Non-cell-autonomous pathogenic mechanisms in amyotrophic lateral sclerosis

Abstract

Amyotrophic lateral sclerosis (ALS) is the most common adult-onset paralytic disorder, characterized mainly by a loss of motor neurons (MNs) in the CNS. Over the past decades, thanks to intense investigations performed in both in vivo and in vitro models of ALS, major progress has been made toward gaining insights into the pathobiology of this incurable, fatal disorder. Among these advances is the growing recognition that non-neuronal cells participate in the degeneration of MNs in ALS, which could transform our understanding of the neurobiology of disease and the ability to devise effective disease-modifying therapies. In this review, we examine the contribution of non-cell-autonomous processes to the pathogenesis of ALS, with a focus on glial cells and in particular on astrocytes.

Clues that non-cell-autonomous processes contribute to ALS

ALS is an adult-onset neurodegenerative disorder characterized mainly by the degeneration of MNs in the cerebral cortex and/or brainstem and spinal cord. ALS is a fatal disorder, leading to the death of patients from, in most instances, respiratory failure typically between 2 and 5 years after diagnosis [1]. Despite decades of intense research, to this day no cure is available, and up to now the US FDA has approved only three drugs (riluzole under various formulations, the combination of dextromethorphan HBr and quinidine sulfate, and edaravone) with mild disease-modifying benefits, at best. Although ~90% of ALS cases occur without a family history [referred to as sporadic ALS (sALS)], ~10% are linked to inherited mutations in more than 25 disparate genes [2]. Among these, mutations in superoxide dismutase 1 (SOD1) and hexanucleotide repeats in C9orf72 are the most prevalent [3]. Less frequent genetic causes include, for example, mutations in fused in sarcoma/translocated in liposarcoma (FUS/TLS), TAR DNA binding protein 43 (TARDBP), TANK-binding kinase 1 (TBK1), or valosin-containing protein (VCP) [3]. Despite this genetic diversity in ALS, mutations in SOD1 have received the lion’s share of attention, in part because transgenic rodents carrying mutant SOD1 (mSOD1) recapitulate most of the clinical and neuropathological hallmarks of ALS [4]. However, these transgenic rodents do not pheno-copy sALS, in that, for instance, the degeneration of cortical MNs, which is fundamental for the diagnosis of the disease [1], has been inconsistently documented [5,6] and the biochemical hallmarks of TDP-43 protein pathology observed in the majorly of sALS cases [7] are, at best, partially recapitulated in these animals [8,9]. Despite these caveats, the mSOD1 mouse model has led to a number of insights into the pathobiology of ALS not only at the molecular but also at the cellular level, as discussed in this review. Also discussed in this review are results from cell-based models of ALS, such as human induced pluripotent stem cells (iPSCs) differentiated into MNs and expressing various pathogenic mutations.

The longstanding neurocentric view of ALS stems, in part, from the notion that the degeneration of MNs in ALS is caused by cell-autonomous processes that are independent of extracellular influences. However, mounting evidence indicates that non-neuronal cells may play a role in the pathogenesis of a number of different neurodegenerative disorders [10], including ALS. For instance, genetic variants in triggering receptor expressed on myeloid cells 2 (TREM2) were found to increase the risk for neurodegenerative disorders such as ALS [11]. In an attempt to find cell types involved in ALS occurrence, Saez-Atienzar et al. reported that OPALIN-, FCHSD2-, and LAMA2-expressing oligodendrocytes are also associated with increased risk of developing the disease [12]. Furthermore, overexpression of mSOD1 or deletion of C9orf72 and TBK1, whose gene products are highly expressed in glial and immune cells, are found to alter the functions of these cells [13].

