Proteostatic imbalance and protein spreading in amyotrophic lateral sclerosis

Abstract

Amyotrophic lateral sclerosis (ALS) is a fatal neurodegenerative disorder whose exact causative mechanisms are still under intense investigation. Several lines of evidence suggest that the anatomical and temporal propagation of pathological protein species along the neural axis could be among the main driving mechanisms for the fast and irreversible progression of ALS pathology. Many ALS‐associated proteins form intracellular aggregates as a result of their intrinsic prion‐like properties and/or following impairment of the protein quality control systems. During the disease course, these mutated proteins and aberrant peptides are released in the extracellular milieu as soluble or aggregated forms through a variety of mechanisms. Internalization by recipient cells may seed further aggregation and amplify existing proteostatic imbalances, thus triggering a vicious cycle that propagates pathology in vulnerable cells, such as motor neurons and other susceptible neuronal subtypes. Here, we provide an in‐depth review of ALS pathology with a particular focus on the disease mechanisms of seeding and transmission of the most common ALS‐associated proteins, including SOD1, FUS, TDP‐43, and C9orf72‐linked dipeptide repeats. For each of these proteins, we report historical, biochemical, and pathological evidence of their behaviors in ALS. We further discuss the possibility to harness pathological proteins as biomarkers and reflect on the implications of these findings for future research.

This review summarizes the mechanisms involved in aggregation of ALS‐associated proteins and in their transmission between cells, representing pathological hallmarks of this neurodegenerative disease.

Glossary

Aβ

Amyloid‐β

AD

Alzheimer’s disease

ALS

Amyotrophic lateral sclerosis

CJD

Creutzfeldt‐Jacob disease

cKO

Conditional knock‐out

CNS

Central nervous system

CPS

Conventional protein secretion

CSF

Cerebrospinal fluid

C9orf72

Chromosome 9 open reading frame 72

DPR

Dipeptide protein

eIF2α

Eukaryotic initiation factor 2α

ER

Endoplasmic reticulum

ERAD

Endoplasmic reticulum‐associated degradation

ESCRT

Endosomal sorting complex required for transport machinery

ESE

Early‐sorting endosomes

EV

Extracellular vesicle

FF

Fast‐twitch fatigable

FFR

Fast‐twitch fatigue‐resistant

FTD

Frontotemporal dementia

FUS

Fused in sarcoma

G3BP

Ras GTPase‐activating protein‐binding protein

hnRNP

Heterogeneous nuclear ribonuclear protein

HRE

Hexanucleotide repeat expansion

Hsp

Heat‐shock protein

Htt

Huntingtin

IDR

Intrinsically disordered region

ILV

Intraluminal vesicle

iPSC

induced pluripotent stem cell

jALS

Juvenile amyotrophic lateral sclerosis

LCD

Low‐complexity domain

LMN

Lower motor neuron

LSE

Late‐sorting endosome

MN

motor neuron

MVB

Multivesicular bodies

NFL

Neurofilament light chain

NLS

Nuclear‐localization signal

NTD

N‐terminal domain

PD

Parkinson’s disease

PERK

Protein kinase RNA‐like ER kinase

PLD

prion‐like domain

pNFH

phosphorylated neurofilament heavy chain

PROTAC

Proteolysis targeting chimera

PrP

Prion protein

pTDP‐43

Phosphorylated TDP‐43

RAN‐T

Repeat associated non‐AUG translation

RBP

RNA‐binding protein

RNP

Ribonucleoprotein

RRM

RNA‐recognition motif

sALS

Sporadic amyotrophic lateral sclerosis

SFR

Slow‐twitch fatigure‐resistant

SG

Stress granule

SIMOA

Sensitive single molecule assay

SOD1

Superoxide dismutase 1

TDP‐43

TAR DNA‐binding protein 43

TNT

Tunneling nanotube

UMN

Upper motor neuron

UPS

Ubiquitin‐proteasome system

The progressive nature of neurodegenerative diseases

Neurodegenerative disorders, such as Alzheimer’s disease (AD), Parkinson’s disease (PD), frontotemporal dementia (FTD), and amyotrophic lateral sclerosis (ALS), are incurable neurological conditions characterized by the irreversible loss of specific neuronal subtypes. Despite the apparent cell type selectivity, these disorders share common features, including (i) focal onset, (ii) progressive neurodegeneration, and (iii) multisystemic involvement at a late stage (Seeley, 2017). In fact, while pathology typically ensues with the localized loss of a subset of neurons, degeneration gradually progresses in a spatiotemporal fashion, ultimately affecting extended areas of the central nervous system (CNS) and involving different neural cell types. For example, AD has been documented to advance through distinct stages, and topological degeneration across the brain has been mapped (Ewers et al, 2008). PD has been primarily associated with the selective loss of dopaminergic neurons in the substantia nigra pars compacta. However, recent evidence has demonstrated that although the disease becomes clinically evident upon loss of about 60–70% of nigrostriatal dopaminergic neurons, pathology is concomitantly found in other areas of the central and peripheral nervous system (Halliday et al, 2010). Indeed, besides exhibiting classical motor symptoms, PD is frequently associated with cognitive impairment and dementia, coherently with the involvement of non‐dopaminergic loci (Poewe et al, 2008). Similarly, ALS predominantly affects upper and lower motor neurons, but glial cells and other neuronal populations have also been observed to display pathological hallmarks of the disease. Moreover, cortical areas other than the motor cortex may undergo progressive degeneration causing FTD‐related symptoms (Hudson, 1981). Importantly, FTD patients occasionally develop motor neuron disease, and for this reason, ALS and FTD have been suggested to be at the edges of the same disease spectrum (Couratier et al, 2017). These observations have been possible thanks to the combination of clues derived from human post‐mortem material (Braak & Braak, 1991; Strong et al, 2005; Kalaitzakis et al, 2008), longitudinal neuroimaging of living patients (Ewers et al, 2008; Seeley et al, 2009; Brettschneider et al, 2015; Müller & Kassubek, 2018; Young et al, 2020), and clinical monitoring (Perlmutter, 2009; Turner et al, 2010; Agüera‐Ortiz et al, 2010; Fang et al, 2017). Additionally, the development of in vitro and in vivo disease models has helped to elucidate, at least in part, the molecular basis of disease pathogenesis (Luk et al, 2012; Ruiz‐Riquelme et al, 2018; Kim et al, 2019; McAlary et al, 2019). Although the exact mechanisms underlying neurodegeneration are multifactorial and not fully understood, a unifying feature of neurodegenerative diseases is the presence of protein aggregates in degenerating neurons and glia (Ross & Poirier, 2004; Shaw, 2005). For instance, AD is characterized by extracellular deposits of amyloid‐β (Aβ) as well as intracellular neurofibrillary tangles of tau protein; PD exhibits Lewy bodies of aggregated α‐synuclein; ALS and FTD display a variety of aggregates including TAR‐DNA‐binding protein 43 (TDP‐43), fused in sarcoma (FUS), Cu²⁺/Zn²⁺ superoxide dismutase 1 (SOD1), dipeptide repeats (DPRs), p62 and ubiquitinated proteins. Due to the widespread presence of aggregates and misfolded proteins, these disorders have been designated as “proteinopathies”.

The observation that affected cells exhibits progressive accumulation of insoluble aggregates with a precise spatiotemporal distribution along the CNS has led to postulate that protein aggregation and progression of disease could be two consequential, interconnected events. In particular, it has been hypothesized that proteinaceous aggregates may initiate disease via gain of toxic properties, including sequestration of critical components of physiological processes such as DNA damage response and RNA metabolism, or via obstruction of axonal trafficking and induction of inflammation (Ilieva et al, 2009). Subsequently, disease progression would be sustained by the gradual collapse of protein homeostasis across the different cells of the neural axis as the result of cell‐to‐cell transmission of pathogenic protein species and aggregates. In this manuscript, we review several lines of evidence supporting the concept of protein misfolding, seeding and spreading in neurodegeneration, and focusing our discussion on ALS. We report the experimental findings converging on the protein propagation hypothesis and reflect on the implications of these studies in the context of disease monitoring and treatment.

Amyotrophic lateral sclerosis affects the motor system

In spite of early case reports from the 1820s, the first systematic description of ALS dates back to the 1860s, when the French neurologist Charcot correlated the disease to its distinct neuropathology (Kumar et al, 2011). Nevertheless, ALS gained public recognition only in the 1930s, when the baseball player Lou Gehrig, renowned for his prowess as a hitter and for his durability, which earned him his nickname “The Iron Horse”, was forced to quit the game due to developing progressive muscle weakness. ALS is the most widespread motor neuron (MN) disorder, with an average incidence of 2 cases per 100,000 people per year on a global scale (Talbott et al, 2016; Gowland et al, 2019). The pathology affects men ~1.5 times more than women, and ethnicity has been recently found to play a role in its occurrence (McCombe & Henderson, 2010). For instance, compared with an average incidence of > 3 cases per 100,000 individuals in Europe, the incidence of ALS is much lower in Asia (0.8 cases per 100,000 individuals), where mean survival time is also longer (48 months as opposed to 24 months (Hardiman et al, 2017)). ALS is usually diagnosed between 55 and 65 years of age. As such, age has been indicated as the main risk factor for ALS and more in general neurodegenerative diseases (Niccoli et al, 2017). However, ALS may also affect younger individuals, giving rise to “juvenile” ALS (jALS). Early‐onset ALS accounts for a relatively small number of cases and has been speculated to be linked to mutations that fast‐track age‐related pathogenic mechanisms (Turner et al, 2012). Of all the reported cases of ALS, approximately 90% are sporadic (sALS), i.e., of apparently undefined family history or characterized by de novo somatic mutations. The remaining cases are inherited (familial ALS or fALS) and directly caused by genetic mutations that typically follow an autosomal dominant pattern (Chen et al, 2013). The identification of disease‐associated loci has been a groundbreaking discovery in the ALS research field, as it has provided the starting point for the discovery of a number of pathogenic mechanisms at the roots of the disease. SOD1, or ALS1, was the first ALS‐associated gene to be discovered in studies dating back to 1993 (Rosen et al, 1993). Being SOD1 an antioxidant enzyme, oxidative stress was at first explored as a potential cause of ALS. However, ALS‐linked SOD1 variants were also found to deposit into intracellular inclusions, indicating the existence of pathogenic mechanisms resulting from protein misfolding. Yet, only 15–20% of fALS stem from mutations in SOD1 (Pansarasa et al, 2018). In 2006 and 2009, respectively, genes encoding the RNA‐binding proteins TDP43 and FUS were added to the list of ALS candidate genes and each are currently known to account for 4–5% of fALS cases (Neumann et al, 2006; Kwiatkowski et al, 2009; Kapeli et al, 2017). Importantly, both proteins can aggregate in insoluble inclusions. In particular, TDP‐43 aggregates have been found in about 95% of all ALS cases (Hergesheimer et al, 2019). However, in spite of the common structure and function of these proteins, TDP‐43 inclusions are typically FUS‐negative and vice versa, suggesting that they may be triggered by independent hits eventually converging into the disruption of shared downstream pathways (Maniecka & Polymenidou, 2015). In 2011, an aberrant hexanucleotide repeat expansion within the chromosome 9 open reading frame 72 (C9ORF72) gene was found to be the main genetic determinant of fALS (Renton et al, 2011). About 40% of fALS but also ~7% of sALS have been associated with the aberrant expansion in C9ORF72 (Iacoangeli et al, 2019). Currently, more than 25 genetic loci have been linked to ALS, and it is generally accepted that the disruption of protein homeostasis acts as a critical component of disease pathogenesis (Nguyen et al, 2018).

