Bioinformatic Analyses of the Ataxin-2 Family Since Algae Emphasize Its Small Isoforms, Large Chimerisms, and the Importance of Human Exon 1B as Target of Therapies to Prevent Neurodegeneration

1. Introduction

Ataxin-2 was discovered as the disease protein responsible for Spinocerebellar Ataxia type 2 (SCA2) [1,2,3], and became widely known because its deletion mitigates the disease progression of TDP-43-triggered Amyotrophic Lateral Sclerosis (ALS) or Fronto-Temporal Dementia (FTD) [4,5], and perhaps also a Tau-triggered form of Parkinson plus syndrome that was named Progressive Supranuclear Palsy (PSP) [6,7,8,9]. SCA2 is a rare neurodegenerative process with autosomal dominant inheritance, starting in the spinal cord, brainstem, and cerebellum and leading to a typical combination of muscle cramps, deficient locomotor coordination, intention tremor, and slowed saccadic eye movements, so that patients over the course of 6-25 years become wheelchair-bound, bedridden, and die from swallowing problems [10,11,12,13,14]. The ages of clinical manifestation occur earlier in successive generations of a pedigree, and the molecular cause of this “anticipation of onset age” is the expansion of a polyglutamine (polyQ) repeat domain near the N-terminus of Ataxin-2 [15,16,17]. In most healthy individuals worldwide, this domain would contain 22Q, normally encoded by (CAG)₈-CAA-(CAG)₄-CAA-(CAG)₈ [18]. The CAA interruptions in this repeat stabilize the length during genome replication [19], and once they are lost, the DNA polymerase will have difficulty in copying the exact length, leading to somatic mosaicism, anticipation over generations, and eventually the appearance of a pure CAG-repeat of excessive size that shows neurotoxicity with accumulating properties [20,21,22,23]. At intermediate sizes of 31Q to 33Q, this neurotoxicity becomes relevant in a polygenic manner, affecting only the most vulnerable neurons (spinal and cortical motor neurons in ALS and FTD, respectively, or the nigrostriatal dopaminergic pathway in Parkinson’s disease and PSP) [4,24,25,26,27,28,29,30,31]. From the threshold of 34Q upward, the neurotoxicity is strong enough to act in a monogenic manner with increasing severity in the multi-system neurodegeneration known as SCA2, either with spinocerebellar deficits upon adult onset [32,33,34], or with developmental delay and seizures upon pediatric onset caused by very long polyQ domains [35]. Expansions where one CAA interruption is still present have been reported several times to be associated with a clinical picture of Parkinson’s disease or of ALS [36,37,38,39,40]. Preliminary data mainly from animal models also suggest that Ataxin-2 modifies the disease course of other spinocerebellar ataxias (SCA1 / SCA3), adult motor neuron disease (ALS/FTD), and tauopathy-associated Parkinsonism [4,5,6,7,9,16,28,41,42,43,44,45,46,47].

However, it is completely unclear how Ataxin-2 exerts this role as an important modifier gene for neurodegenerative processes. Ataxin-2 localizes normally to the cytoplasm. A minor quantity of Ataxin-2 associates with the Golgi apparatus [48,49,50], modulates endoplasmic reticulum (ER) dynamics [51], and associates with the endocytosis complex at the plasma membrane [52,53], where it mediates the fast recycling of the Notch receptor [54]. Thus, the role of Ataxin-2 for membrane dynamics was documented early on but was not investigated intensely, while an additional role of Ataxin-2 for RNA has caught most of the attention. The main quantity of Ataxin-2 is present at the rough ER, where translation of membrane proteins and secreted proteins occurs [55], and associates with polyribosomes [56]. Some Ataxin-2 was reported to interact with actin and fimbrin [57,58,59], and furthermore, some evidence in favor of microtubule association was published [8,60,61,62,63], so a role of Ataxin-2 for cell protrusions is apparent. The diffuse distribution of Ataxin-2 throughout the cytoplasm changes dramatically in periods of cell stress, when it relocalizes with damaged RNA to the cytoplasmic stress granules, where RNA surveillance and triage occur [64,65,66]. Particularly oxidative stress is known to cause the maximal recruitment of RNA and translation machinery to stress granules, and one report suggested that Ataxin-2 influences the fate of superfluous mitochondria as the main source of oxidative stress [67], a notion that is strengthened by evidence that Ataxin-2 controls the expression levels of PINK1 as the protein kinase which phosphorylates PARKIN to initiate the autophagic elimination of mitochondria, and that Ataxin-2 associates with PARKIN as the ubiquitin ligase which tags mitochondrial outer membrane proteins as targets of removal [68,69,70].