While the aforementioned genetic data provide impetus to the concept of non-cell autonomy in ALS, mechanistic support came initially from chimeric mice [14]. These animals, which comprise cells expressing either wild type or mSOD1, revealed the striking finding that wild-type MNs surrounded by non-neuronal cells expressing mSOD1 can acquire ubiquitin-positive protein aggregates [14], a sign of neuronal damage in this ALS model [15]. Furthermore, unlike the germline expression of mSOD1, the selective expression of this mutated enzyme in MNs is associated with either no [16] or a late [17] ALS-like phenotype. As for transgenic mice with MN-restricted deletion of mSOD1, it is noteworthy that these animals showed attenuation, but not abrogation, of the ALS-like phenotype: these mice lived longer, but eventually developed terminal paralysis [18]. Together, these findings support the notion that non-MN cells may contribute to the degeneration of neighboring MNs in ALS. Here, we review emerging findings consistent with this non-cell-autonomous pathogenic paradigm of ALS. However, the focus of this review should not be misconstrued as an attempt to dismiss the recognized role of cell-autonomous mechanisms in ALS pathogenesis; our goal is to highlight that there may be more to ALS pathogenesis than initially thought. Thus, the reader interested in the cell-autonomous processes of ALS is referred to reviews on this topic for more details [19–21].

Non-cell-autonomous deleterious effects on MNs

From the growing interest in the non-cell-autonomous hypothesis in neurodegenerative disorders and particularly in ALS, a number of elegant studies in diverse cell types have emerged. Among these different cell types, it is worth mentioning first that overexpression of mSOD1 in skeletal muscles, the effector targets of spinal MNs, causes local deleterious effects ranging from metabolic to morphological alterations [22], but whether these muscle abnormalities can engender MN degeneration remains debatable [23–25]. By contrast, a clearer picture exists in support of the contribution of glial cells to the demise of MNs in ALS. However, by the sheer number of publications on astrocytes in ALS, it should come as no surprise that the focus of this review would predominantly be on astrocytes, which should not be taken as suggesting that other glial cells such as microglia or oligodendrocytes [12,26] or bloodborne immune cells [13] are not important to the non-cell-autonomous pathogenesis of ALS. For instance, TDP-43 inclusions were detected in both spinal MNs and oligodendrocytes in patients with sALS and, even prior to overt loss of MNs, spinal cord sections from transgenic mSOD1 mice showed a substantial number of spinal hypertrophic mature oligodendrocytes that were positive for active Caspase-3 [27]. In the same study, the authors found that the newly generated oligodendrocytes in spinal cords from transgenic mSOD1 mice exhibited lower expression of myelin basic protein (MBP) and monocarboxylate transporter-1 (MCT1), suggesting that these cells may have myelination and metabolic defects [27], which can be envisioned to affect the wellbeing of neighboring MNs.

Moreover, it should be noted that while overexpression experiments on ALS-related gene mutations in specific cell types have yielded valuable insights, conditional deletion, passive transfer of cells, and bone marrow grafting strategies aimed at reducing the expression of ALS mutations in specific cells, or at reducing the number of specific cell types expressing these mutations, have generated more enlightening results. In aggregate, the outcomes of most of these studies, especially those targeting mSOD1, were mostly congruent in that both reducing mutant proteins per cell and having fewer cells expressing mutant proteins attenuated the ALS-like phenotype [28], raising the idea that not a single but rather several non-neuronal cell types may participate in the degenerative process in ALS, including astrocytes [29,30], microglia [18,31], and oligodendrocytes [32]. However, three sets of findings warrant further attention. First, mice with Schwann cell-specific mSOD1 excision, contrary to a similar conditional strategy in other glial cells, were reported to have an accelerated ALS-like phenotype [33]. Second, replacing peripheral mSOD1 macrophages with wild-type macrophages had marginal effects in transgenic mSOD1 mice, unless wild-type macrophages were also engineered to produce less reactive oxygen species and were grafted not at presymptomatic but at early symptomatic stages [34]. Third, astrocyte-specific deletion of mSOD1 resulted in longer survival compared with germline transgenic mSOD1 counterparts [29,30]. However, a delay in the onset of symptoms was observed only on the silencing of SOD1^G37R, and not SOD1^G85R [29,30]. Perhaps this preclinical discrepancy typifies the enigmatic variability of clinical phenotype observed in patients carrying different SOD1 mutations [35] and is related to the fact that SOD1^G85R, which has impaired metal-binding and reduced catalytic activity, causes greater and earlier alterations in astrocytes than SOD1^G37R, which has normal metal-binding and catalytic activity [36–38]. Incidentally, given the distinct characteristics of these two SOD1 mutants, it would have been interesting to see whether excising SOD1^G85R in Schwann cells has the same effect on the disease phenotype in this ALS mouse model as SOD1^G37R [33]. Thus, while additional investigations are needed to elucidate the basis of these different observations, they already herald that the non-cell-autonomous hypothesis is multifaceted and likely to be more complex than perhaps expected.