The end product of the pathological cascade caused by ALS mutations is neurotoxicity, with prime involvement of the motor system (Nijssen et al, 2017) (Fig 1). Motor signals originate by the upper motor neurons (UMNs) of the motor cortex, which integrate excitatory and inhibitory stimuli from the surrounding cortical areas and translate them into signals that will either promote or inhibit voluntary movement (de Carvalho et al, 2014; Zayia; & Tadi, 2020). After decussating, the axons of UMNs reach out to the lower motor neurons (LMNs) of the brain stem (corticobulbar tract) and the spinal cord (corticospinal tract). Spinal MNs have been classified into the following: (i) alpha, (ii) beta, and (iii) gamma (Stifani, 2014). Alpha MNs are critical for muscle contraction and have been, in turn, divided into (i) slow‐twitch fatigue‐resistant (SFR), (ii) fast‐twitch fatigue‐resistant (FFR), and (iii) fast‐twitch fatigable (FF). These subtypes differ in terms of metabolism and firing activity, which appear to be major determinants of MN vulnerability in ALS (Khademullah et al, 2020). ALS affects both UMNs and LMNs. While the word “amyotrophic” denotes the muscular atrophy resulting from the loss of neuromuscular junctions, “lateral sclerosis” refers to the degeneration of the corticospinal tract, typically followed by extensive gliosis and tissue fibrosis (Fig 1) (Rowland & Shneider, 2001).

Figure 1. Neuroanatomical propagation of ALS pathology.

(A) Schematic of the central nervous system (CNS), coronal plane, with the brain motor cortex highlighted in dark red. Upper motor neurons (UMNs) descend contralaterally and synapse with lower motor neurons (LMNs) in the spinal cord. (B) In the human brain, ALS pathology spreads sequentially in a corticofugal fashion from the motor cortex (darker red) to the surrounding cortical areas (lighter red). The brain is displayed in lateral view. (C) LMNs contact skeletal muscles to control contraction. The neuromuscular junction of ALS patients is disrupted due to peripheral axonal degeneration. Black arrows indicate the direction of pathological spread, which is likely a combination of dying forward and dying back phenomena.

The neuroanatomical spread of ALS symptoms and pathology progressively extends beyond the motor system

In most cases, the clinical symptoms of ALS appear first in one of the limbs (“limbic” onset), followed by their spread to the contralateral or ipsilateral limb (Turner et al, 2010). But ALS can also manifest itself with dysarthria/dysphagia (“bulbar” onset). Interestingly, symptoms have a higher likelihood to develop from the bulbar area to the limbs than vice versa, and bulbar onset tends to have a worse disease course than limbic onset (Ravits et al, 2007; Wijesekera et al, 2009). Only a minority of cases begins with respiratory muscle involvement, and this type of onset has the poorest prognosis (Swinnen & Robberecht, 2014). Classic motor symptoms of weakness, fasciculation, and paralysis are often accompanied by cognitive defects and emotional liability. In fact, 3–5% of ALS patients develop FTD, exhibiting behavioral changes and frontotemporal deficits (Ingre et al, 2015). When breathing muscles are affected, artificial ventilation becomes essential for survival, and respiratory failure represents the main cause of death (Radunovic et al, 2017). A number of scales, such as the ALS Functional Rating Scale (ALSFRS), the Milano‐Torino (MiToS), and the King’s staging system (Tipton‐Burton, 2011; Chiò et al, 2015), have been developed to assess clinical disease progression and confirm ALS as a complex multi‐system disorder (Abrahams et al, 1996).

Clinical manifestations are the sum of observable pathological signs linked to disease. However, these manifestations do not immediately reflect the extent of cellular pathology. Increasing evidence suggests ALS to be primarily a disease of the cerebral cortex, characterized by anterograde spread proceeding along with cortical connections (Baker, 2014). Yet, other studies have reported axonal transport defects in LMNs as well as diffused muscle denervation in the presence of unaltered numbers of spinal cord MNs (Fischer et al, 2004; Dadon‐Nachum et al, 2011). This has led to the “dying back” hypothesis, which considers ALS as a pathology of LMNs progressing from the neuromuscular junction to the neuronal soma, followed by retrograde transmission to the CNS. It is likely that the “dying back” and “dying forward” hypotheses are not mutually exclusive processes, but rather phenomena that take place simultaneously, equally contributing to disease spread and symptoms. Within the same individual, propagation of pathology may occur contiguously or non‐contiguously (Kanouchi et al, 2012). Contiguous propagation requires cell‐to‐cell contact and is driven by cell proximity. Because clinical symptoms become evident when a localized group of neurons is affected, this type of spread may explain the development of focal lesions and their subsequent amplification at a local level. However, propagation may also occur in a non‐contiguous fashion via long‐range interactions, including trans‐synaptic contacts and non‐synaptic release of effector biomolecules such as extracellular vesicles in the biological fluids, e.g., the cerebrospinal fluid. This model would explain regional propagation. Although regional spread may also be caused by the independent development of multiple local sites of pathology, it is currently believed that disease propagation results from the combination of both contiguous and non‐contiguous spread (Kanouchi et al, 2012). To sum up, the clinical heterogeneity of disease onset suggests an initial stochastic involvement of neuronal populations along the neural axis. However, the convergence of clinical phenotypes at a later stage supports the evidence that progression proceeds in an orderly fashion through the CNS connectome, eventually extending beyond motor manifestations.

ALS mutations often affect ubiquitous proteins but do not impact all cells equally, which explains the stereotypical propagation of pathology across neurons. Neurons are highly specialized and delicate cells that become symptomatic when pathological insults exceed a relatively low threshold of tolerance, leading first to their dysfunction and then demise. However, substantial evidence has demonstrated that distinct neuronal subtypes exhibit varying ability to cope with stress and resist neurodegeneration. UMNs and LMNs are the best characterized, but the list of vulnerability in ALS extends to other neurons including, for instance, spinal inhibitory interneurons (Hossaini et al, 2011; de Carvalho et al, 2014; Rudnick et al, 2017). In ALS with dementia, pathology has been additionally found in the dentate gyrus of the hippocampus, as well as in the superficial layers of the frontotemporal cortex and the entorhinal area (Wightman et al, 1992; Ikemoto et al, 1997). Interestingly, even within MNs there are different degrees of vulnerability. For example, it is primarily FF alpha‐MNs that degenerate in ALS (Lalancette‐Hebert et al, 2016). In contrast, cranial nerves III, IV, and VI, which control ocular movements, remain mostly intact during disease, which is what allows patients to communicate with the external world with the aid of computer devices during advanced disease stages (Caligari et al, 2013; Nijssen et al, 2017). Furthermore, innervation of the pelvic floor sphincter muscle is typically preserved, precluding incontinence in ALS patients (Carvalho, Schwartz, & Swash, 1995). The molecular mechanisms underlying neuronal susceptibility throughout disease onset and progression are still poorly understood, but they are thought to be linked to a combination of extrinsic factors (non‐cell autonomous) and intrinsic properties (cell‐autonomous) (Nijssen et al, 2017; Ragagnin et al, 2019). In particular, conditions leading to imbalances in protein homeostasis appear to play a critical role in priming selective vulnerability and sustaining propagation of pathology, as demonstrated by the neuroanatomical spread of protein aggregates throughout disease.

Protein aggregation is a central pathological hallmark of ALS

Followed by Bunina bodies and hyaline conglomerate inclusions of neurofilaments, ubiquitinated inclusions, either skein‐like (filamentous) or Lewy body‐like (spherical), are the most common aggregates found in ALS patients (Bunina, 1962; Chou et al, 1996; Rouleau et al, 1996; Ince et al, 1998). Irrespective of their morphology, ubiquitinated inclusions are also positive for p62 (Mizuno et al, 2006), which is suggestive of defects in protein turnover. Thus, it is now broadly accepted that the disturbance of protein homeostasis (proteostasis) is a key hallmark and perhaps determinant of numerous neurodegenerative disorders, including ALS (Blokhuis et al, 2013). Importantly, the relevance of proteostatic pathways in ALS has been substantiated by extensive genetic evidence, such as that several ALS‐associated proteins, including p62, valosin‐containing protein (VCP), ubiquilin‐2 (UBQLN2), optineurin (OPTN), and TANK‐binding kinase 1 (TBK1) are involved in these pathways (Webster et al, 2017). Proteostatic impairment underlies protein aggregation, which in turn can further hamper cellular proteostasis. Recent evidence suggests that aggregated proteins may be released in the extracellular space probably as the ultimate attempt of the cell to overcome the proteotoxic burden (Turner et al, 2005). Subsequently, released proteins would be taken up by neighboring cells, and the internalization of these proteins could amplify proteostatic imbalance in recipient cells as well as hampering other cellular processes, thus triggering a chain reaction that facilitates propagation of protein pathology. The initial aggregation of most ALS‐associated proteins and aberrant peptides typically occurs (i) as a result of imbalances in the protein quality control systems, or (ii) due to intrinsic properties encoded in their sequence (Figs 2 and 3).

Figure 2. Defects in protein quality control lead to protein aggregation.

A critical balance of molecular players prevents protein aggregation in physiological conditions. Molecular chaperones, the ubiquitin‐proteasome system (UPS) and autophagy synergistically cooperate to avert accumulation of misfolded proteins and aggregation‐prone species. Additionally, misfolded proteins in the endoplasmic reticulum (ER) induce ER stress, which potentiates the protein quality control machinery and targets misfolded proteins to degradation via the ER‐associated protein degradation (ERAD) pathway. If misfolded proteins accumulate, they may form higher order structures, such as oligomers and insoluble aggregates. Imbalances in protein quality control represent a major cell autonomous mechanism in ALS pathogenesis.

Figure 3. Intrinsically disordered domains: from SGs to insoluble aggregates.

Intrinsically disordered domains (IDRs) in ALS‐associated proteins (Gly‐rich, GQSY‐rich regions and polyQ tracts) are responsible for liquid–liquid phase separation (LLPS), a process underlying SGs assembly. Many ALS‐proteins carry mutations located within their IDRs. These mutations, combined with conditions of prolonged stress, might prevent SG dissolution. Persistent SGs may in turn serve as seeds for a liquid‐to‐solid phase transition leading to the deposition of insoluble protein aggregates.

Aggregation following imbalances in the protein quality‐control machinery

A composite network of players synergizes to preserve normal proteome conditions throughout the cell’s life, including molecular chaperones, the proteasomes, and the autophagic system. The chaperone system, including proteins of the heat‐shock protein (Hsp) family, is designed to counteract the natural tendency of proteins to aggregate, both during translation and after cytotoxic stress (Barral et al, 2004) (Fig 2). Molecular chaperones recognize hydrophobic amino acid sequences exposed by unfolded or misfolded proteins and provide crucial aid in favoring the rearrangements needed for the acquisition of the correct three‐dimensional protein conformation (Braakman & Bulleid, 2011). This activity is fundamental in the crowded environment of the cell, where high concentrations of macromolecules cause an important portion of the available volume to be physically occupied, and these conditions thermodynamically favor the interaction of solutes, facilitating aggregation (Ellis, 2001). Interestingly, some neuronal subtypes, such as MNs, intrinsically express Hsps at particularly low levels throughout life (Batulan et al, 2003), which may account for their increased susceptibility to protein misfolding pathology. As a compensatory mechanism, evidence indicates that neurons may actively take up Hsps released in the extracellular space by glial cells (Robinson et al, 2005; Taylor et al, 2007a).

When molecular chaperons are deficient, overwhelmed, or dysfunctional, the probability of protein aggregation increases. The surveillance pathways of the ubiquitin‐proteasome system (UPS) and autophagy will attempt to compensate for the generation of aggregates by inducing their clearance (Fig 2). The UPS is primarily involved in the basal turnover of proteins, which ensures a tight control over their expression levels (Huang & Figueiredo‐Pereira, 2010). However, the UPS also plays a crucial role in the proteolysis of persistently misfolded proteins through their poly‐ubiquitination and degradation via the 26S proteasome (Livneh et al, 2016). This critical role was demonstrated by multiple sources. Early indications came from studies analyzing proteasome mutants, which showed that cells carrying defective proteasomes had drastically reduced degradation rates and accumulated abundant ubiquitinated proteins under stress (Heinemeyer et al, 1991). In the context of ALS, pharmacological inhibition of the proteasomes via MG132 or lactacystin could induce significant accumulation and aggregation of TDP‐43 protein in cultured cells (Nonaka et al, 2009; van Eersel et al, 2011). Similarly, MG132 treatment effectively caused the aggregation of mutant SOD1 and exacerbated p62 levels in mutant FUS cells (Benmohamed et al, 2011; Marrone et al, 2019). Importantly, proteasomal activity declines with age, increasing the risk of protein accumulation and aggregation (Saez & Vilchez, 2014).