It is still unclear how these localizations reflect Ataxin-2 functions and interactions in specific pathways. As novel insights, this study analyzes Ataxin-2 orthologs to document isoforms and the evolution of different domains and regions, thus elucidating their regulation and functional context. The discovery of extra domains that were integrated with Ataxin-2 into chimeric proteins during evolution, and their explanation as transcriptional readthrough across neighbor genes, triggers the notion that mutations in Ataxin-2 have complex effects on its isoforms, but also on the function of genomic neighbors.

The bioinformatic efforts in the current manuscript must take into account how multi-domain proteins such as Ataxin-2 became necessary, and how extra-domains are still getting added even in vertebrates. Recent studies in Euglena gracilis microalgae observed that cells with additional energy from endosymbionts can increase their cytoplasmic volume and show highly flexible cell shapes [71]. However, the diffusion of substrates, cofactors, and proteins becomes insufficient for efficient metabolism in such larger cells. This causes selection pressure that favors assembling multi-component complexes non-covalently, or even better, placing several domains from single pathways within one protein, rather than relying on diffusion. In this manner, the assembly of the LSm domain, LSmAD, PAM2 motif, and IDRs into one large protein would maximize the throughput in a chain of reactions, which were important for eukaryotic life. The classical domain combination that constitutes Ataxin-2 is clearly able to respond to oxidative stress and to manage some form of damage control, for RNAs and possibly other vulnerable compounds.

Importantly, diverse eukaryotic species live in different niches, so they had to adapt their metabolic processes further. Therefore, additional domains were appended to the core features of Ataxin-2 in a minority of organisms. The spectrum of variation between these additional domains and functions reflects changes in stressors, as well as alternative stress responses among eukaryotic phyla. In our view, available data on this “tinkering of nature” [72,73] can be exploited to understand the core and accessory mutation effects not only for Ataxin-2, but for any poorly understood protein. The compilation of such additional domains and the analysis of statistically enriched features among them might contribute to understanding how nature modifies and optimizes the core role of an enigmatic protein like Ataxin-2. This approach could elucidate how Ataxin-2 impacts several overlapping pathways. The wealth of sequencing data on species from all kingdoms of life, which is available in databases such as UniProt/UniParc, NCBI/GenBank and EMBL, can provide many novel insights into physiological connections between metabolic responses and the specific abiotic stressors in each habitat. This approach consumes enormous amount of time when performed manually, but should be amenable to automation, and is applicable to many scientific questions.

2. Results

2.1. Overall Structure of Ataxin-2 Family Members Across Evolution

The multi-domain protein Ataxin-2 family has three conserved sequence motifs in a defined order [74,75,76]. Their structures are dominated by the conspicuous LSm fold (see https://alphafold.ebi.ac.uk/entry/A0A370Q1D4, or https://www.rcsb.org/structure/AF_AFA0A044URM1F1, both last accessed on 1. December 2025), with the LSm-associated domain (LSmAD) also standing out in alpha-fold predictions, while most remaining Ataxin-2 is characterized by intrinsically disordered regions (IDRs), where the linear PAM2 motif is usually embedded within the C-terminal half. The three successive domains - surrounded by IDRs and in less than 5% of species by extra domains of known function - are shown schematically in Figure 1a and are meticulously assessed in text paragraphs below.

Many functional insights are known for two of the three highly conserved domains, but the third least studied domain, was found to be characteristic of Ataxin-2:

(1)

Firstly, just after the ancestral N-terminus, the Like-SM (LSm) fold has a typical structure of small five-strand anti-parallel beta sheets (β1, β2, β3, β4, β5), with SH3-type barrel tertiary structure. The ancestral quaternary structure is characterized by assembly into an LSm hexameric or heptameric ring, with U-rich RNA oligonucleotides binding inside the LSm torus lumen. Overall, the LSm domain in human ATXN2 and ATXN2L (synthetic full-length entries UniProt Q99700 and Q8WWM7) comprises 78 amino acid residues. LSm domains have descended from bacterial proteins like Escherichia coli Hfq and YlxS, as well as archaeal Sm1 / Sm2, which serve as RNA chaperones and in ribosomal pathways [77,78]. In eukaryotes, various Sm proteins in the nucleus form a heteroheptameric ring that is crucial for intron splicing [79]. Like-Sm proteins were subsequently also described in the cytoplasm, where LSM2-16 combine the RNA-binding sequence with a methyl-transferase domain [80,81], whereas Ataxin-2 combines the RNA-binding sequence with the LSmAD and the PAM2 motif [76,82]. Our datamining effort confirmed that the LSm sequence of Ataxin-2 has relevant similarity to the LSm domains of LSM2-16, so BlastP searches frequently confuse different families, and the sequence variability of the LSm domain is so strong in low organisms that current InterPro and Pfam algorithms fail to detect it in approximately one third of Ataxin-2 orthologs. The LSm/LSmAD region in Ataxin-2 binds to the RNA helicase DDX6 [64]. Together with the RNA helicase DDX6 and the LSm-containing factor LSM12, Ataxin-2 was found to influence circadian posttranscriptional regulation and olfactory habituation in fly neurons [83,84]. It is therefore important to note that an RNA helicase domain was chimerically added to Ataxin-2 orthologs in several species (see below for individual protein database entries).