Mechanisms driving non-cell-autonomous deleterious effects on MNs

In light of that discussed earlier, one may wonder how non-neuronal cells contribute to the demise of neighboring MNs. To begin to address this question, many investigators have used astrocytes expressing mSOD1 as a prototypical non-neuronal cell model that can readily be studied both in vitro and in vivo. Salient findings generated by these studies include the report that mSOD1 astrocytes, both in vitro using rodent primary and embryonic stem cell (ESC)-derived MNs [39–46] and in vivo by the transplantation of mouse mSOD1 expressing glia-restricted precursors into the spinal cords of wild-type rats, led to a decrease in spinal MN numbers [47]. However, can these studies be translated to humans? This question has been partly addressed by the following in vitro studies: (i) mouse mSOD1 astrocytes were associated with a loss of human ESC-derived MNs [42,48]; (ii) human astrocytes derived from neural progenitors (NPs) of postmortem sALS and mSOD1 patients caused the loss of mouse ESC-derived MNs [49]; and (iii) sALS patient-derived astrocytes mediated by an unknown mechanism caused the loss of human ESC-derived MNs [50]. Last, it was shown that media conditioned with either mouse mSOD1 or human sALS astrocytes were similarly associated with loss of spinal MNs [39,50], suggesting that these non-cell-autonomous deleterious effects do not rely on cell contact and may be shared by the sporadic and the genetic forms of ALS. However, iPSC-derived astrocytes expressing C9ORF72, TDP-43, or VCP mutants failed to prove toxic to co-cultured wild-type MNs [51–53]; admittedly, one cannot exclude the possibility that the negative results obtained with these non-mSOD1 mutations stem from the fact that iPSC-astrocytes may not be identical to primary or NP astrocytes.

Also worth mentioning is the critical feature of ALS that not all spinal neurons and not all MN subpopulations are equally susceptible to the disease process [54], which, in the context of non-cell autonomy, has received relatively limited attention. Relevant to this point, our group has reported that ALS astrocytes are toxic to spinal MNs but not to Lim1/2 or LH2 interneurons [39]. However, it remains unknown (at least to our knowledge) whether MNs in the spinal Onuf’s or brainstem oculomotor nuclei, which are typically spared in ALS patients and in transgenic mSOD1 mice [54], are resistant to the non-cell-autonomous deleterious effects. Addressing this question may be valuable for both our understanding of the pathobiology of ALS and its treatment, as this may shed light on whether the differential susceptibility results solely from MN-intrinsic properties or involves signaling from neighboring non-neuronal cells.

Two mechanistic hypotheses, although not mutually exclusive, can be proposed to explain the contribution of non-neuronal cells to the demise of MNs in ALS (Figure 1): a loss of beneficial functions or the acquisition of noxious properties. Both possibilities are discussed later separately, but it must be recognized that to obtain an accurate picture of the non-cell-autonomous contribution to ALS pathogenesis, one may have to also consider additional variables such as the stage of the disease and the cell types being studied. It was shown that key molecular mechanisms in glial cells may be neuroprotective in presymptomatic animal models of ALS and become neurotoxic at later stages of the disease [55] and that non-cell-autonomous effects of cells like neural crest-derived Schwann cells [33] may not be identical to those of neural tube-derived oligodendrocytes [27,32]. Thus, some of the discussion later may have to be taken with caution, as the non-cell-autonomous hypothesis of ALS calls for a great deal of temporal and cell-specific contextual nuance.