Protein degradation is further sustained by autophagy (from the Greek “self‐eating”), which is a quality control system centered on the catabolic activity of lysosomal organelles and the presence of double‐membrane vesicles called “autophagosomes” acting as cargo transporters (Fig 2) (Ravikumar et al, 2010). Commonly, autophagy is initiated by the formation of a cup‐shaped autophagosome precursor at the level of endoplasmic reticulum (ER)‐mitochondria contact sites (Hamasaki et al, 2013; Yoshii & Mizushima, 2017). The exposure of LC3‐II protein on both sides of the double‐membrane facilitates cargo recruitment via the adaptor proteins p62 and optineurin, which simultaneously recognize LC3‐II and poly‐ubiquitinated protein domains (Birgisdottir et al, 2013). Subsequent to sorting and closure, mature autophagosomes are directed to the lysosomes, where substrates are degraded by acidic hydrolases, followed by the reconstitution of free lysosomes (Chen & Yu, 2017; Yim & Mizushima, 2020). Defects at any stage of this highly regulated multi‐step process result in the aberrant accumulation of undigested products, to which neurons are particularly susceptible (Onyenwoke & Brenman, 2015). Previous works have investigated the implications of autophagic dysfunction in neurodegeneration. For instance, Komatsu and colleagues generated a mouse model in which the autophagy‐related gene 7 (Atg7) was conditionally knocked out (Atg7 cKO) (Komatsu et al, 2005). These mice exhibited multiple abnormalities, including deformed mitochondria and protein aggregates. Additionally, neuromuscular junctions were structurally and functionally affected. When Rudnick et al (2017) crossed the Atg7 cKO mice with an SOD1 mutant background, they detected accelerated denervation and a more dramatic accumulation of p62‐positive ubiquitinated aggregates. Similar findings were observed using a number of other animal and cell culture models upon chemical inhibition of the autophagic pathway (Shen et al, 2011; Nalbandian et al, 2015; Marrone et al, 2019).

If misfolded proteins accumulate in the ER, the ER stress response triggers an integrated intracellular signaling pathway that attempts to restore proteostasis by increasing the expression of molecular chaperones, enhancing autophagic clearance, and/or upregulating the ER‐associated degradation (ERAD) machinery (Fig 2), which retro‐translocates target proteins to the cytosol and marks them for proteasomal degradation. In addition, the ER sensor and effector molecule protein kinase RNA‐like ER kinase (PERK) phosphorylates the eukaryotic initiation factor 2α (eIF2α) inducing stress granule formation and reducing de novo protein synthesis as a measure to counteract elevated protein levels (Liu & Kaufman, 2003). Importantly, a distinguishing trait of rapidly degenerating FF alpha MNs in the spinal cord is the presence of marked ER stress, which further confirms the aberrant accumulation of misfolded protein species in ALS (Saxena et al, 2009). Disruption of the protein quality control systems is invariantly associated with a dangerous increase in protein levels. It has been suggested that several proteins are inherently metastable in the cytosol and that when their concentrations exceed a critical threshold, such “supersaturated” proteins are driven to aggregation (Ciryam et al, 2015). Interestingly, numerous proteins appear to be expressed at levels that are near or even above their critical values (Tartaglia et al, 2007). For these reasons, loss of protein solubility is not a surprising consequence of perturbations of this finely tuned equilibrium. Recent evidence has shown that proteins found in ALS aggregates share the peculiar feature of being supersaturated in spinal MNs (Yerbury et al, 2019). Yerbury and colleagues showed that the spinal MN proteome is characterized by greater supersaturation than other ALS‐resistant neurons, such as oculomotor neurons. Additionally, affected neurons reacted to pathology by downregulating supersaturated proteins at the transcriptional level, which provided additional evidence as to why spinal MNs may be so vulnerable to disease.

Aggregation propensity can be encoded in a protein’s sequence

Multiple studies have proposed that impaired protein degradation, in concert with prion‐like behaviors, is the driving force of protein aggregation and spread in ALS. Indeed, proteins that form inclusions in ALS post‐mortem tissues share some key features with the infectious prion protein (PrP) at the core of transmissible spongiform encephalopathies, such as Creutzfeldt‐Jacob disease (CJD) (Prusiner, 1998). Prions are proteins that can adopt aberrant conformations (misfolding) in response to external stimuli or in the presence of certain mutations (Prusiner, 1982; Legname et al, 2004; Moore et al, 2009; Colby et al, 2009). Misfolded proteins thus act as “seeds” of aggregation by promoting the conversion of normal proteins into their erroneous conformation. This process is self‐amplifying and develops via the induction of β‐sheet structures that act as a nidus for protein assembly into oligomers (small assemblies of misfolded proteins) and fibrils/amyloids (extensive formations of β‐sheet stacks) (Fontaine & Brown, 2009; Stöhr, 2012; Wang & Roberts, 2018). This process is usually unfavored because a high amount of energy is required to convert the naïve protein into its metastable counterpart with exposed β‐sheets, but it can be accelerated by the existence of pre‐formed oligomers/fibrils. Once the protein has acquired a metastable state, aggregate formation is thermodynamically favored, as it brings the system back to low free energy (Espargaró et al, 2008). Both oligomers and fibrils can eventually be transmitted from cell to cell, perpetuating this aberrant cycle (Bartz et al, 2000). Notably, prions acquire a wide range of aberrant conformations termed “strains”, which reflects their different priogenicity and aggregation propensity (Morales et al, 2007). Prp prions strains might co‐occur in the same patient; interestingly slow and fast progressor strains can influence each other fate (Sigurdson et al, 2019). Also, prions are characterized by a high degree of infectivity as they are not only transmitted among individuals of the same species but can also spread among different species (Scialò et al, 2019).

ALS and other neurodegenerative disorders have been defined as prion‐like diseases because aggregated proteins found in post‐mortem tissue show morphological hallmarks similar to prion proteins. Many studies have also shown the propensity of ALS‐causative proteins and proteins variants to be cell‐to‐cell transmitted. However, an important difference is that these proteins lack inter‐individual infectivity (Polymenidou & Cleveland, 2011; Aguzzi & Altmeyer, 2016; Scheckel & Aguzzi, 2018). Interestingly, misfolded mutant SOD1 is capable of recruiting properly folded wild‐type counterparts and to form aggregates. Moreover, multiple strains of mutated SOD1 have been found in patient motor neurons lysates ascribing a prion‐like behavior to mutant SOD1 (Bergh et al, 2015). Different is the case of the aberrant dipeptide repeat proteins associated to the C9orf72 mutation. Even though they have high aggregation propensity, they do not display prion‐like behavior as they are not endogenous proteins (Boeynaems et al, 2017). TDP‐43 and FUS represent two very interesting paradigms to study prion‐like mechanisms in ALS. They are both proteins containing intrinsically disordered regions (IDRs), which allows them to adopt a range of metastable states from unstructured to partially structured. Because of this property, IDRs can be at the basis of the formation of protein assemblies arising from the establishment of non‐specific interactions between different IDRs or between folded and unfolded domains (Chiesa et al, 2020). TDP‐43 and FUS also exhibit low‐complexity domains (LCDs), whereby “low‐complexity” indicates the presence of a limited repertoire of amino acids, such as polyglutamine (polyQ) repeats, glutamine/asparagine (Q/N)‐rich motifs, clusters of hydrophobic, and aromatic residues, and amino acid patches displaying charged side‐chains (Lin et al, 2016; Pak et al, 2016) (Fig 3). It has been shown that the specific amino acid sequence of LCDs dictates their behavior as prion‐like. Prion‐like domains (PLDs) exhibit sequence similarities with yeast proteins that were found to propagate as alternative states, such as Sup35p (Alberti et al, 2009; Wang et al, 2018). PLDs are modular sequences of defined composition. Thus, it has been possible to develop algorithms that predict the prion‐propensity of proteins based on their amino‐acid sequence, even though not all the predicted proteins can actually turn into prions, suggesting that the position of PLDs within the protein sequence plays an important role (Alberti et al, 2009; Chiesa et al, 2020). Importantly, PLDs govern the ability of these proteins to interact with other similar proteins and phase‐separate, and have been shown to critically drive stress granule (SG) formation (Murakami et al, 2015; Elbaum‐Garfinkle & Brangwynne, 2015).

SGs are membrane‐less organelles composed of mRNA and RNA‐binding proteins (RBPs), including translation initiation factors (Fig 3). These cytoplasmic ribonucleoprotein (RNP) complexes stall the protein translation machinery while protecting mRNAs from degradation during harmful conditions, hence their appearance upon cellular stress (Protter & Parker, 2016). After phosphorylation of eIF2α and polysome displacement, released mRNAs engage a number of interactions with the Ras GTPase‐activating protein‐binding protein (G3BP), which undergoes a conformational change that sustains the coalescence of G3BP proteins and nucleates SG formation (Guillén‐Boixet et al, 2020). These condensates recruit additional client proteins via microtubule‐associated transport, and protein–protein interactions promote further clustering into macroscopically visible SGs of up to 5 μm in diameter (Nadezhdina et al, 2010; Leung, 2016). Upon termination of sublethal stress, SGs are dismantled by molecular chaperones and polysomes are re‐established, reactivating translation (Mateju et al, 2017). Due to their dynamic makeup, SGs have been described as liquid‐like compartments originating from liquid–liquid phase separation (Hyman et al, 2014; Harrison & Shorter, 2017). Indeed, SGs form when RNPs establish energetically favorable intramolecular interactions, thus demixing from the cytoplasm while remaining in a liquid state. Molecules within SGs are motile and in constant exchange with the surrounding environment, as attested by FRAP experiments showing a rapid signal recovery after photobleaching (Kedersha et al, 2000; Kedersha et al, 2005). However, since liquid–liquid demixing concentrates proteins within discrete subcellular compartments, SGs have been proposed to participate in the nucleation of protein aggregates by facilitating liquid‐to‐solid phase transitions (Patel et al, 2015). Definitive proof for the conversion of SGs into protein aggregates in disease models is still lacking, but supporting evidence for this suggested pathomechanism stems from the identification of SG markers within patient inclusions (Liu‐Yesucevitz et al, 2010; Mackenzie et al, 2017). Additionally, most of the RBPs linked to ALS and other neurodegenerative disorders are recruited to SGs in cell culture models, indicating that the SG may act as “crucibles” of ALS pathogenesis (Fig 3) (Hua & Zhou, 2004; Guil et al, 2006; Colombrita et al, 2009; Vanderweyde et al, 2012; Li et al, 2013; Boeynaems et al, 2017; Marrone et al, 2018). Of note, side by side with SG‐centric studies, it is becoming increasingly evident that protein aggregation may also occur independent of SG formation, as recently shown by a handful of cell‐based works that reported the deposition of FUS and TDP‐43 assemblies outside of the context of SGs (Marrone et al, 2018; Gasset‐Rosa et al, 2019; Mann et al, 2019). While this remains a controversy in the field, it is reasonable to speculate that aggregate formation may occur via both mechanisms.

Aggregates are transmitted from cell to cell by multiple mechanisms

The unique structure and biophysical properties of each protein determine their route of transmission (Agnati & Fuxe, 2014). Mechanisms of protein aggregate transmission are not CNS‐specific, but they are present in most cell types. Even though some of these routes have been described only in vitro cultures of neuronal cell type, they can very likely explain aggregates spread in ALS. Further studies are thus strongly needed to corroborate these first lines of evidence.