(2)

Secondly, after a disordered bridge region of usually 50-60 amino acids, the LSmAD sequence stands out with a predicted alpha-fold structure. However, experimental analysis showed only a modest presence of α-helical structural elements, with a considerable degree of flexibility, devoid of tertiary structure and without RNA binding capacity [77]. Human Ataxin-2 LSmAD sequence (amino acids 409-477 in the synthetic full-length UniProt entry Q99700) contains a putative clathrin-mediated trans-Golgi signal (residues 414–416) and an ER exit signal (residues 426–428). Indeed, experimental analysis confirmed that deletion of 42 residues within LSmAD causes Golgi dispersion [50]. Overall, the ancient protein module comprising LSm and LSmAD with their connecting bridge sequences extends across some 250 amino acid residues in a very stable size across evolution, while most length variability of Ataxin-2 orthologs is due to IDR composition and length across the C-terminal half, and sometimes in short N-terminal regions. Here, it is important to note that our datamining effort found practically all LSmAD-containing sequences to represent Ataxin-2 orthologs, so the extremely well-conserved LSmAD domain is the unique characteristic feature of the Ataxin-2 family, and perfectly suited for the BlastP search for orthologs.

(3)

Thirdly, the 14 residues-short linear motif known as PAM2 was named as Poly(A)-binding protein interacting Motif 2. It connects to an MLLE sequence in the PABP C-terminus (also known as CTC, short for carboxy-terminal conserved domain), in dependence on nearby phosphorylation sites [62,78,79,80,81]. In plants, the PAM2 motif extends over 19 residues that contain a tandem duplicate of the core sequence [76]. Interestingly, its location is always outside globular domains [82], at approximately three-fifths of the protein length. It clearly functions to interact with mRNA 3’ tails, and it exists as a component of over a dozen different eukaryotic proteins [82], several of which are known for their regulation of mRNA translation versus decay [79]. These protein families include PAIP1/2, LARP4, eRF3/GSPT1/2, TTC3, USP10, PAN3, GW182, Tob1/2, and other factors, so our datamining effort found its usefulness for Ataxin-2 ortholog searches to be limited, and particularly in low organisms the sequence variability of this motif makes its automated recognition by current InterPro and Pfam algorithms doubtful. The PAM2 motif was shown to prevent the phase separation of Ataxin-2 in cellular growth periods, while it localizes to the translation apparatus at the rough ER, promoting the relocation of Ataxin-2 to stress granules after cellular damage [55,56,83].

2.2. Compilation of Ataxin-2 Family Protein Sequences Until Excavata, Amoebozoa, and Algae

Conservation of Ataxin-2 sequences across evolution was used to judge the relevance of each domain and feature for physiology and pathology. Using the BlastP algorithm to search UniProt-UniParc, NCBI, and EMBL databases for ATXN2 homologs, the longest or otherwise relevant isoforms for many organisms were compiled in Table S1, but it was impossible in the current study to compile all isoforms and assess their genomic coding sequences. Separate tabs in this table show either an overview of the longest orthologs (Table S1a), or specifically the N-terminal intrinsically disordered domain 1 (IDR1) before the polyQ domain (Table S1b), sequences across the polyQ domain with its flanking residues and interrupting residues (Table S1c), the IDR1 sequences around and after the polyQ domain (Table S1d), structured sequences around and after the LSm motif (Table S1e), structured sequences around the LSmAD motif (Table S1f), the intrinsically disordered domain 2 (IDR2, Table S1g), sequences across the PAM2 motif within IDR2 (Table S1h), ordered sequences before and after IDR3 with C-terminal variability (Table S1i). Some ATXN2L orthologs were compiled in Table S2. While this compilation cannot be exhaustive, it covers the main classes of eukaryotic organisms.