Figure 1. Illustration of the non-cell-autonomously based pathogenesis of amyotrophic lateral sclerosis (ALS).

Mounting evidence indicates that nonneuronal cells such as glial and immune cells could contribute to the degenerative process of ALS. However, according to this pathogenic hypothesis, not one but rather a host of different molecular mechanisms originating from non-neuronal cells may contribute to the demise of motor neurons (MNs). These non-cell-autonomous deleterious mechanisms can be divided into two main categories, which are not mutually exclusive: those due to a loss of beneficial effects (left box) and those due to a gain of toxic effects (right box). The non-cell-autonomous loss of beneficial effects involves a series of alterations in glial cells, which include reduced glutamate uptake and efflux lactate that is essential for MN energy supply, as well as impaired molecular injury responses. As for the non-cell-autonomous gain of toxic effects, it is believed that both inflammatory factors and non-inflammatory factors such as amyloid precursor protein (APP) fragment may be released by glial cells, which in turn engage death-signaling pathways in neighboring MNs. Abbreviations: C1Q, complement component 1q; DR6, death receptor 6; EphB1, ephrin type-B receptor 1; IL-1α, interleukin 1 alpha; MCT1, monocarboxylate transporter-1; NF-κB, nuclear factor kappa B; TNFα, tumor necrosis factor alpha.

Loss of beneficial function

The first hypothesis pertains to the loss of beneficial function and stems from the idea that the disease process does not necessarily kill non-neuronal cells but rather subverts their normal functions (e.g., supplying nutrients and neurotrophic factors, buffering ions), thus impairing their supportive/protective effects on neighboring MNs (Figure 1). Among potential loss-of-function mechanisms linked to MN degeneration in ALS, the reported decreases in the glutamate transporter GLT1 and the lactate transporter MCT1 in glial cells are both appealing [56–59]. Based on these findings, it is expected that a reduced uptake capacity for glutamate by ALS astrocytes exposes MNs to chronic excitotoxicity and reduced efflux of lactate from ALS oligodendrocytes starves MNs of energy-generating substrates, in both instances leading eventually to neurodegeneration. Further findings consistent with the loss-of-function hypothesis include investigations in neuron–astrocyte co-culture systems [60] that revealed marked downregulation in mSOD1 astrocytes of about 34 genes including Osteopontin, Sparcl1, Sepp1, Apolipoprotein D, and Pleiotrophin, interpreted by the authors as suggesting that mutant glia have a compromised molecular response to injury. In keeping with this interpretation are the following three observations: (i) the ephrin-B1 reversed signaling of signal transducer and activator of transcription 3 (STAT3)-dependent protective and anti-inflammatory responses to MN injury is blunted in human iPSC astrocytes from ALS patients and in transgenic mSOD1 mice [61]; (ii) human iPSC astrocytes from healthy subjects improved the survival of co-cultured MNs, a supportive effect that is lost in human iPSC astrocytes from patients carrying either the VCP^R191Q or the VCP^R155C mutation [53]; and (iii) several ubiquitously mutant genes causing ALS affect autophagy [13], which, in turn, may alter not only MN but also glial functions, including their responsiveness to pathological stimuli [62].

Gain of toxic function

Although the findings discussed earlier are consistent with an impairment in the supportive/protective and quality control properties of glial cells in ALS, the second hypothesis, which pertains to a gain of toxic effect, has received even greater attention (Figure 1). This is due in part to the widely accepted concept that immune and glial cells have the capacity to mount a neuroinflammatory response that may exert deleterious effects on neighboring cells such as MNs; for example, by the production of reactive oxygen/nitrogen species, complement components, and cytokines and chemokines [13]. Relevant to this concept are the demonstrations of morphological and immunostaining alterations consistent with activated microglia, reactive astrocytes, and T cell infiltration in areas of neurodegeneration in postmortem samples from ALS patients [13]. Furthermore, increased binding for translocator protein radioligands can be detected by positron emission tomography in ALS-relevant brain regions of living patients, suggesting that microglial activation in ALS occurs early over the course of the disease [13]. The same neuroinflammatory changes seen in ALS patients are seen in mouse models of ALS, even prior to the onset of weakness [13]. These findings suggest that features of neuroinflammation are detected in both patients with and animal models of ALS early enough to potentially play a meaningful role in the overall disease process.