Exocytosis/endocytosis

Misfolded and aggregated proteins can be actively secreted by cells or released from dead or dying cells (Fig 4). The conventional protein secretion (CPS) pathway is a process that involves vesicles trafficking from the ER to the Golgi apparatus, and eventually to the plasma membrane. Proteins aimed for secretion are engulfed into vesicles following post‐translational modifications, and subsequently released into the extracellular space upon fusion with the plasma membrane (Emmanouilidou & Vekrellis, 2016; Viotti, 2016; Demaegd et al, 2018). In vitro and in vivo evidence [for comprehensive reviews see (Kennedy & Ehlers, 2011; Bankaitis, 2015; Demaegd et al, 2018)] as well as information gathered from the analysis of human biofluids strongly support the notion that exocytosis of proteins physiologically happen in the human CNS (Kroksveen et al, 2011; Begcevic et al, 2016). Substrates of the CPS pathway are typically monomeric proteins, but further evaluation is needed to understand whether oligomers and fibrils may also be externalized via this route. In the last two decades, unconventional protein secretion pathways have been described for a number of proteins. Unconventional protein secretion pathways bypass the Golgi system to directly traffic the protein to the plasma membrane. Most of these pathways are only partially understood but unconventional receptor‐mediated secretion routes have been described in vitro for some proteins involved in neurodegeneration such as endogenous and pathological Tau (Merezhko et al, 2018) and for misfolded α‐synuclein (Kennedy & Ehlers, 2011). Proteins released in the extracellular space are then taken up by surrounding recipient cells, predominantly via macropinocytosis (Zeineddine et al, 2015; Yerbury, 2016). It has been proposed that the interaction between secreted proteins and the recipient neuronal plasma membrane stimulates the formation of large membrane ruffles that fold back into the cell giving rise to un‐coated vesicles. Due to the lack of coating, these vesicles are easily disrupted by the loaded aggregates, which are released into the cytoplasm of the host cell where they could start seeding further protein aggregation.

Figure 4. Mechanisms of cell‐to‐cell protein transmission.

(A) Overview of protein aggregation and mechanisms of secretion: exocytosis, extracellular vesicles (EVs, including ectosomes and exosomes), tunneling nanotubes (TNTs). (B) Protein release from a dying cell. Extracellular aggregates might be the result of active secretion from living cells or passive secretion from dying cells. (C) Secreted proteins are internalized via different routes, producing serious downstream effects in recipient cells. Acquired misfolded proteins/oligomers/aggregates may induce cross‐seeding by interacting with newly synthetized proteins, hamper the degradation systems and/or hijack physiological cellular processes. (D) Circulating misfolded proteins/oligomers/aggregates can be exploited as potential biomarkers. Proteins released in the extracellular space by neurons and glial cells are collected in the interstitial fluid (ISF) and reach the bloodstream, where they can be detected and measured.

Tunneling nanotubes

Tunneling nanotubes (TNTs) are a type of wiring transmission (Agnati & Fuxe, 2014), whereby intercellular communication occurs via a well‐defined intervening structure (Fig 4). Their existence has been demonstrated in vitro using multiple cell types including neurons, astrocytes and microglia. Their existence and relevance in vivo have been demonstrated in CNS‐unrelated cell types (such as myeloid cells as reviewed in (Dupont et al, 2018)) while their existence in the human CNS is still under investigation. TNTs are thin tubes (50‐200 nm in diameter) that directly connect the cytoplasm of two distinct cells granting the exchange of soluble factors, proteins and organelles. Interestingly, TNTs are transient structures, whose lifespan ranges from 30 to 60 min to several hours (Gurke et al, 2008; Davis & Sowinski, 2008). Their formation typically relies on actin polymerization, which initially induces the formation of protrusions. TNTs maturity is then achieved via the connection of protrusions deriving from two different cells, and the shuttling of cargo remains unidirectional and actin‐mediated (Gurke et al, 2008). The biological functions of TNTs in the CNS are still largely unknown. Neuron–neuron connections by TNTs have been shown to enable the exchange of receptors, while astrocyte–astrocyte connections have proven the exchange of mitochondria (Valadi et al, 2007; Rostami et al, 2017). There is also discrete evidence of neuron‐astrocyte TNTs (Sun et al, 2012). For their role, TNTs have been investigated in neurodegenerative diseases as a mean of toxic proteins transmission, especially in the case of tau, α‐synuclein (which is already part of the TNT structure), huntingtin, and PrP (Gousset et al, 2009; Tardivel et al, 2016; Abounit et al, 2016). Even though no study has investigated the involvement of these structures in the transmission of proteins in the context of ALS, TNTs undoubtedly represent an intriguing pathway.

Extracellular vesicles

Extracellular vesicles (EVs), initially referred to as exosomes, were originally defined as “platelet dust” following identification of microparticles driving blood coagulation (Johnstone et al, 1989). To date, the term EVs identifies a broad class of secreted double‐membrane organelles heterogeneous in size, composition, biogenesis, and function. The protein, lipid, and nucleic acid content of EVs reflects that of the cells of origin, including the pathophysiological variations occurring during their lifespan. Based on their biogenesis, EVs can be classified into two categories: (i) ectosomes and (ii) exosomes. Ectosomes originate by external budding of the plasma membrane. Their size ranges from 50 nm to 1–2 μm, and organelles belonging to this class include microparticles, microvesicles, and large vesicles (Théry et al, 2018; Hessvik & Llorente, 2018; Mathieu et al, 2019; Kalluri & LeBleu, 2020). Exosomes are generated by the fusion of the multivesicular bodies (MVBs) with the plasma membrane (Fig 4). In turn, MVBs derive from early sorting endosomes (ESEs) which stem from inward invagination of the plasma membrane. ESEs develop into the late sorting endosome (LSE) after processing in the ER and Golgi apparatus. Eventually, inward budding of the LSE membrane generates intraluminal vesicles (ILVs), which are the characteristic feature of mature MVBs. When MVBs do not fuse with autophagosomes/lysosomes for autophagic degradation, they are released in the extracellular space in the form of exosomes (Raposo & Stoorvogel, 2013; Villarroya‐Beltri et al, 2016; Record et al, 2018; Latifkar et al, 2019). Four different complexes, named “endosomal sorting complexes required for transport machinery” (ESCRT) 0‐I‐II‐III typically orchestrate the key steps of exosomes biogenesis (Wollert et al, 2009; Hurley & Hanson, 2010; Hanson & Cashikar, 2012). However, EVs biogenesis is also known to occur in an ESCRT‐independent pathway that primarily relies on sphingomyelinase 2, implying an important contribution of lipid rafts in the formation of EVs (Trajkovic et al, 2008; Stuffers et al, 2009). Following release, the distinction between exosomes and ectosomes is blurry, and the mechanisms of uptake by recipient cells are shared. Uptake is a rapid and active process that occurs through a plethora of unspecific mechanisms, including clathrin‐mediated endocytosis, macropinocytosis, and phagocytosis. As inhibition of each of these pathways reduces but not completely abates the uptake of EVs, these mechanisms are speculated to come in place simultaneously. Interestingly, EVs may also enter cells by direct fusion with the plasma membrane following receptor interaction, a process in which the EV content is directly released into the cytoplasm (Parolini et al, 2009).

In the CNS, EVs are produced by all cell types and mediate a number of biological processes under both physiological and pathological conditions (Delpech et al, 2019; Blanchette & Rodal, 2020). In vitro and in vivo experiments have shown that EV production is a critical pathway for CNS cells to communicate and modify each other’s fate (Yang et al, 2018). Astrocyte‐produced EVs have been shown to modulate neuronal firing, microglia activation, and maturation of oligodendrocyte progenitors into oligodendrocytes (Lombardi et al, 2019). Oligodendrocyte‐produced EVs guarantee neuronal health (Frühbeis et al, 2013; Frühbeis et al, 2019). In conditions of neuroinflammation, a process which is relevant to ALS and other neurodegenerative disorders, microglia‐produced EVs exert detrimental effects on neurons and prime astrocyte reactivity (Krämer‐Albers et al, 2007; Gosselin et al, 2013; You et al, 2020). Both animals and humans EVs produced in the CNS can be found in the cerebrospinal fluid (CSF) and surprisingly also in the blood. It has been shown that EVs can cross the blood brain barrier (BBB) (Saint‐Pol et al, 2020). Being an effective means of transmission between cells, the shedding of EVs has been proposed as a major route of transcellular spreading of misfolded proteins and small aggregates. For example, Aβ peptide, α‐synuclein, and several ALS‐associated proteins have been observed in EVs extracted from patient biofluids (Iguchi et al, 2016; Lee et al, 2019; Yuan et al, 2019). Importantly, different studies have shown that EV secretion participates in the maintenance of cellular homeostasis, with MVBs being central factors in the interplay between EV production and autophagy. MVBs have short half‐lives as they generally readily fuse with lysosomes. However, under specific conditions, they may fuse with the plasma membrane. Modulation of the fate of MVBs occurs at multiple levels: post‐translational modifications, such as ISGylation, control EVs release (Villarroya‐Beltri et al, 2016; Desdín‐Micó & Mittelbrunn, 2017). Additionally, proteins belonging to the kinesin family participate in coordinating the distribution of MVBs toward either the lysosomes or the plasma membrane (Heisler et al, 2018). It has been shown that upon inhibition of MVB fusion with lysosomes, pathological proteins were release via EVs at a higher rate (Poehler et al, 2014; Hessvik et al, 2016; Minakaki et al, 2018). On the other hand, inhibition of autophagosome formation via 3‐MA caused a reduction in the number of released EVs as a consequence of the hampered formation of MVBs, suggesting shared mechanisms of biogenesis (Wang et al, 2019). Pharmacological stimulation of autophagy led to lower rates of EV externalization, while inhibition of autophagy by misfolded proteins induced increased cellular levels of toxic proteins, urging the EV secretory pathway to be recruited as the last resort to counteract the proteotoxic stress (Fussi et al, 2018; Abdulrahman et al, 2018). Interestingly, EV release was also found to be facilitated by Golgi fragmentation after prolonged proteasome/autophagy inhibition (Yang et al, 2017). Given this evidence, elucidation of the mechanisms underlying the loading of pathogenic proteins into EVs is critical. However, a thorough understanding of these processes is further complicated by the technical challenges linked to EV research. For instance, resolving EV content and size at a single‐particle resolution is still technically unfeasible. Additionally, different EV classes share multiple markers, which makes them undistinguishable after release (Lotvall et al, 2014; Yáñez‐Mó et al, 2015). Several unanswered questions remain to be addressed in the context of the spatiotemporal spread of protein aggregates. For instance, do aggregate‐containing EVs derive from a specific cell population? And what is the interplay between the presence of mutated misfolded proteins and EVs production and secretion? Future experiments and technical advances will improve our understanding of these processes.

Aggregation, transmission, and propagation of ALS‐linked pathogenic proteins

Superoxide dismutase (SOD1)

The human SOD1 gene is located on chromosome 21q22.11 and encodes for a 153‐amino acid polypeptide of 16 kDa in weight (Fig 5) (Milani et al, 2011). During translation, SOD1 monomers are partially folded. The definitive three‐dimensional conformation is reached upon binding the metal cofactors zinc and copper, as well as associating with a second monomer and forming intramolecular disulfide bonds (Milani et al, 2011). Thus, the functional Cu/Zn SOD1 enzyme is a 32 kDa homodimeric protein involved in scavenging the free radicals naturally produced by cellular respiration and other processes. SOD1 has been described as one of the most stable proteins ever observed (Rakhit & Chakrabartty, 2006). For instance, the complete protein was shown to retain enzymatic activity in the presence of highly denaturing conditions, such as 10 M urea and 4% SDS (Forman & Fridovich, 1973), as well as to resist to melting temperatures of up to 100°C thanks to the contribution of metal ions (Lepock et al, 1985). However, Chia and colleagues demonstrated that wild type SOD1 recombinant protein can rapidly aggregate in conditions that disrupt the stability of the dimer, such as low pH and increasing concentrations of guanidine hydrochloride (Chia et al, 2010). A number of ALS causative SOD1 mutants, including G93A, G37R, A4V, and G85R, were shown to spontaneously fibrillize with much shorter lag times than the wild‐type protein. Aggregates enriched in β‐sheet structures are known to act as templates to convert normal proteins into their misfolded counterpart. Interestingly, electron microscopy of SOD1 aggregates revealed a predominantly amyloid‐like morphology (Chia et al, 2010; Shvil et al, 2018). Because all the available protein was eventually converted into fibrils, the group demonstrated the ability of fibrils to trigger fibrillization in an autocatalytic manner after seeding. In addition, pre‐formed fibrils of mutant SOD1 could cross‐seed the wild type SOD1 protein, suggesting the possibility of template‐dependent propagation. Last but not least, the group tested spinal cord homogenates from SOD1^G93A mice using in vitro fibrillization assays and recapitulated previous observations (Chia et al, 2010).