It is difficult to pinpoint the origin of Ataxin-2, given that precursors of Ataxin-2 in the databases cannot reliably be distinguished from fragmented Ataxin-2 due to sequencing problems, proteolysis, or isoforms. Furthermore, the absence of Ataxin-2 orthologs from species or organism classes may simply be due to incomplete sequence documentation, so only crude conclusions are possible. But as the prime result, Ataxin-2 orthologs were clearly identified in all eukaryotic kingdoms until amoebozoa, protozoa, and algae, but not in archaebacteria or eubacteria. Details are provided in the following paragraphs of this chapter.

Figure 1b depicts the evolution and domain structure of Ataxin-2 proteins in all kingdoms of life. Among amorphea-amoebozoa, Ataxin-2 homologs as in UniProt entries A0A151ZGE7 and Q55DE7 with 900-1100 aa in Dictyostelium are already complete with N-terminal polyQ stretches, LSm, LsmAD, and a typical PAM2 motif recognized by the InterPro or Pfam algorithms.

Among protozoa, Ataxin-2 homologs like A0A1Z5KDA2 and A0A9N8E681 with 700-900 aa in stramenopiles (the most ancient among them have chloroplasts surrounded by four membranes as a testimony to secondary endosymbiosis [84]) have this multi-domain structure and C-terminal polyQ stretches. In phaeophytes (brown algae), as a photosynthetic class among stramenopiles, UniProt-UniParc entries A0A6H5L901, D7FVW6, and D7G1N4 in Ectocarpus species represent typical Ataxin-2 orthologs with short polyQ stretches upstream from LSm.

Interestingly, A0A0G4IZN6 with 400 aa in rhizaria contains a structured sequence not recognized as LSm by current algorithms, then LSmAD, IDRs with PAM2 embedded (PAM2 sequence takpkLNANAKTFT-MSAAAKEFV), but no polyQ stretch.

In naegleria from the excavata/discoba-schizopyrenida branch (that alternates between amoeboid and flagellate stages, having mitochondria with discoid cristae similar to Euglenozoa [85]), the UniProt entries A0A6A5CC11, A0AA88H5V6, and D2V0R5 with about 900 aa represent complete Ataxin-2 orthologs with N-terminal polyQ stretches.

In the closely related euglenozoan branch (the only photosynthetic branch among the excavata), smaller complete orthologs are documented. A0A0L1KIJ6 with >800 aa in Perkinsela shows a potential LSm coexisting with LSmAD and a primitive PAM2 motif (LNPNATAFLP) without polyQ stretches; A4H699 or A4HUM3 or Q4QHA3 in Leishmania (potential PAM2 precursor core sequence PNPSATPFVP), A0A1X0NQC5 in Trypanosoma (potential PAM2 core FNPAATPYTP), W6KMY5 in Phytomonas (potential PAM2 core PNPAAAPFVP), A0A0M9FU89 in Leptomonas (potential PAM2 core PNPSATPFVP), all with >500 aa, show a LSm and LSmAD coexisting with a C-terminus that contains several polyQ stretches. Trypanosomal A0A061J3B7 with 279 aa, and A0A0S4IS39 with 213 amino acids, show LSm and LSmAD coexisting without further elements. Given that trypanosomes were reported to have exchanged genetic material with green algae and other archaeplastida [86,87], orthologs of Ataxin-2 were studied in detail among algae.

Ataxin-2 orthologs are also documented among alveolata (A0A813ABV5 with >1100 aa that contain an extra 2-hydroxyacid dehydrogenase domain, with homology to bacterial SerA protein that is essential for the biosynthesis of L-serine as precursor of phospholipids and sphingolipids). Among cryptista, one protein with significant LSmAD homology is documented (L1I7C3 with 522 aa).

In contrast, Ataxin-2 homologs appeared absent from the diplomonad and parabasalid-metamonad branches of excavata protists, which lack mitochondria, so these protists depend on nutrients and energy from hosts. In the monocercomonoides species from excavata, endosymbiont mitochondria were completely lost again [88], and an Ataxin-2 homolog could not be detected, while the microsporidia that lost mitochondria but retained mitosomes with FeS-metabolism-dependent oxidative stress [89] did preserve the presence of Ataxin-2 orthologs (e.g., A0A4Q9L9Z0 and A0A086J3B3). These findings support the concept that Ataxin-2 is needed to protect cells from the oxidative stress emanating from endosymbionts.