Conventionally, neuroinflammation in ALS is regarded as secondary to the loss of MNs and hence not as an initiator of ALS but rather as a potential amplifier of its progression [13]. However, this view may have to be amended in light of the demonstrations that overexpression of mSOD1 [63,64] as well as C9orf72 haploinsufficiency [65,66] is associated with a hyperresponsive phenotype of microglia for the former and myeloid cells for the latter. Of note, three main mechanisms have been proposed to explain the pathogenic role of C9orf72 in ALS: loss of function of the C9orf72 protein and toxic gain of function from C9orf72 repeat RNA or from dipeptide repeat proteins produced by repeat-associated non-AUG translation [3]. Although, as indicated earlier, loss of normal C9orf72 function causes immune dysregulation, whether the RNA or dipeptide toxic gain of function triggers, in glial and immune cells, an abnormal inflammatory phenotype remains, to our knowledge, to be established. In the case of the loss of C9orf72, this maladaptive inflammatory response revolves around the loss of suppression of activators of the stimulator of interferon genes (STING) protein, since blocking STING abates the hyperactive type I interferon responses in C9orf72^−/− immune cells as well as splenomegaly and inflammation in C9orf72^−/− mice [65]. As for mSOD1 overexpression, the hyperresponsiveness of microglia is attributed to aberrant activation of nuclear factor kappa B (NF-κB) and enrichment of α2-Na/K ATPase-α-adducin in glial cells, since inhibition of this transcription factor or deletion of this protein complex attenuates the non-cell-autonomous toxicity on MNs [67–69]. In a more complex scenario, it has been proposed that microglia in diseases such as ALS, by releasing interleukin 1 alpha (IL-1α), tumor necrosis factor alpha (TNFα), and complement component 1q (C1q), coax astrocytes to adopt a neurotoxic phenotype, which, in turn, promotes neurodegeneration [70,71]. However, at this time it remains uncertain how the aforementioned molecular events that drive the nonneuronal cell-mediated MN degeneration occur in response to ALS mutations.

Besides the possible contribution of non-neuronal cells to MN degeneration via neuroinflammation, it should be considered that glia could also contribute to ALS pathogenesis via the release of neurotoxic factors other than reactive oxygen/nitrogen species, complement components, and cytokines and chemokines. For example, the use of neuron–astrocyte co-cultures revealed that, while both mutant and wild-type mouse astrocytes are reactive in culture, SOD1^G93A astrocytes display genomic profiles distinct from those of SOD1^WT astrocytes [60]. Likewise, human sALS astrocytes induce MN degeneration in vitro despite the fact that a set of common proinflammatory cytokines are not differentially expressed between cultured sALS astrocytes and their non-neurological disease counterparts [50]. Collectively, these findings support the idea that the non-cell-autonomous toxicity may not necessarily or solely rely on common neuroinflammatory pathways.

An approach to identify the toxic factors released by non-neuronal cells is to examine early pathogenic changes in MNs exposed to, for example, ALS astrocytes. By interrogating a mouse brain interactome (i.e., regulatory interactions between transcription factors and their targets), our group has used gene expression profile signatures obtained from ESC MNs cultured in medium conditioned with mSOD1-expressing astrocytes to identify master regulators of neurodegeneration, 11 of which were confirmed experimentally [72]. Among these, we found that the silencing of eight genes (Tgif1, Nfkb1, Epas1, Psmc3ip, Tcf7, Tcf3, Zfp36l1, and Zdhhc2) had a neuroprotective effect that was specific for the toxicity induced by mSOD1-expressing astrocytes and did not affect the survival of neurons exposed to control astrocytes. Given the aforementioned role of NF-κB in glial cells, it is worth noting that, here, the engagement of Nfkb1 – hence the NF-κB pathway – was intrinsic to MNs on exposure to mSOD1-expressing mouse astrocytes and protected against the toxicity of mSOD1 astrocytes as well as sALS patient-derived astrocytes. Even more striking is the demonstration by others that the genetic inhibition of NF-κB specifically in MNs mitigates neuropathology in transgenic mice expressing either mutant TDP-43 or mutant SOD1 [73]. Taken together, these transcription factors constitute potential therapeutic targets as well as early biomarkers for ALS and suggest that NF-κB, in both glia and MNs, may be involved in ALS pathogenesis.