Figure 5. Table illustrating domain composition of the ALS proteins discussed in this review.

Major protein domains are color‐coded. Numbers indicate amino acid residues at their boundaries. For each protein, primary biological functions are indicated. QGSY = glutamine‐glycine‐serine‐tyrosine‐rich motif, G‐rich = glycine‐rich, RRM = RNA recognition motif, RGG = arginine‐glycine‐glycine‐rich motif, ZnF = zinc finger domain, NLS = nuclear localization signal, DPRs = dipeptide protein repeats.

SOD1 mutations have been suggested to cause structural instability, leading to the collapse of the homodimeric complex and its subsequent aggregation (Cleveland & Liu, 2000; de Araújo Brasil et al, 2019). For instance, decreased affinity of mutant SOD1 for copper/zinc ions, as well as reduction of disulfide bonds in fALS have been suggested to contribute to the formation of aggregation‐prone SOD1 species (Tiwari & Hayward, 2003). In cell culture models, aggregates of mutant SOD1 recombinant protein could induce misfolding and aggregation of the corresponding wild type native SOD1 protein following uptake from the culture medium, which may explain the identification of SOD1 mutant/wild‐type co‐aggregated states in fALS patients (Bruijn et al, 1998; Münch et al, 2011; Furukawa et al, 2013). Interestingly, the templated conversion of human wild type SOD1 to a misfolded state has been linked to the exposure and aberrant oxidation of the tryptophan residue at position 32 (W32) (Taylor et al, 2007b; Grad et al, 2011). Replacement of W32 with the evolutionary conserved and less prone to oxidation serine (found, for instance in mice), markedly diminished conversion (DuVal et al, 2019). Using a number of SOD1‐reporter lines, the Cashman laboratory confirmed the induction of reporter protein aggregation following exposure of cells to human spinal cord homogenates from mutant SOD1‐fALS and sALS patients (Pokrishevsky et al, 2017). Misfolded SOD1 deposits were shown to trigger ER stress and impair proteasome function, thus hampering the ability of cells to eliminate misfolded proteins and triggering a vicious cycle of protein accumulation and aggregation (Kikuchi et al, 2006; Nishitoh et al, 2008; Bendotti et al, 2012). Additional studies have explored the mechanisms of cell‐to‐cell transfer of SOD1 protein aggregates, highlighting numerous possible routes. Several groups suggested that mutant SOD1 may transit between cells via two non‐mutually exclusive mechanisms, including the release of protein aggregates from dying cells followed by micropinocytosis, and the exchange of exosomal vesicles between living cells (Münch et al, 2011; Grad et al, 2014; Zeineddine et al, 2015; Benkler et al, 2018). The transfer of aggregates did not seem to require cell‐to‐cell contacts, because incubation with conditioned medium from cells bearing aggregated SOD1 efficiently transferred pathology to healthy cells, and transfer efficiency drastically decreased after dilution of the conditioned medium (Münch et al, 2011).

In line with the discussed central role of protein quality control in preventing pathology, overexpression and accumulation of human wild‐type SOD1 led to aggregation, neurotoxicity, and reduced lifespan in transgenic mice (Graffmo et al, 2013). In this study, human wild‐type SOD1 was expressed at levels comparable to human variants carrying the G93A mutation. However, the first evidence of in vivo SOD1 transmission was reported by Ayers and colleagues in 2014 using a different mutant (Ayers et al, 2014). The group generated heterozygous SOD1^G85R‐YFP mice, which did not exhibit any ALS symptoms before 20 months of age. Yet, when newborn mice were challenged with intraspinal homogenates from terminally ill mutant mice, they were shown to develop rapid MN pathology concomitant with the appearance of detergent‐insoluble aggregates of endogenous SOD1 protein. In a follow‐up study centered on a single unilateral injection into the sciatic nerve of test mice, inclusion pathology was shown to spread in a step‐wise process, moving from the ipsilateral dorsal root ganglion to the lumbar spinal cord, and further proceeding rostrally toward the cervical spine (Ayers et al, 2016). Inclusions were eventually observed in the brainstem and midbrain neurons that synapsed with the affected spinal MN, suggesting trans‐synaptic propagation of protein pathology. More recently, the Marklund laboratory reported that transgenic mice expressing human mutant SOD1 protein may generate different strains of SOD1 aggregates, and that injection of such strains into the lumbar spinal cord of asymptomatic animals could induce templated SOD1 aggregation and rostrally spreading pathology (Bergh et al, 2015). In a subsequent study, they explored the prion‐like properties of human‐derived aggregates by delivering autopsy material from SOD1^G127X patients into the spinal cord of mice expressing a human SOD1 transgene (Ekhtiari Bidhendi et al, 2018). Of note, inoculations led to spreading aggregation resulting in fatal ALS‐like pathology, demonstrating for the first time that human SOD1 has prion‐like properties.

Human post‐mortem tissue has shown the presence of aggregated SOD1 in the glia. Using an antibody raised against misfolded SOD1 (misSOD1), Forsberg and Andersen (2011) showed that patients homozygous for D90A‐SOD1 presented misSOD1 aggregates in astrocytes, microglia, and oligodendrocytes. Glial cells have been suggested to contribute to ALS pathogenesis by increasing neuronal susceptibility to a variety of potential insults. For instance, transplanting SOD1^G93A astrocyte precursors into the spinal cord of wild‐type mice led to local MN loss accompanied by mild motor deficits (Papadeas et al, 2011). Interestingly, engrafted SOD1^G93A astrocytes were shown to induce the appearance of ubiquitin‐positive aggregates selectively in the spatially interacting MNs, suggesting that mutant astrocytes may release toxic factors triggering neuronal pathology. The nature of such factors is still the subject of extensive studies. However, toxic factors may include aggregates of misfolded proteins. Evidence from cell culture models suggested that mutant SOD1 highjacks the secretory pathways by reducing the amount of secreted proteins except for the specific enrichment of mutant SOD1 (Fig 6) (Basso et al, 2013). The same study also observed that astrocyte‐derived extracellular vesicles (exosomes) could effectively deliver mutant SOD1 to MNs and cause their death. One more recent work confirmed that astrocytes and neurons are the main source of misfolded SOD1 released in EVs (Silverman et al, 2019). The release of SOD1 in the extracellular space appears to be a physiological process (Ogawa & Furukawa, 2014), as attested by the presence of both wild‐type and mutant SOD1 in the CSF of healthy individuals and fALS patients (Zetterstöm et al, 2011). However, the pathological accumulation of mutant protein in the cytoplasm may require increased release of mutant SOD1 as a strategy to safeguard intracellular homeostasis. In fact, it was shown that impaired secretion of mutant SOD1 in NSC‐34 cells caused the formation of intracellular aggregates followed by cytotoxicity (Turner et al, 2005), suggesting that active secretion of mutant SOD1 is beneficial for cell survival. Yet, mutant SOD1 is secreted in conjunction with chromogranins (Urushitani et al, 2006), a known trigger of microgliosis, and misfolded SOD1 has been described to itself activate microglia through CD14, toll‐like receptor 2 (TLR2) and TLR4, causing neurotoxicity via release of reactive oxygen species and proinflammatory cytokines (Liu & Wang, 2017). Taken together, these studies identify mutant SOD1 secretion as a double‐edged sword in ALS pathogenesis.

Figure 6. Table illustrating the different routes of extracellular secretion related to the main ALS proteins.

Investigated disease models are reported and referenced. Physiological/wild‐type protein; Pathological/mutated protein.

Fused in sarcoma (FUS)

The FUS gene is located on chromosome 16p11.2 and encodes a 526‐amino acid protein of 75 kDa with DNA/RNA‐binding ability. FUS shuttles between nucleus and cytoplasm to fulfill a multiplicity of tasks, including DNA damage response, transcriptional activation, splicing, mRNA export, and translation initiation (Deng et al, 2014). FUS is highly aggregation‐prone, and this strong aggregation propensity is encoded in its protein sequence. In fact, in silico analysis of the FUS secondary structure has revealed that, except for two α‐helical regions at residues 295–305 and 342–352, the FUS protein is intrinsically disordered across its entire length (Fig 5) (Marrone et al, 2020). Recent studies have demonstrated that the interaction with RNA molecules (Maharana et al, 2018), other RBPs (Marrone et al, 2019), molecular chaperones as well as post‐translational modifications (Monahan et al, 2017; Qamar et al, 2018; Hofweber et al, 2018; Ding et al, 2020), are among the main factors contributing to the cellular solubility of FUS. Thus, perturbations in these interactions underlie the formation of FUS biomolecular condensates. The main driving force for the phase separation and liquid‐to‐solid phase transition of FUS is its N‐terminal glutamine‐glycine‐serine‐tyrosine (QGSY) PLD IDR LCD (Gitler & Shorter, 2011; Dunker et al, 2013; Mier et al, 2020). Previous works have shown that replacing Y in the FUS LCD with non‐aromatic residues disrupted the ability of FUS to phase separate, implicating Y residues in the liquid–liquid demixing of FUS (Lin, Currie, & Rosen, 2017). More recently, Q residues have been suggested as another major contributor for the establishment of intermolecular contacts and the subsequent induction of partitioning (Murthy et al, 2019). Studies with purified recombinant proteins have shown that under conditions of low salt concentrations, full‐length wild‐type FUS can self‐assemble and give rise to liquid‐like condensates of high sphericity and low stiffness (Qamar et al, 2018). This suggested that ALS‐mutations may accelerate the kinetics of phase separation and liquid‐to‐solid phase transition. For instance, Patel et al (2015) found the G156E mutation to encourage the conversion of liquid FUS droplets into an aggregated state, as demonstrated by the formation of sea urchin‐like structures after prolonged incubation. Similar observations were made by Nomura and colleagues, who additionally found FUS^G156E aggregates to exhibit fibrillar morphology and positivity for amyloid marker thioflavin T (Nomura et al, 2014). Importantly, when the group mixed a small amount of FUS^G156E recombinant protein with wild‐type FUS, this seeded the aggregation of the wild‐type protein in insoluble pellets, suggesting the ability of aggregation‐prone mutant FUS to self‐propagate (Nomura et al, 2014). G156E disturbs the N‐terminal LCD of FUS, but the majority of FUS‐linked ALS mutations affect its C‐terminus, with residues 510–526 (Shang & Huang, 2016) being the most frequently mutated. These mutations have been found to impair the interaction between the proline‐tyrosine nuclear localization signal (PY‐NLS) of FUS and its nuclear import receptor (transportin‐1), which has been associated with the cytoplasmic mislocalization of this predominantly nuclear protein. Indeed, binding affinity studies have shown that FUS NLS mutations exhibit variably reduced affinity for transportin‐1, which was shown to correlate with the degree of cytoplasmic accumulation as well as with disease onset and progression (Niu et al, 2012). For these reasons, FUS C‐terminal mutations are believed to induce aggregation by supersaturation rather than by altering the biophysical properties of FUS. However, we recently found that at least one C‐terminal mutations, P525L, directly affects the solubility of FUS, increasing the aggregation tendency of the protein independent of its subcellular localization (Marrone et al, 2020).