Among algae, many entries for Ataxin-2 orthologs exist from chlorophytes (green algae) with well above 1000 aa (e.g., A0A8J4CZB7 or A0A383VPF7) that contain LSm, LSmAD, polyQ stretches shortly after LSmAD, and typical PAM2 motifs embedded in the long C-terminal IDRs, as well as a short N-terminal IDR, in close structural and length similarity to human ATXN2L and ATXN2. In the particularly interesting green algae Cymbomonas tetramitiformis, whose ancestors are thought to have played a crucial role in the primary endosymbiosis of plastids [90], A0AAE0L4G1 is documented as a typical Ataxin-2 ortholog of 668 aa without polyQ stretch. Similarly, among archaeplastida-rhodophytes (red algae), the three entries A0A1X6NXA5, A0A5J4YT85, and A0A2V3IVX2 (noteworthy for a C-terminal uninterrupted 30Q repeat) represent Ataxin-2 orthologs.

However, no Ataxin-2 homologs were detected in glaucophytes (green-blue algae, which have plastids surrounded by an inner membrane containing a remaining peptidoglycan layer, suggesting they are still synthesized by plastids as by ancestral cyanobacteria, having mitochondria with the typical eukaryotic flat cristae surrounded by a double membrane, and using asexual reproduction). They are considered the basal archaeplastida that engulfed cyanobacteria in an act of primary endosymbiosis, within a terrestrial freshwater ecosystem [91,92,93,94,95]. Therefore, Ataxin-2 orthologs apparently were not needed for the engulfment of mitochondria or plastids. Instead, the appearance of Ataxin-2 might be associated with the gained ability to synthesize robust inner membranes by eukaryotes rather than by plastids. Apparent absence of Ataxin-2 homologs was also observed in the mesostigma branch of green algae, which contains primitive plate-like chloroplasts, lacking a periplastidal ER [96,97], and was therefore placed at the basis of divergence between chlorophyta (green algae) and streptophyta (land plants). This finding agrees with the notion that Ataxin-2 presence correlates with eukaryotic synthesis of organellar membranes.

Overall, the appearance of Ataxin-2 ancestors occurred only after glaucophytes had developed the endosymbiosis of mitochondria and primary endosymbiosis of plastids (Figure 1b). Ataxin-2 ancestors are only documented since rhodophytes learnt to cope with the oxidative stress due to sun exposure and mitochondria/chloroplast presence [98]. Rhodophytes still have chloroplasts without external ER or unstacked (stroma) thylakoids, but developed α- and β-carotene (provitamin-A), lutein, and zeaxanthin as new antioxidative pigments [99], and reproduce sexually, while lacking centrioles and flagella. Carotenoids play a basic role as membrane reinforcers in view of their rigid conjugated double-bond backbone, as phylogenetic precursors of sterols [100,101]. When marine life forms subsequently moved to freshwater habitats and underwent terrestrialization, duplications of the Ataxin-2 gene occurred (Figure 1c).

2.3. Gene Duplication ATXN2 / ATXN2L in Animals, and CID3 / CID4 in Plants, Upon Entering Freshwater and Land

Datamining with STRING homology searches revealed that the gene duplication giving rise to land animal ATXN2 versus ATXN2L proteins is not only documented from mammals via Xenopus laevis to Danio rerio but is documented during earlier evolution until lampreys (UPI0014023775 vs UPI001403D160 in Petromyzon marinus, UPI002AB6496F vs UPI002AB6167E in Lethenteron reissneri) (Suppl. Material S1). Lampreys start life as larvae in freshwater, irrespective of whether they later live in coastal sea areas, rivers, or lakes. ATXN2 orthologs among fish species share more similarity with each other than with ATXN2L orthologs (Figure 2a), with specific residues being characteristic of ATXN2 versus ATXN2L (Figure 2b). The Pro vs Ser difference at the beginning of the L3 sequence within the LSm fold would cause a bend in the peptide chain and possibly result in a beta-turn only in ATXN2 orthologs, whereas the Ser/Ala vs Pro difference in the center of L5 may cause such a bend or beta-turn only in ATXN2L orthologs (Figure 2b). This suggests that different interactors and functions for ATXN2 versus ATXN2L have evolved. Thus, two copies of Ataxin-2 with differing point mutations are first detected in chordate lampreys as jawless fish that derive from the Devonian period, some 360 million years ago [102]. The gene duplication without distinguishing sequence variants may also be present in other chordates, like hagfish. Indeed, three rounds of genome duplication events are known to have occurred during fish evolution, one of them during the Devonian [103]. With many genes duplicated, chordates were enabled to develop notochords and neural tubes with dorsal versus ventral specifications, which are established via signaling with retinoic acid (a carotenoid vitamin-A derivative) to control homeobox expression and somite formation [104,105,106,107,108,109,110,111], a pathway that establishes anterior-posterior gradients and is particularly pathway for the development of hindbrain, cerebellum and spinal cord that are affected most strongly in SCA2 and ALS [112].