The precise nature of the factors released by ALS astrocytes that are toxic to MNs is still under investigation. However, by using a novel bioinformatics pipeline that combines proteomics and genomics data, our group has reported that ALS astrocytes generate an aberrant fragment of amyloid precursor protein (APP) (and perhaps of amyloid-like protein 1), which, by activating death receptor 6 (DR6) expressed on the surface of MNs, triggers neurodegeneration [74]. Nonetheless, since the deletion of DR6 provided only partial protection against neurodegeneration in transgenic mSOD1 mice, it is likely that the non-cell-autonomous arm of ALS pathogenesis is driven by additional mechanisms. This astrocyte APP/DR6 toxicity was shown to depend on the engagement of the cell death Bax pathway. Interestingly, while MNs exposed to sALS- and familial (f)ALS-derived astrocytes display signs of apoptosis [39,50], anticaspase strategies both in vivo [75] and in vitro [39] provided only limited benefit, suggesting that MN death may occur via a Bax-dependent, caspase-dispensable cell death program. Relevant to this view are the demonstrations that the toxicity of both mouse and human ALS astrocytes was mitigated by silencing components of the necroptotic cell death pathway [50,76]. However, whether this form of cell death is instrumental in the demise of MNs in vivo remains uncertain [76–78].

Glial cell heterogeneity

The premise that glial cells, and microglia in particular, are heterogeneous has been increasingly acknowledged with the rising awareness of their complexity and diversity of functions [79]. Morphological differences between glial cells, such as astrocytes in gray and white matter, as well as differences in generic surface markers among glial cells, have been known for decades. It is only recently that strategies capable of capturing glial heterogeneity rather than assessing the state of selected inflammatory factors have started to arise. Bulk RNA-seq approach to whole-tissue samples and acutely purified glial cells from ALS patients and rodent models of the disease [12,80–83] and, more recently, single-cell RNA-sequencing in the spinal cord [84] and brainstem [85] of transgenic mSOD1 mice have revealed previously unobserved heterogeneity in glial populations. Although these approaches have begun to provide insights into glial cell heterogeneity in ALS, it is also critical to determine where these different phenotypically defined clusters of glial cells are anatomically localized in diseased tissues. Emerging technologies such as spatially resolved transcriptome-wide profiling have the power to reveal both the regional heterogeneity of glial cells and the spatiotemporal dynamics of disease-relevant signaling pathways. For example, in transgenic mSOD1 mice, Tyrobp/Trem2 signaling appears to be dysregulated at the presymptomatic stage in microglia that closely abut motor axons in the ventrolateral white matter [86]. In addition, subpopulations of glial cells may have distinct spatiotemporal responses over the course of the disease, depending on their location in tissue [86]. Of note, these glial subpopulations appear to be differentially responsive to perturbations in autophagy in MNs [86]. Going forward, such studies are likely to play a critical role in elucidating region-specific glial heterogeneity and, even more importantly, whether such regional differences may explain, at least in part, the differential susceptibility of MNs in ALS [54] mentioned earlier. Also worth noting is that proteome heterogeneity in glial populations in ALS may not necessarily be reflected in changes in mRNA expression. Along these lines, the emergence of single-cell proteomics techniques used in ALS tissues [87] and in macrophage-like cells [88] may ultimately prove valuable.