Mutations affecting the FUS C‐terminus are among the most studied. In cell culture experiments, FUS C‐terminal mutations have been shown to increase the recruitment of FUS to SGs upon stress (Vance et al, 2013; Lenzi et al, 2015). Cytoplasmic mislocalization has also been linked to an impairment in the protein quality control systems, as demonstrated by increased p62 levels (Marrone et al, 2019). Drugs stimulating the clearance of aberrantly accumulated cytoplasmic FUS reduced SG recruitment (Marrone et al, 2018), as well as restoring disrupted proteostasis (Marrone et al, 2019), suggesting that increased cytoplasmic FUS levels are a crucial trigger of disease pathogenesis. Importantly, the presence of FUS cytoplasmic inclusions in spinal MNs of ALS patients carrying FUS NLS mutations correlate with a simultaneous reduction in nuclear FUS levels (Vance et al, 2013). Based on this observation, Vance and colleagues hypothesized that mutant FUS protein may seed the aggregation of wild‐type FUS. To explore this scenario, they co‐transfected cells with differentially tagged wild‐type and mutant FUS‐expressing plasmids and found that wild‐type FUS localized in both nucleus and cytoplasm upon co‐expression, suggesting that mutant FUS may physically sequester its wild‐type counterpart. Immunoprecipitation confirmed the interaction, which was not dependent on RNA‐binding, but rather resulted from the physical interplay between proteins (Vance et al, 2013). We recently showed that FUS pathology is linked to alterations in multiple RBPs and that FUS aggregates in patient specimens occasionally colocalize with other ALS‐associated proteins, such as heterogeneous nuclear ribonuclear protein A1 (hnRNPA1), which are simultaneously depleted from the nucleus (Marrone et al, 2019). We speculate that these events may be linked to the ability of aggregated FUS to template the conversion of other RBPs harboring PLDs. However, the propagation of FUS aggregates, both in vitro and in vivo, is largely unexplored and further experiments are needed to elucidate the dynamics of FUS pathology and its spread. Because evidence is still lacking, there are not enough elements to provide a detailed description of the mechanisms of cell‐to‐cell transmission for FUS. The reasons for the absence of research on FUS may be linked to its relatively recent discovery as well as the limited number of ALS cases associated with FUS mutations. However, FUS is an excellent example of prion‐like protein, and particularly in vivo studies may offer extremely valuable evidence for the prion‐like potentials of ALS.

TAR DNA‐binding protein (TDP‐43)

TDP‐43 pathology is the most common feature of ALS/FTD, in addition to being present in some cases of AD and PD (Arai et al, 2006; Neumann et al, 2006; McAleese et al, 2017; Matej et al, 2019). The TDP‐43 encoding gene is located on chromosome 1p36.2, and its protein product belongs to the family of hnRNPs, whose role in ALS and FTD is detailed in (Bampton et al, 2020). Proteins of these family mediate a number of critical functions in the nuclear and cytoplasmic compartments, including DNA repair and replication, transcription, pre‐mRNA splicing, mRNA nuclear export and retention, mRNA stability, and translation (Buratti & Baralle, 2001; Buratti et al, 2010; Buratti & Baralle, 2010). As other RBPs, TDP‐43 encompasses an RNA‐binding domain and a C‐terminal LCD (Fig 5) (Kuo et al, 2009; Wang et al, 2018; Franzmann & Alberti, 2019). However, TDP‐43 additionally exhibits a small N‐terminal domain (NTD, 1–80) that mediates TDP‐43 physiological oligomerization, which is essential for the implementation of its functions in the context of RNA splicing (Afroz et al, 2017). Downstream of the NTD, two RNA‐recognition motifs (RRMs) (80‐278 a.a.) recognize UG‐repeats within RNA strands in a sequence and length specific manner. In fact, mutations in this motifs result in a loss of RNA‐binding capacity (Buratti & Baralle, 2001; Kuo et al, 2009; Polymenidou et al, 2011; Chen et al, 2019). By binding to RNA, TDP‐43 modulates the splicing of numerous genes, such as FUS and TDP‐43 itself, which is particularly interesting since it allows TDP‐43 to repress its own transcription and modulate its own protein levels (Ayala et al, 2011; Polymenidou et al, 2011; Bhardwaj et al, 2013). The TDP‐43 LCD (278–414 a.a.) is constituted by tandem QN‐rich sequences, which confer TDP‐43 prion‐like properties (Furukawa et al, 2011; Budini et al, 2012; Mompeán et al, 2014) and the ability to form SGs by phase separation (Li et al, 2018; Babinchak et al, 2019). Loss of the LCD is associated with profound alterations in the cellular transcriptome, implying that TDP‐43 nuclear functions are also critically dependent on this domain. In 2006, two parallel studies identified TDP‐43 as the major component of ubiquitin‐positive, tau‐negative, and synuclein‐negative inclusions in the CNS of individuals affected by ALS/FTD (Arai et al, 2006; Neumann et al, 2006). TDP‐43 was also hyperphosphorylated (p409/410) and aberrantly cleaved. Interestingly, inclusions of pTDP‐43 were additionally detected in the cytoplasm of glial cells (pTDP‐43 immunoreactive glial cytoplasmic inclusions were found only in oligodendrocytes, while other glial cells were negative (Nishihira et al, 2008; Takeuchi et al, 2016) as well as in different muscle tissues (Sorarú et al, 2010; Cykowski et al, 2018). The nuclei of cells presenting inclusions were depleted of TDP‐43. ALS‐causing mutations in TDP‐43 primarily affect the C‐terminal domain (Kabashi et al, 2008; Pesiridis et al, 2009; Buratti, 2015; Kapeli et al, 2017), and TDP‐43 pathology has been extensively characterized (Hasegawa et al, 2008; Zhang et al, 2009; Lee et al, 2012; Cykowski et al, 2017) and linked to both a loss of function subsequent to TDP‐43 nuclear depletion (Iguchi et al, 2013; Schmid et al, 2013; Skoko et al, 2016; Koza et al, 2019; Melamed et al, 2019) and a toxic gain of function following its aggregation in the cytoplasm (Johnson et al, 2008; Couthouisa et al, 2011; Figley & Gitler, 2013; Crippa et al, 2016; Cicardi et al, 2018; Guo et al, 2018a).

In vitro experiments have shown that both the N‐terminal and C‐terminal domains of TDP‐43 are essential for its phase separation. McGurk and colleagues demonstrated that the liquid demixing of TDP‐43 is driven by the interaction between poly‐ADP ribose and the NTD of TDP‐43, whose deletion completely abates coalescence (McGurk et al, 2018; Wang et al, 2020). A study by the Fawzi lab suggested that the interplay between C‐terminal α‐helices of two or more molecules of TDP‐43 acts as an additional driver of TDP‐43 phase separation. In fact, mutations in this domain caused the rapid formation of less dynamic droplets, suggesting a liquid‐to‐solid phase transition, while the C‐terminal fragments alone immediately underwent aggregation, assembling in amorphous structures or fibrils (Johnson et al, 2009; Conicella et al, 2020). TDP‐43 aggregation may also result from the disruption of cellular proteostasis irrespective of the presence of pathological mutations. For instance, pharmacological inhibition of the proteasome was demonstrated to induce the cytoplasmic accumulation of wild type TDP‐43 and its C‐terminal fragments (Scotter et al, 2014; Ishii et al, 2017). Similarly, we demonstrated that inhibiting autophagy overloads cells with aggregates, while stimulating autophagy facilitates the clearance of preformed inclusions (Scotter et al, 2014; Crippa et al, 2016; Cicardi et al, 2018). Interestingly, cryo‐EM analysis showed that TDP‐43 is subject to polymorphisms and can deposit in fibrils of different shapes (Cao et al, 2019). Accordingly, TDP‐43 aggregates exhibit conformational dissimilarities across patients, confirming that TDP‐43 may seed different strains (Laferrière et al, 2018; Porta et al, 2018). Consistent with a potential prion‐like role of TDP‐43, aggregates extracted from the frontal cortex of TDP‐43 FTD patients have been shown to initiate pathology (mislocalization, aggregation, hyperphosphorylation) both in vitro and in vivo (Nonaka et al, 2013; Porta et al, 2018).

The temporal and spatial distribution of TDP‐43 lesions across the CNS (Jamshidi et al, 2020) as well as the presence of TDP‐43 in the CSF (Steinacker et al, 2008; Kasai et al, 2009; Noto et al, 2011) (both monomeric and aggregated forms) have sparked investigations into the mechanisms behind the release and uptake of TDP‐43. In biofluids, TDP‐43 was found both as a free protein and incorporated into the EVs, suggesting the existence of different mechanisms of secretion, including the possibility that these proteins may be released by the breakdown of dead cells (Feneberg et al, 2014; Sproviero et al, 2018; Feneberg et al, 2018; Kasai et al, 2019). Similarly, Ishii et al, showed TDP‐43 aggregation and transmission in cultured cells through externalization and consequent uptake of TDP‐43 aggregates Ishii et al (2017). Irrespective of the secretion mechanisms, multiple studies have used brain homogenates or patient CSF extracts to investigate the modalities of cellular uptake. Newly internalized TDP‐43 aggregates had the ability to recruit intracellular proteins and template the formation of new inclusions in neurons (Porta et al, 2018). An additional study demonstrated that living neuronal and non‐neuronal cells actively secrete soluble and oligomeric TDP‐43, a fraction of which is enclosed into EVs (Fig 6). EV‐enclosed TDP‐43 was internalized by recipient cells more readily than the fraction secreted as free protein. Also, TDP‐43 was taken up at the neuronal synapse followed by axonal transport (Feiler et al, 2015). Additionally, a zebrafish model showed that TDP‐43 axonal spreading is exacerbated by impaired microglia (Svahn et al, 2018). One study evaluated the impact of modulating EV production in a TDP‐43 mutant mouse model (TDP‐43^A315T). Interestingly, only primary neuron‐, but not microglia‐ and astrocyte‐derived EVs contained TDP‐43. The A315T C‐terminal mutation did not cause increased release of TDP‐43 within EVs, but induced a general increment in the number of vesicles produced, suggesting a role for misfolded/aggregated TDP‐43 on EV biogenesis/release. In mice bearing the A315T mutation, hampering the release of EVs both genetically and pharmacologically led to the premature insurgence of motor and behavioral symptoms, suggesting that EV release is critical to alleviate neurons from the burden of misfolded proteins (Iguchi et al, 2016). Another study interestingly showed that CSF‐derived EVs from ALS and ALS/FTD patients contain TDP‐43 C‐terminal fragments (Ding et al, 2015). A recent study provided the first proof of concept that CSF is a route of pathology spreading in ALS; they observed that human ALS CSF infused in the CNS of mice overexpressing human TDP‐43 induce TDP‐43 pathology onset accompanied by decreased cognitive functions (Mishra et al, 2020).

Because the surveilling microglia phagocyte debris, it is plausible that neuronal death may activate the neighboring microglia causing it to engulf the released aggregates. Along these lines, Leal‐Lasarte et al (2017) observed an overactivation of the inflammatory machinery resulting in the copious production of pro‐inflammatory cytokines upon exposure to TDP‐43 aggregates. In vitro systems have also demonstrated TDP‐43 aggregate transmission from neurons to astrocytes; For instance, Smethurst et al (2020) showed that TDP‐43 aggregates could be exchanged between iPSC‐derived MNs and astrocytes and that astrocytes exhibited reduced vulnerability to aggregated TDP‐43 compared with MNs. They speculated that, at least in the first stages of the disease, astrocytes may exert a protective role on MNs by reducing TDP‐43 cytoplasmic mislocalization and aggregation. Yet, another work demonstrated that astrocytes with TDP‐43 aggregation exhibited dysregulated metabolism and induced neuronal death due to lack of trophic support (Smethurst et al, 2020; Velebit et al, 2020). To sum up, current evidence on TDP‐43 cell‐to‐cell transmission strongly highlights the importance of these pathways on disease progression. However, a more complete understanding of the modalities of TDP‐43 transmission is still required.