The question of whether ATXN2 or ATXN2L is the duplicated copy was assessed by genomics. The relevant human chromosome locus on 12q contains the gene order PPP1CC – CCDC63 – MYL2 – VHRT – CUX2 – PHETA1 – SH2B3 – ATXN2 – U7– BRAP – ACAD10 – ALDH2 – MAPKAPK5 – TMEM116 – ERP29 – NAA25 (underlined genes share the same transcriptional direction; U7 refers to the 63 bp uridine-rich non-coding snRNA ENST00000607576.1, a potential binding target of Sm domains, which is positioned on the opposite strand downstream from polyQ-encoding exon 1, and upstream from exon 2 of human ATXN2, as the start of the LSm sequence encoding exon cluster), whereas the locus on 16p contains the gene order SULT1A1 – NPIPB8 – EIF3C – NPIPB9 – ATXN2L – TUFM – SH2B1 – ATP2A1 – RABEP2 –CD19, according to the human genome annotation in the UCSC genome browser. The synteny of ATXN2 with SH2B3, and of ATXN2L with SH2B1, is conserved since Nile tilapia fish. Before the gene duplication event, in fish such as Callorhinchus milii (elephant shark) or Oryzias latipes (medaka), the genome contains only the SH2B3 – ATXN2 – BRAP locus. It is tempting to speculate about a functional interaction of ATXN2 with SH2B3 and ATXN2L with SH2B1, given that all these proteins contain proline-rich motifs and associate with GRB2 during PI3K/AKT growth signaling at neuronal membranes [53,113,114,115,116]. Such a potential competition for GRB2 would be a relatively recent development during evolution, given that SH2B3 is no neighbor gene of ATXN2 in the genomes of lancelet fish, D. melanogaster flies, or C. elegans nematodes, and is therefore not crucial for the ancient functions of Ataxin-2. However, it is conceivable that SH2B3, ATXN2, BRAP are jointly relevant for human diseases such as SCA2, ALS, cancer, and longevity, either via their gene variants in linkage disequilibrium, or via protein interactions [29,117,118]. Overall, ATXN2 clearly evolved from ancient Ataxin-2 in its original genomic context, while ATXN2L represents the duplicated copy in a new genomic environment.

ATXN2L is thought to be missing in birds. However, non-flying but feather-bearing kiwi birds in northern New Zealand have ATXN2 and ATXN2L entries in UniProt (Apteryx mantelli A0ABM4FEG3 and A0ABM4G1B9). Apparently, ATXN2L got lost before the evolution of flight, because ostriches, emu, and other kiwi species do not record ATXN2L sequences. It is known that large segmental deletions occurred when the macroevolution of avian genomes favored constrained size and became relatively static [119].

Among plants, two Ataxin-2 homologous gene copy translation products are not only documented in eudicot brassicales such as Arabidopsis thaliana (Q94AM9 vs Q8L793) and Capsella rubella (R0HKF4 vs R0GVL5) and other flowering land plants, but already in paleodicots / ANA-grade angiosperms such as amborellales (U5D2G8 vs U5D208 in Amborella trichopoda) and in gymnosperms such as conifers-araucariales (UPI0005EA7401 vs UPI0005EA70B3 in Araucaria cunninghamii) (Suppl. Material S2). The ancestral angiosperms, like amborellales as well as gymnosperms, underwent enormous diversification after two whole genome duplication events since 319 million years ago during the late carboniferous period [120], which was possibly the ancestral event where CID3/CID4 appeared. While the Ataxin-2 orthologs in conifers still have >800 amino acids, similar to chlorophyte Ataxin-2 orthologs, CID3 and CID4 in flowering plants show a length of around 600 residues, consistent with the notion that some of their stress resistance properties were lost during evolution to flowering plants due to unusually short IDRs at either end of the proteins.

Conifers were the first trees among land plants. They are still the dominating plant species, e.g., in the taiga of Siberia, where they show exceptional resilience to winters with temperatures below the freezing point, managing to stay evergreen. Their cold hardiness involves the sensing of low temperatures and a short photoperiod, to trigger molecular adaptations, such as (i) improved membrane fluidity, with the development of epicuticular wax (composed mainly of nonacosane, which is used as a component of paraffin, as 29-carbon straight-chain alkane that is derived from very-long-chain fatty acids [121]), (ii) the mobilization of solutes such as glycogen to lower the freezing point and minimize cold shock of vulnerable RNA, (iii) downregulation of photosynthesis and cell growth in parallel to upregulated expression of cold-induced proteins [122].