Thus far, several interesting questions remain unresolved. Is glial cell heterogeneity driven mainly by intrinsic cellular programs or do cell-to-cell interactions play a role? Can the subsequent distinct identities lead to different responses in the context of ALS? Relevant to this point is the proposal that there might be two major subtypes of reactive astrocytes: LPS-induced neurotoxic A1 astrocytes and ischemic penumbra-associated neuroprotective A2 astrocytes [70]. In this in vitro model, as mentioned in the gain of toxic function section, microglia induce the astrocyte A1 phenotype through the release of IL-1α/TNFα/C1q, resulting in the death of neurons and oligodendrocytes. However, we believe that there may be more shades of astrocyte phenotypes than this strict dichotomy, given the transcriptomic and proteomic heterogeneity of these cells that is starting to be reported in various human diseases of the nervous system.

Studies that do not discriminate between astrocyte subtypes show a general toxicity of astrocytes derived from sALS patients or expressing ALS-linked mutations, suggesting that heterogeneity might not be the primary driver of neurodegeneration. Moreover, the proposed region-biased toxic feedback loop created by the activation of microglia and astrocytes may reflect secondary mechanisms triggered by neuronal damage in the immediate neighborhood [13]. In addition, the fact that differentially susceptible MNs are located in close proximity to one another possibly implies that differences in their vulnerability to degeneration may primarily rely on cell-intrinsic mechanisms, even if, as suggested earlier, interactions with non-neuronal cells are involved. Thus, if the heterogeneity of glial cells may not be instrumental in driving the neurodegenerative disorder, it may still be critical to the modulation of MN vulnerability.

Neuron–non-neuronal cell crosstalk

Despite the possible non-cell-autonomous contribution to the pathogenesis of ALS, disease-related changes in non-neuronal cells in ALS may not arise entirely cell autonomously but rather rely, at least in part, on MN–glia crosstalk. To explore such a possibility, investigators designed an in vitro co-culture system comprising an astrocyte monolayer physically separated from MNs in the same dish to allow constant communication through the culture medium [60]. RNA-sequencing of both cell types separately after co-culture for different periods of time showed that mSOD1 expression elicited the perturbation of similar pathways in MNs and astrocytes as well as a dynamic interaction between the two cell types. Interestingly, many of the dysregulated genes encode cell-surface proteins or proteins involved in cytoskeletal transport, stress, and injury response. One pathway found to be perturbed in both co-culture systems as well as in spinal cords of transgenic mSOD1 mice was the transforming growth factor beta (TGF-β) signaling pathway, which is known to regulate astrogliosis and to play a neuroprotective role on acute injury [89,90]. TGF-β₁ secreted by reactive astrocytes, by contrast, was later shown to induce protein-aceous aggregates and MN death [91]. The impact on this pathway in the co-culture system was seen at the level of the cell surface. Moreover, in mSOD1 mice, elevated expression of the TGF-β type II receptor was dysregulated in spinal cord well before symptom onset and correlated with reactive astrogliosis in the surrounding astrocytes. Also, relevant to this vicious cycle model is the proposed breakdown in neuronal–glial communication in ALS due to alterations in the ephrin type-B receptor 1 (EphB1)–ephrin-B1 ligand reversed signaling presenting as a primary neuronal injury cue triggering a STAT3-protective astrocyte phenotype [61]. Further supporting the notion that abnormal signals emanating from MNs can impact the behavior of glial cells is the demonstration that the conditional deletion of TBK1 in MNs reduces the responsiveness of glial cells to pathological stimuli [92]. More speculative, but no less interesting, is the idea inspired by the demonstration that the expression of matrix metallopeptidase-9 (MMP-9) in MNs correlates with their propensity to degenerate in ALS [93]. Since glial responsiveness is influenced by the composition of the extracellular matrix they contact, it is then possible that the MN-released MMP-9, via its proteolytic activity, modulates, for example, the activation of microglia and the ensuing MN degeneration. Taken together, these studies support the interconnected network of pathways in neuron–glia communication. They also provide insights into how these pathways, on dysregulation, may modulate the effects of non-neuronal cells on neighboring MNs and ultimately contribute to neuronal death and even MN differential susceptibility.

