP-A1 and TDP-43 to SGs, while higher PARylation levels delay the disassembly of SGs. Considering the close proximity between the PAR-binding motif and the RRMs in hnRNP-A1, their PARylation has an impact on the interaction with RNA. In addition, K298 PARylation regulates the nucleocytoplasmic transport of hnRNP-A1 and stress induced-K298 PARylation may serve as a nuclear export signal (Duan et al., 2019).