Importantly, there are no specific residues that are characteristic of either CID3 or CID4. Instead, both gene copies in each organism were more similar to each other than to the orthologs in other species (Figure 2c). Thus, they appear to have co-evolved in different organisms, adapting similarly to the changes of interactor molecules in dependence on various ecological niches. Apparently, both were conserved and remain expressed because a higher gene dosage was beneficial, but they didn’t assume different roles.

A curious exception to the high similarity of both Ataxin-2 orthologs within each species is observed in amborellales, where one gene copy has a sequence with homology to flowering land plants. Amborellales are known to undergo extensive horizontal gene transfer from neighboring plants [123], and this might provide an explanation.

Overall, events that duplicated the Ataxin-2 gene occurred twice during macroevolution while moving out of marine life, with both duplications becoming conserved with independent fates. On the one hand, one event among chordates of shallow waters was inherited by land animals, resulting in ATXN2 versus ATXN2L acquiring separate functions reflected by different protein lengths. On the other hand, another Ataxin-2 gene duplication occurred among conifers as early tree forms with unusual stress resilience, resulting in CID3 versus CID4, which lost some protein length and did not develop consistent distinguishing features. This is compatible with the notion that mechanical stress due to gravitational forces, more sunlight intensity, heat and cold exposure, and oxidative stress made it advantageous to have two gene copies when marine life forms underwent terrestrialization. The fact that flying birds lost the Ataxin-2 gene duplication might suggest that the higher weights of organismal bodies on land and the mechanical force of waves in shallow water require strengthened cell membranes. In contrast, life-forms that are suspended by buoyant forces in deep ocean water or in the air have less demand on membrane stability.

A comparison of the LSm and LSmAD sequences in human ATXN2 versus ATXN2L reveals very similar predicted structures, while pinpointing the putative crucial differences (Figure 3a/b). Additional comparison of the polyQ domain and the LSm domain in the human species versus their first appearance in the red alga Gracilariopsis chorda underlines the strong conservation of predicted structures, and presumably also their functions (Figure 3c/d).

2.4. Genomic Comparison of Exon-Intron Structure for Human and Murine Ataxin-2 Versus Ataxin-2-like

Mapping of the individual exons within genomic sequences (as documented at https://www.ncbi.nlm.nih.gov/gene/6311, last access on August 15, 2025) shows that human ATXN2L exons on chromosome 16 (Suppl. Material S3) stretch across 13,823 bp, with large blocks of non-coding sequences after exon 1, then (with the LSm domain encoded by exons 3-5) large non-coding blocks again after exon 6, then (with LSmAD encoded by exons 7-8) after exon 10, and again (with PAM2 motif encoded by exon 15) after the final exon 24B.

Human ATXN2 exons on chromosome 12 (previously studied in [124]) have a similar distribution overall (as shown at https://www.ncbi.nlm.nih.gov/gene/11273, last access on August 15, 2025), but they are found across a >10-fold larger genomic area with 146,713 bp (Suppl. Material S4), suggesting that much more sophisticated control over expression and splicing of exon clusters has become possible. Massive blocks of non-coding sequences are found after exon 1 (where the polyQ repeat is encoded), then (with the LSm domain encoded by exons 3B-5) again after exon 5, a smaller block after exon 6, then (with LSmAD encoded by exons 7-8) smaller blocks each after constitutive exon 9, alternative exon 10, constitutive exon 11, and again massive blocks after exon 14 (encoding IDR2, with the PAM2 motif encoded by exon 16), exon 18A, exon 18B, exon 20, exon 21, and the final exon 25.

The extraordinary similarity of translation starts, aligned exon boundaries (Suppl. Material S5, employing CLUSTAL Omega 1.2.4), and of the genomic clustering of several exons around the LSm, around the LSmAD, around the PAM2 motif, and the C-terminal intrinsically disordered sequences may reflect a joint regulation and separate functions in common for each cluster.