Chromosome 9 open reading frame 72 dipeptide repeats (C9orf72 DPRs)

The identification of mutations in the C9orf72 gene by two independent groups was a groundbreaking discovery (Renton et al, 2011; Dejesus‐hernandez et al, 2011). This mutation is responsible for ~11% of all ALS cases and ~13% of all FTD cases (Abramzon et al, 2020). In rare instances, C9orf72 mutations are also the cause of other neurodegenerative diseases such as AD, PD, and CJD (Jiao et al, 2013; Geut et al, 2019). The name refers to the chromosomal location of the gene (9p21, open reading frame 72), and the disease‐causative mutation is a GGGGCC hexanucleotide (G₄C₂) repeat expansion (C9‐HRE). C9‐HRE ranges between 3 and 30 repetitions in healthy individuals, while it can grow to hundreds or even thousands of repeats in patients. Repeat length has been suggested to correlate with age of disease onset and severity, but the technical challenges linked to determining the HRE length and the heterogeneity of the disease have made it cumbersome to establish a univocal correlation (Van Mossevelde et al, 2017).

Three major mechanisms have been linked to C9‐HRE: (i) C9orf72 protein haploinsufficiency, (ii) the formation of RNA foci, and (iii) the synthesis of DPRs by the aberrant translation of the HRE. C9orf72 transcript exists in three variants. In variant 1 and 3, the HRE is located in the intronic region between two alternatively spliced exons. In variant 2, the HRE affects the promoter (Balendra & Isaacs, 2018). Early reports showed that instability of variant 2 transcripts was linked to the formation of G‐quadruplex secondary structures and R‐loops, resulting in C9orf72 protein haploinsufficiency (Gijselinck et al, 2012; Fratta et al, 2012; Reddy et al, 2013). The exact role of this protein is still a matter of debate. The C9orf72 gene is composed of 11 exons (Fig 5); C9orf72 transcript is translated in both a long (55 kDa) and a short (25 kDa) isoform and has been proposed to act as a GDP‐GTP exchange factor for Rab‐GTPases proteins involved in vesicle trafficking and autophagosome formation (Liang et al, 1997; Zhang et al, 2018a; Shao et al, 2020; Tang et al, 2020). Studies conducted in vitro and in different in vivo models showed that C9orf72 haploinsufficiency causes dysfunction and aggregation of lysosomes, Golgi defects, and endosomal impairment (Farg et al, 2014; Corrionero & Horvitz, 2018; Shi et al, 2018). At a systemic level, haploinsufficiency has been associated with an autoimmune/inflammatory phenotype without motor dysfunctions in mice (O’Rourke et al, 2016; Burberry et al, 2016; Atanasio et al, 2016; Sudria‐Lopez et al, 2016). In caenorhabditis elegans and zebrafish models C9orf72 depletion results in motor deficits, but further studies are needed to better define whether the observed phenotype is due to compromised neuronal development (Therrien et al, 2013; Ciura et al, 2013). However, in presence of toxic proteins, C9orf72 has been shown to synergistically lead to impaired neuronal development, increased neuronal toxicity, and motor dysfunction (Lagier‐Tourenne et al, 2013; Sellier et al, 2016; Jiang et al, 2016; Zhu et al 2020). RNA foci formation and DPR accumulation are gain of function mechanisms, and their concomitant accumulation is linked to dysfunction in the nuclear pore compartment (Haeusler et al, 2014; Fox & Tibbets, 2015; Freibaum et al, 2016; Jovičić et al, 2016). RNA foci are clusters of HRE‐RNA and proteins that form in the nucleus and cytoplasm of neurons (Burguete et al, 2015; Yu et al, 2015). Astrocytes, microglia, and oligodendrocytes of post‐mortem tissue display RNA foci even though to a lesser extent and more challenging to detect compared with neurons (Lagier‐Tourenne et al, 2013; Mizielinska et al, 2013; Gendron et al, 2013). One study reported the formation of RNA foci in oligodendrocytes derived from C9‐ALS/FTD iPSCs (Livesey et al, 2016). RNA foci have been suggested to act as seeds of protein accumulation by trapping transcription and splicing factors, eventually affecting cell viability (Lee et al, 2013; Donnelly et al, 2014). C9‐HRE RNA can fold into G‐quadruplex secondary structures which have been associated with multiple toxic mechanisms; in neurons they recruit components of the nucleolus causing nucleolar stress and cell death (Haeusler et al, 2014); in drosophila models, the formation of these structures is at the basis of nucleoplasmic transport dysregulation which is a major cause of neurodegeneration (Zhang et al, 2015). The C9orf72 HRE lacks a canonical AUG start codon, but HRE‐containing C9orf72 RNA transcripts are translated via repeat associated non‐AUG translation (RAN‐T), which is an unconventional mode of translation that could occur in eukaryotes under certain conditions (Zu et al, 2010). Being independent of the start codon, RAN‐T recognizes the C9‐HRE in both directions and along the three reading frames, producing five different DPRs: poly‐GA and poly‐GR from the sense strand, poly‐PA and poly‐PR from the antisense strand, poly‐GP from both strands (Gendron et al, 2013; Gitler & Tsuiji, 2016). All DPRs have been detected in post‐mortem brain tissues from C9orf72 patients in neuronal and non‐neuronal cells (Schludi et al, 2015). The scientific community is currently striving to understand whether DPRs play a crucial role in disease besides being a validated pathological hallmark. Many studies have approached this topic trying to modulate RAN‐T. Silencing the small ribosomal subunit RPS25 leads to decrease of RAN‐T and rescues toxic phenotypes in multiple models (Yamada et al, 2019). Importantly, neuronal RAN‐T can be pharmacologically accelerated by using drugs that maintain cells in conditions of prolonged stress, such as excessive glutamate stimulation or ER inhibition (Westergard et al, 2019). These findings suggest that secondary hits may be required to build up aberrant DPRs. DPRs are non‐functional, unstructured proteinaceous products, quite challenging to assimilate to prion‐like molecules because they do not lose any initial conformation, nor do they template the misfolding of a native counterpart. Additionally, DPR‐containing inclusions have shown an amyloid‐like structure in patient samples, but the nature of these inclusions is still matter of debate (Edbauer & Haass, 2016). Nevertheless, understanding the aggregation dynamics of DPRs and their impact on the compartments in which they deposit is of critical importance.

Among the five DPRs, poly‐GR and poly‐PR are the best characterized. They are positively charged due to the presence of the arginine (R) residues, which affects their overall biophysical properties and behavior within cells (Moens et al, 2019). Both (R)‐rich DPRs are neurotoxic, as demonstrated by work conducted in 2014 (Mizielinska et al, 2014). This initial discovery led to the study of the pathways at the basis of the observed neurodegeneration. In 2016, researchers showed that the poly‐GR and poly‐PR interactome was enriched in nuclear and cytoplasmic proteins encompassing an LCD. Poly‐GR and poly‐PR were found in nuclear membrane‐less structures, such as Cajal bodies and nuclear speckles, as well as the nucleolus (Lee et al, 2016). The recruitment of R‐rich DPRs in these organelles disrupted their structure and/or dynamics, resulting in toxicity. For instance, alterations in nucleolar dynamics have been linked to the activation of apoptotic cell death (Freibaum et al, 2016). Nevertheless, further studies are needed to substantiate these connections. Of note, some of the findings obtained in vitro were also replicated in vivo in mouse models as well as in human autopsy material. For example, histopathology of spinal cord sections revealed that overexpressed poly‐PR co‐localized with the nucleolus and other structures positive for euchromatin and heterochromatin markers (Zhang et al, 2019). Poly‐GR and poly‐PR have additionally been shown to accumulate in SGs and modify their dynamics or, even more interestingly, trigger their assembly (Lee et al, 2016; Boeynaems et al, 2017; Hartmann et al, 2018). In particular, poly‐GRs have been observed to colocalize with components of the translation machinery and alter the formation of SGs, a finding that was validated in C9‐ALS patient tissue (Tao et al, 2015; Hartmann et al, 2018; Zhang et al, 2018b; Chew et al, 2019). However, it remains to be determined whether these alterations are directly linked to neuronal dysfunction and death. Poly‐PR and poly‐GR expression in neurons also hampers nuclear import machinery (Hayes et al, 2020) and ribosomal function (Moens et al, 2019). Importantly, defects in the nuclear pore complex have been observed in patients neurons also in absence of RAN‐T products; an important recent study has characterized the NPC impairment identifying POM121 haploinsufficiency as the initiator of the NPC defect observed in c9‐ALS neurons (Coyne et al, 2020).

With respect to the remaining DPRs, poly‐GA aggregates were characterized by a study conducted in 2018, which defined their morphology by cryo‐electron microscopy as twisted ribbons characterized by densely packed regions. The group also observed proteasome impairment concomitant with the formation of aggregates (Guo et al, 2018b). In line with this evidence, another study described that poly‐GA cytoplasmic aggregates caused TDP‐43 (Khosravi et al, 2017) mislocalization (Nonaka et al, 2018), phosphorylation, and aggregation (Guo et al, 2018a) linked to proteasome dysfunction. Interestingly, a more recent study showed that targeting poly‐GA with antibody therapy could rescue proteasome impairment as well as improve cellular function and cognitive and motor phenotypes in mice (Guo et al, 2018b; Nguyen et al, 2020). Poly‐PA and poly‐GP DPRs have not been associated to any dysfunctional phenotype or toxicity so far, thus they have been poorly characterized.

Importantly, a publication from Gendron and colleagues recently showed the presence of polyGP in the CSF of patients with C9‐HRE, leading to the assumption that DPRs may be secreted and spread along the neural axis via the CSF (Fig 6) (Gendron et al, 2017). In vitro experiments performed by Westergard et al, further demonstrated that DPRs are actively released by cells in both an EV‐dependent and independent manner. They showed that DPRs are transmitted from neuron to neuron as well as from neurons to astrocytes. Additionally, they observed that internalized DPRs were transported along axons both in an anterograde and a retrograde fashion (Westergard et al, 2016). Another work focused on the inter‐neuronal propagation of the dipeptide repeat protein poly‐GA, both in its soluble and aggregated form. Poly‐GA internalization caused TDP‐43 mislocalization and accumulation of poly‐GA in ubiquitinated inclusions, due to impairment of proteasomal activity in the poly‐GA recipient cells (Khosravi et al, 2020). The group fell short of showing whether an impairment of the degradation systems could cause an increase in poly‐GA release, therefore failing to link potential age‐related phenomena, such as the impairment of degradation systems, to the spread of DPRs. Regardless, this initial evidence has laid the foundations for an in‐depth analysis of C9orf72 disease progression and has encouraged the pursuit of a number of studies aimed at unraveling the mechanisms of propagation of pathogenic protein aggregates in C9orf72‐linked ALS/FTD. In this respect, many questions remain unanswered. For instance, as noted before it is still unclear whether the impairment of the degradation systems may prime and amplify the spread of these proteins, or whether uptake of the DPRs may initiate cytotoxicity in the recipient cells. Moreover, the identification of the DPR species with the highest degree of toxicity and spreading propensity would aid in designing tailored, more focused therapies targeting DPRs. Finally, the role of DPRs transmission into non‐neuronal cells is still an unexplored topic.