Concluding remarks and future perspectives

Despite the many imperfections of the in vivo and in vitro models of ALS and the debate about how authentically ESCs and IPSCs emulate their primary counterparts, we would argue that the findings discussed here provide compelling evidence for a role of non-neuronal cells in neurodegeneration in this complex, fatal MN disease. However, several interesting unresolved questions remain to be addressed (see Outstanding questions). Nonetheless, while until now much of the attention on non-cell autonomy in ALS has been on the neuropathobiology of the disease, one should not underestimate the importance of this concept in guiding therapy. According to this framework, for therapies to be effective, rather than targeting MNs specifically they may have to be active in both neuronal and non-neuronal cells. This is exemplified in the ongoing clinical trials in which knockdown of SOD1 or C9orf72 using antisense oligonucleotides aims to decrease the pathogenic gene in many cell types of the CNS. Moreover, the studies discussed in this review stress the fact that some of the identified non-cell-autonomic processes may be shared by the sporadic and different genetic forms of ALS, while others may be mutation specific. Thus, the exploitation of non-cell autonomy for therapeutic purposes may have to rely on a precision medicine approach where the treatment may have to target specific factors in specific patients who harbor specific mutations. In contrast to this pathogenic variability, as discussed earlier, the differential susceptibility of MNs transcends the issue of sporadic versus genetic disease and the type of gene mutations, as the same cluster of MNs is consistently lost or spared among all forms of ALS [54]. Thus, looking forward, perhaps the greatest impact that further investigations on non-cell autonomy may have for ALS patients is in determining whether non-neuronal cells – besides killing neighboring MNs per se – can modulate the propensity of MNs to degenerate.

Outstanding questions.

An important goal for future research is to elucidate why glial and immune cell non-cell-autonomous effects in ALS impact MNs preferentially. Do the deleterious actions of glial and immune cells in ALS occur only in areas where MNs reside or are MNs specifically susceptible to these alterations?

Also critical to elucidate are non-cell-autonomous responses in a time-dependent manner, adding another dimension to their functional complexity. Along this line, at which stage throughout their development do glial and immune cells acquire their toxicity for MNs?

Regarding astrocyte and microglial heterogeneity, what would be the optimal way to therapeutically target these cells and shift the balance from toxic to protective? How can deeper understanding of the factors discriminating between toxic and protective subtypes be obtained?

What are the critical upstream events that initiate the non-cell-autonomous effects on MNs? Are these events triggered in glia and immune cells autonomously (i.e., alterations emerging in a given cell type in the absence of any extrinsic influences) or non-autonomously (i.e., in response to alteration in cell-to-cell communications with other glial cells or even MNs themselves, causing a feedforward deleterious cascade)?

How do other cell types besides astrocytes (e.g., oligodendrocytes, microglia, T cells) orchestrate a network of interconnected cells that contribute to the response that ultimately contributes to the demise of MNs?

In ALS, both the PNS and CNS are affected. It will thus be interesting to compare the non-cell-autonomous responses occurring in the PNS and CNS. What is the relative role of each in the ALS paralytic phenotype?

How do different ALS-linked mutations manifest in terms of non-cell-autonomous responses, for different types of glial and immune cells?

Highlights.

Whereas the pathophysiology of amyotrophic lateral sclerosis (ALS) involves the selective degeneration of motor neurons (MNs), non-neuronal cells are increasingly acknowledged as critical determinants of MN death and survival.

Extensive co-culture experiments suggest that the deleterious effects of astrocytes on MNs involve loss of neuroprotective function as well as gain of toxic function.

Astrocytic heterogeneity may contribute to the mix of neuroprotective and neurotoxic effects imposed on neighboring MNs and can itself be influenced by other cell types.

The intricate communication between neurons and astrocytes bridges cell-intrinsic and -extrinsic factors, and this crosstalk is perturbed by mutant superoxide dismutase 1 (SOD1).