The work presented here suggests that circRNA expression may be profoundly impaired in CNS and skeletal muscle, the two tissues most affected by ALS. The observed dysregulation of circRNA levels seems to argue for a pattern that is tissue-dependent: whereas in the ALS spinal cord and brain circRNAs are generally downregulated, in the skeletal muscle their abundance tends to increase (Fig. 6 and Fig. 7).
The abundance of circRNAs can be altered mainly due to three causes: (1) circRNA biogenesis (regulated by RBPs, Alu elements and ADAR enzymes), (2) circRNA degradation (DBR1, RNase L, m6a modification, among others), and (3) release of circRNAs in extracellular vesicles [3].
CircRNA downregulation in CNS could be due to an altered function of RBPs, a common ALS hallmark [72]. Motor neurons silenced or mutated for FUS exhibit decreased levels of circRNAs [18, 19]. Whereas FUS WT binds to the flanking introns of circRNAs and regulates their expression, it seems that FUSP525L mutant could preferentially bind to introns with inverted Alu repeats flanking circRNAs which would inhibit circRNA biogenesis [19]. Decreased levels of CNS circRNAs both in the SOD1G93A mice and in sporadic ALS patients, as observed here, could be a consequence of a common etiopathogenic mechanism, such as misfolded proteins (FUS, TDP-43, SOD1). On the other hand, opposing alterations in circRNA levels between CNS and muscle could be due to the different response of neuronal and muscle cells to misfolded proteins in ALS [73–77]. It has been shown that muscle cells more efficiently eliminate misfolded proteins than neurons [73–76]. Thus, depending propensity for aggregate formation or management of misfolded proteins in different tissues, ALS-linked RBPs may exert distinct effects on CNS and muscle in ALS, which could differently affect circRNA biogenesis.
Notably, a possible therapeutic approach for ALS has been proposed based on the regulation of circRNA levels through DBR1 inhibition [25]. This enzyme catalyzes the debranching of lariat introns that are formed during pre-mRNA splicing, which may result in accumulation of cytoplasmic intronic circRNAs (ciRNAs) [2, 4]. The absence of DBR1 enzymatic activity is sufficient to rescue mutant TDP-43 toxicity in yeast, in human neuroblastoma cells and in rodent primary cortical neurons [25]. This likely occurs because the accumulated ciRNAs act as cytoplasmic decoys to titrate mutant TDP-43 relieving important cellular RNAs and RBPs that mislocalized TDP-43 normally sequesters from their normal function in ALS [25]. These results could support the hypothesis stated above, according to which neurons respond worse to misfolded proteins and would result in decreased physiological levels of circRNAs that could not be beneficial to the CNS. On the contrary, muscle may present an increase in the physiological levels of circRNAs, which could contribute to reduce the toxicity of misfolded proteins and show greater compensatory regenerative capacity to damage. In this regard, it is also possible that the observed decrease in circRNA levels reflects their increased degradation in the CNS of ALS model and patients. Although the degradation of circRNAs has hardly been studied, it has recently been shown that m6a methylation of RNA and activity of RNase L could promote the global elimination of circRNAs [3]. Interestingly, m6a hypermethylation of protein coding and non-coding RNA species have been observed in the spinal cord of sALS patients [78] and the expression of RNase L transcript is upregulated in iPSC-derived motor neurons from ALS patients [79]. In this respect, RNase L-mediated degradation of circRNA has been found to be necessary for the activation of protein kinase R (PKR) [80], which has been detected at higher levels in spinal cord [81] as well as in hippocampus from ALS patients [82], and its inhibition has been proposed as a promising therapeutic approach [82]. Therefore, increased levels of circRNAs in the CNS of ALS could be beneficial, as it could decrease TDP43-mediated toxicity [25] and PKR activation [80].
It is noteworthy that decreased expression of circRNAs in the CNS and increased levels in muscle has been observed in other pathologies and disease models. CircRNA levels are decreased in the spinal cord of spinal cord injury (SCI) models[83, 84] and in brain cortex of patients with schizophrenia [85], whereas increase is observed in myopathies such as DM1 [13, 14]. Based on these findings it may be plausible, regardless of the possible cause (RBPs, m6a methylation...), that changes in circRNA abundance could reflect a compensatory response to neuronal (e.g., neuroinflammation, reactive glia, oxidative damage, glutamate excitotoxicity, ADAR-mediated RNA editing) and muscle damage (muscle weakness and wasting) common to ALS and ALS mimic-pathologies. In this regard, it has been observed that there is a connection between the deregulation of global circRNA abundance and immune microenvironment [86, 87].
During the preparation of this manuscript, a comparable tissue-dependent pattern of circRNAs observed in CNS and muscle of ALS patients was reported by Tsitsipatis [21]. While the mentioned study examined different circRNAs from those investigated in the present article (with the exception of circFBXW7), their findings and conclusions aline closely with ours. Interestingly, some of the circRNAs that we selected through a literature review independent of the Tsitsipatis work overlap with the data of their sequencing analysis in muscle biopsies from ALS patients (circXPO1, circSCLT1, circSPECC1, circMGA, circHIPK3, circUBXN7, circSAMD4A and circHIPK2). Three of these circRNAs were not detected in the quadriceps muscle from SOD1G93A mouse model, possibly due to lower expression levels and different methodological sensitivity. Nevertheless, considerable overlap between the results and general conclusions of these two studies strengthen the value of these findings and reinforce the importance of circRNAs in ALS pathology.