Secreted proteins and aggregates could serve as powerful biomarkers

Biomarkers are biomedical manifestations that mirror permanent or transient pathological conditions and can be accurately and reproducibly measured. These encompass “almost any measurement reflecting an interaction between a biological system and a potential hazard, which may be chemical, physical, or biological” (WHO International Programme on Chemical Safety. Biomarkers and Risk Assessment: Concepts and Principles). Thus, dysfunctional processes underlying disease may be harnessed as sources of biomarkers. Two major requirements of biomarkers include (i) relevance and (ii) validity. While “relevance” refers to the clinical significance for the disease of interest, “validity” indicates that the “predictive‐power” of a biomarker should be evaluated as population parameters change over time and in case, reshaped (Strimbu & Tavel, 2010). Biomarkers allow to objectively monitor disease progression in addition to or in replacement of clinical manifestations. For instance, because primary endpoints, such as increased survival, occur infrequently, ALS clinical trials often use biomarkers as surrogate clinical endpoints to evaluate target engagement and treatment efficacy. Unfortunately, most biomarkers reliably correlate with disease only after clinical onset. Therefore, early biomarkers are critically needed to shorten the time of diagnosis, thus anticipating the start of treatment and broadening the therapeutic window (Strimbu & Tavel, 2010). Because ALS is a very heterogenous disease encompassing different manifestations and dysfunctions, biomarkers are needed to stratify ALS subtypes as well as to identify phenoconverters (Benatar et al, 2018). As a single biomarker is very unlikely to address all expectations, the diagnosis and prognosis of ALS will probably rely on a combination of biomarkers. To identify ALS biomarkers, it is critical to: (i) detect a biological process which is impaired in disease; (ii) dissect a measurable parameter within the identified process (protein, lipid, metabolite); (iii) develop a method to easily access and measure the identified parameter. This is pivotal for neurodegenerative disorders, such as ALS, where the collection of solid biopsies is both impracticable and unethical. In this context, liquid biopsies have gained critical importance, as they represent the sole means to probe patients with no harm (Ilie et al, 2014; Poulet et al, 2019).

Current biomarkers of ALS pathology include neurofilament light chain (NFL) and, less commonly, phosphorylated neurofilament heavy chain (pNFH), whose levels in the CSF and plasma/serum of patients directly correlate with axonal damage and neuronal degeneration. NFL and pNFH are accurate biomarkers for a large number of neurodegenerative disorders, but lack the ability to distinguish ALS from other neurological disorders as well as to identify different ALS subtypes (Benatar et al, 2018; De Schaepdryver et al, 2018; Poesen & Van Damme, 2019; De Schaepdryver et al, 2019). Therefore, they could be regarded as biomarkers of neurodegeneration and neuronal death more broadly. MicroRNAs (miRNAs) circulating in the blood or incorporated into the EVs have been recently added to the list. For instance, multiple studies have evaluated the suitability of mir‐206 as a predictive tool for ALS diagnosis (Williams et al, 2009). Increased levels of mir‐206, which is well‐established biomarker for Duchenne’s muscular dystrophy, have been linked to muscle waste (Cacchiarelli et al, 2011; Toivonen et al, 2014). Mir‐143‐3p and mir‐374b‐5p have also been tested, but their validity has yet to be demonstrated (Toivonen et al, 2014; Waller et al, 2016; Joilin et al, 2019; Dolinar et al, 2019). Among all processes that are dysregulated in ALS, protein pathology represents an intriguing source for biomarkers, as proteins released in the extracellular space are collected in biofluids. For example, waste products from the CNS convey into the CSF and are further drained in the blood, which is then cleared in the kidneys causing waste products from the CNS to be also found in urines (Laterra et al, 1999). The CSF is the best source for biomarkers of CNS pathologies. However, access to it is rather invasive and potentially harmful. Therefore, blood is often preferred, while urines could be used as a sort of last resort. Proteins associated with the pathogenesis of ALS have been detected and measured in biological fluids as free soluble proteins, incorporated within EVs, or both. Because of the broad spectrum of TDP‐43 pathology, circulating levels of TDP‐43 have been carefully examined. One study assessed the presence of TDP‐43 in CSF and blood and identified EV‐containing TDP‐43 in the CSF. However, no significant differences were measured between ALS patients and control subjects (Feneberg et al, 2014). A subsequent study identified C‐terminal fragments of TDP‐43 in CSF‐derived EVs but did not investigate whether any correlation existed between the presence of these TDP‐43 fragments and the progression of ALS. In 2019, using the sensitive single molecule assay (SIMOA) technology, a group was able to measure higher levels of TDP‐43 in the biofluids of affected ALS patients. They also proposed that using a combination of NFL and TDP‐43 measurements would provide a more reliable diagnostic marker (Kasai et al, 2019). DPRs have also been detected in the CSF of C9‐ALS patients using an ultra‐sensible ELISA immunoassay as well as SIMOA arrays (Gendron et al, 2017). It was also shown that the poly‐GP DPRs were among the most reproducible and precisely measurable DPRs. Although the group was not able to show differences in the levels of poly‐GPs between asymptomatic and symptomatic patients, they did report that CSF poly‐GP is reduced following gene therapy targeting the C9orf72 expansion in mice (Gendron et al, 2017). As of today, CSF levels of poly‐GP are the only pharmacodynamic indicator able to discriminate between neurodegenerative disorders and ALS granted the presence of the C9orf72 mutation. However, this approach presents limitations, as it can only be used for patients carrying the C9orf72 mutation, as well as being restricted to CSF sampling. No other DPR measurements were done, and an evaluation of the presence of DPRs in plasma/serum as either free protein or incorporated into the EVs has not been yet carried out. Misfolded SOD1 was found in sporadic and familial patients CSF but there was no significant difference with the controls it has thus been disregarded as possible biomarker. Proteomic analysis recently failed to detect FUS in CSF as either free protein or in EVs (Chiasserini et al, 2014; Guha et al, 2019). However, in vitro studies aimed at assessing the presence of FUS in EVs released in the culture medium successfully identified both wild type and NLS‐mutant FUS, and reported R495X NLS‐mutant FUS mutant to be secreted at significantly higher levels (Kamelgarn et al, 2016). This suggested that following cytoplasmic mislocalization and induction of proteostatic imbalances, more FUS protein may be redirected to EV secretion, and indicated that EV‐derived FUS protein levels may be used as biomarkers for FUS proteinopathies. Interestingly, a study by Dormann and colleagues suggested that differences in the methylation levels of FUS may help distinguish between FUS‐ALS and FUS‐FTD (Dormann et al, 2012). In fact, proteinaceous inclusions of FUS‐ALS patients were shown to contain normally methylated FUS, while FUS‐FTD patient inclusions exhibited hypomethylated FUS. One fascinating possibility is that through combined measurement of EV‐derived FUS protein levels and methylation status, it may be plausible to detect FUS pathology and discriminate between ALS and FTD. However, this remains a speculation so far and more studies on FUS are urgently needed.

Generally speaking, more work needs to be done on the biomarkers front. In particular, large patient cohorts will be required to assess relevant biomarkers on a broad scale, and future studies will need to engage multiple patients in clinical centers across several countries (Strimbu & Tavel, 2010).

Conclusions and perspectives of intervention

In this review, we discussed the central role of protein pathology in ALS onset and progression. Numerous mechanisms underly disease pathogenesis, including DNA damage, dysregulated RNA processing, mitochondrial abnormalities, impaired axonal trafficking, and non‐cell autonomous contributions (Wijesekera & Leigh, 2009). However, the distinction between causes and consequences of the pathology has been extremely subtle. The tendency of ALS proteins to aggregate is very likely to occur at an early stage, thereby triggering a cascade of untoward events. For instance, long‐term accumulation of aggregated proteins has been shown to induce DNA double‐strand breaks and nuclear dysmorphia (Lu et al, 2015). Because RBPs are intrinsically aggregation‐prone, aggregates of misfolded proteins may cause their sequestration and interfere with RNA metabolism (Maziuk et al, 2017), which is particularly important in the brain. Additionally, aggregate formation has been linked to oxidative stress in mitochondria (Lin & Beal, 2006). In neurons, aggregates have been shown to physically reduce axonal flux by generating obstructing clumps (Vital et al, 2014). Moreover, aggregated species are a potent trigger of proinflammatory microglia, and astrocytes may act as reservoirs for the internalization and release of misfolded proteins and aggregates (Hook et al, 2015). Importantly, it has become more and more evident that pathogenic proteins, oligomers, and higher‐order assemblies can be exchanged between cells and propagate pathology. For these reasons, we envision therapeutic strategies aimed at preventing or halting seeding and transmission of misfolded protein species as the most likely to succeed in treating this incurable disorder. For instance, considerable research has been focusing on exploring therapeutic opportunities targeting the formation, processing, clearance, and spread of pathological proteins. Because ALS mutations are often associated with the acquisition of toxic functions, several groups have designed strategies to directly prevent the production of mutant protein species by exploiting Watson‐Crick base pairing rules to induce the selective degradation of their transcripts (Smith et al, 2006; Wang et al, 2008; Wu et al, 2009; Foust et al, 2013; Miller et al, 2013; Wang et al, 2014; Dirren et al, 2015; Stoica et al, 2016). A distinct set of studies have focused on boosting the molecular chaperone machinery to combat protein misfolding through the coordinated activation of multiple Hsps (Kalmar et al, 2008; Malik et al, 2013). However, by the time of disease onset, perturbations in protein homeostasis have largely occurred at the cellular level and harnessing the cell’s chaperone systems alone may be of little benefit. Hence, many efforts have been made to enhance protein degradation in order to clear out existing aggregates and reduce aberrant protein levels. Due to the poor druggability of the proteasome, most studies have been oriented toward stimulating autophagic degradation (Spencer et al, 2009; Wang et al, 2012; Castillo et al, 2013; Barmada et al, 2014; Chang et al, 2016; Marrone et al, 2019). However, the recent development of proteolysis targeting chimeras (PROTACs) has redirected the focus to the proteasome as a potential therapeutic site, and this rapidly evolving technology is showing great promise to quickly advance from the bench to the bedside (Sakamoto et al, 2001). More importantly, antibodies and antibody fragments (scFv, nanobodies) are being developed as a fascinating class of therapeutics aimed at targeting aggregates and interrupting the cell‐to‐cell spread of protein pathology underlying the stereotypic progression of neurodegeneration. Numerous works that have focused on tau, α‐synuclein and Aβ pathology are now at different stages of clinical development (Valera et al, 2016; Congdon & Sigurdsson, 2018). However, similar studies have been conducted in the context of ALS. In 2010, Gros‐Louis and colleagues reported on the discovery of monoclonal antibodies that could specifically recognize misfolded SOD1^G93A, but not its wild‐type counterpart, via identification of the usually buried hydrophobic region (Gros‐Louis et al, 2010). Antibody treatment of diseased mice via intracerebroventricular infusion successfully reduced mortality, diminished the burden of aggregated species, and attenuated MN loss. Similar findings were also reported more recently (Maier et al, 2018). The preclinical success of passive immunotherapy was also demonstrated against misfolded TDP‐43 (Tamaki et al, 2018; Pozzi et al, 2019) as well as C9Orf72‐derived DPRs (Zhou et al, 2017; Nguyen et al, 2020). Contrarily, FUS‐immunotherapy is still a largely unexplored topic due to the lack of research on protein transmission. The exact mechanisms conferring protection upon passive immunization are not entirely understood. Antibody‐bound extracellular aggregates may trigger microglia‐mediated phagocytosis of the immunoglobulin‐misfolded protein complexes. Additionally, antibodies and antibody fragments have been shown to penetrate cells and induce Fc‐mediated proteasome degradation (McEwan et al, 2017). Future studies will shed more light on these mechanisms. Importantly, because CSF and blood levels of extracellularly released misfolded proteins act as reliable disease biomarkers, validation of target engagement will be achieved by monitoring circulating proteins using non‐invasive procedures. To obtain the best results, therapy administration should occur as early as possible. Therefore, future research efforts should focus on the detection of pathological clues potentially prior to symptom onset.

Author contributions

Literature search and manuscript drafting: MEC and LM. Manuscript revision and editing; manuscript approval: DT and MA.

Conflict of interest

The authors declare that they have no conflict of interest.

The EMBO Journal (2021) 40: e106389.