Our results corroborate previous findings regarding the relatively high degree of conservation of the circRNAs between mouse and human, and how the altered expression profiles in most cases are captured in both species (Fig. 6) [21]. While it is true that many of the circRNAs differentially expressed in SOD1G93A mice did not have human homologs (or these had not yet been annotated due to the relatively recent discovery of this RNA class), it is important to note that in certain cases with human homologs, the findings obtained from the mouse model were also observed in human ALS samples. This emphasizes the potential value of animal models in translating experimental findings to the benefit of patients. It is also worth noticing that the circRNAs that were not observed to be dysregulated in the patient samples could be because they are altered in the spinal cord but not in the brain. Therefore, although the same trend of decreased circRNA levels in CNS tissues has been observed in ALS, there could be some circRNAs that are downregulated only in the spinal cord or brain.
Finally, in this work we have also studied the differences in the expression of circRNAs according to tissue and sex. The levels of circRNA were found to be higher in the spinal cord than in the skeletal muscle, regardless of the mouse genotype (Fig. 4). This is reasonable since neurons are by far the cell type with the highest number and expression of circRNAs [9]. Therefore, although in ALS circRNA abundance could be decreased in the spinal cord and increased in muscle, their levels would still be higher in the CNS because neurons are enriched in these RNA molecules. This fact could also explain why all the circRNAs we have studied were expressed in the spinal cord but only slightly more than 50% could be quantified in the quadriceps (Fig. 6). The vast majority of circRNAs that we have only been able to quantify in the spinal cord come from circRNA-seq analysis, so it is possible that they could be tissue-specific or highly enriched in the CNS. Moreover, some circRNAs that we selected for their known relevance to the muscle (such as circZNF609 and circSAMD4A), were also found in spinal cord (Fig. 6). As for sex, the expression patterns of circRNAs were maintained in spinal cord and skeletal muscle of SOD1G93A females (Fig. 3), but these variations were more pronounced in males (Figs. 1 and 2). This could be due to sex-dependent differences observed in ALS, since SOD1G93A females tend to have a later disease onset and display modestly increased survival [88]. Consistently, the alterations of lesser extent observed in the end stage females could reflect their better functional status compared with males. Indeed, the levels of circRNAs in females at terminal stage (P120) are more comparable to those obtained for males at late symptomatic stage (P105) in both in quadriceps and spinal cord (Fig. 6).
From the 30 circRNAs studied, 16 were found to be particularly altered in ALS and may therefore be closely related to the etiopathogenesis of this disease (circSCLT1, circXPO1, circUBXN7, circSAMD4A, circMAPK1, circANKRD12, circMAN1A2, circMORC3, circTTBK2, ciRS7, circHIPK3, circASH1L, circHIPK2, circFBXW7, circMGA and circCWC27) (see Table 1 for more details about circRNAs). Some of these circRNAs has been previously associated with other neurodegenerative diseases or myopathies, such as AD [11, 62], PD [52, 66] and DM1 [13, 14]. Ten of these circRNAs have been mechanistically studied, being some involved in inflammation [51, 54, 56, 62, 65, 66, 68], myogenesis [8, 64, 67], and cell cycle and proliferation [39, 60]. In addition, many seem to be involved in the brain development [27]. The function of the remaining circRNAs is still poorly understood. Nevertheless, their host genes are involved in several processes altered in ALS, and some have been identified as modifiers of more than one ALS-causal genes [42]. Interestingly, proteins encoding some host genes have been proposed as therapeutic targets for ALS (XPO1 [43], MAPK1 [57], TTBK2 [28] and HIPK2 [89]). On the other hand, circASH1L was significantly upregulated in the quadriceps from the asymptomatic to the terminal stage (Fig. 2, Fig. 6), so it could be a promising ALS biomarker. Remarkably, its host gene (ASH1L) is involved in myoblast fusion and neuronal morphogenesis and its expression is dysregulated in ALS [48–50]. Other potential biomarkers of ALS and/or muscle damage could be circXPO1, circANKRD12, circMAN1A2, circHIPK2 and ciRS7. Regarding circMAN1A2 (hsa_circ_0000118), it should be noted that another circRNA from the same host gene (hsa_circ_0000119) was found previously altered both in muscle and in spinal cord [21].
In conclusion, the expression of circRNAs in ALS could be characterized by a progressive and tissue-dependent pattern that seems to be opposite in CNS and skeletal muscle (Fig. 6 and Fig. 7). These variations in circRNA levels could be a common signature to all or different forms of ALS and seem to be indicative of neuronal and muscle damage caused by disease progression. This phenomenon could be due to alterations in the biogenesis of circRNAs, being their main players dysregulated in ALS (RBPs [72] and ADAR-mediated RNA editing [90]). Other independent and non-exclusive possibilities to this hypothesis may be an alteration of circRNAs stability (DBR1 [25], RNase L, m6a methylation [78]) or an increase of circRNAs transport from the motor neuron to the muscle [21]. In view of all these findings, the role of circRNAs in the etiopathogenesis of ALS could bear relevance from the earlier stages of the disease, paving the way to consider them as potential diagnostic and prognostic biomarkers. More studies are also needed to further investigate the function of these circRNAs as therapeutic targets of new treatments, as well as their expression in more accessible tissues, such as peripheral blood or plasma, for use as molecular markers of ALS.