In addition to the ancient conserved and aligned exons, human ATXN2 has developed alternative splicing at least for exons 1A, 1B, 2, 3A, 10, 21, 23B, and 25 ([124,125,126], UCSC genome browser). Clearly, the regulation of human ATXN2 appears much more sophisticated than ATXN2L, given that is shows many more genomic non-coding control sequences between exons, at least three alternative translation starts ([127], UniProt, NCBI protein database) within the extended N-terminus containing the expanded polyQ stretch, many options for alternative splicing, longer sequences encoded by exons 11 and 13, and a much more variable C-terminus. Further analysis of murine Atxn2 versus Atxn2l confirmed the strong adherence to exon boundaries but also documented several alternatively spliced exons for Atxn2l (Suppl. Material S5, [128,129]). The fact that Ataxin-2-like remained relatively unchanged during evolution, while Ataxin-2 underwent numerous adaptations to fine-tune its functions, indicates that human Ataxin-2 has acquired additional features and that SCA2/ALS pathogenesis involves diverse mechanisms, which are not necessarily modelled in lower organisms.

Importantly, the human ATXN2 sequence shows two methionine residues as potential start codons at the initial LSm sequence, and database entries confirm that corresponding isoforms are translated. Similarly, the initial LSm sequence of ATXN2L orthologs contains one or two methionine residues. While such an Ataxin-2 translation start right at LSm was normal in ancient species like Trypanosoma, Perkinsela, and other protists, and was still producing the main isoforms among reptiles (Table S1a), a short intrinsically disordered N-terminal extension with IDR1 became usual already among protists, fungi, worms, and insects. An extension of the N-terminal IDR1 occurred progressively for ATXN2 homologs since fish, with one methionine at the LSm start being lost. However, there is database evidence that a minor isoform that starts from the LSm domain has always existed from zebrafish to primates (Table S1d).

2.5. C-Terminal Fragment Isoforms Are Prominent According to Exon Expression Analyses, and C-Terminal Epitopes Are the Target of Current Commercial Antibodies

Interestingly, RNAseq-based analyses of individual exon expression strength in central nervous tissue (as documented at https://www.gtexportal.org/home/gene/ATXN2, last accessed on October 8, 2025) reveal that the exons named 22, 23, 24A, and the C-term alternatively spliced final exon in Suppl. Material S6 has the highest score (in spinal cord 1.48 and 1.26, respectively, encoding C-terminal intrinsically disordered sequences with 7 Met residues). Only about half of these maximum expression levels are observed for exons 15 (containing 1 Met) with scores at 0.644 and exons 16 (encoding PAM2), 17 (3 Met), 18 (2 Met), 19 (3 Met) with scores above 0.8. Less than a quarter of these maximum expression levels are found for exons 5-12 (around the LSmAD sequence, including 1 Met in exon 5, 6 Met in exons 9-12), with scores above 0.407. Around a sixth of the maximum expression levels are apparent for exons 2-4 (encoding the LSm fold start, including two Met), with scores between 0.174 and 0.252. Particularly low expression at 3% of maximum levels is seen for exons 1A and 1B (encoding the N-terminal intrinsically disordered sequence and the polyQ domain, including 2 Met residues), with scores at 0.04.

Also, the GTEx Gene Model based on ENSEMBL-documented transcripts of human ATXN2 (as documented at https://www.gtexportal.org/home/gene/ATXN2, last accessed on October 8, 2025) indicates the existence of many transcripts that comprise only the C-term disordered sequences, or begin at sequences downstream from PAM2, from LSmAD, and from LSm. Transcripts encoding only the LSm domain are always combined with LSmAD, but transcripts encoding only the LSmAD domain (entry ENST00000471866.5) or the sequences between LSmAD and PAM2 (ENST00000546483.1) are documented. Overall, only half of the transcripts, including all alternatively spliced transcripts, span the length of the ATXN2 gene, a tenth spans only the N-terminal parts, while some 40% span across C-terminal parts. The N-terminal parts seem less relevant than models of 3D-structure around the conspicuous LSm-fold would suggest, while the C-terminal intrinsically disordered sequences have a little-appreciated importance. Therefore, despite the notion of synthetic full-length Ataxin-2 being a multi-domain protein of >1300 residues, cells appear to have retained during evolution a great ability to use most of its components selectively.

These transcript findings suggest that short protein isoforms may exist with a Met start codon in the center or towards the C-terminal end of human ATXN2, which were previously considered as partial sequence fragments. Indeed, database entry Q99700.2 reflects a translation start from N-term disordered sequence, NP_002964.4 reflects a start from the polyQ-repeat, NP_001297052.1 reflects a start from the LSm domain, but there are also entries like XP_059732367.1 where the start lies upstream from LSmAD, or EAW97959.1 / EAW97960.1 / A0A2K6UG79 / A0A8C5XNX2 / A0A811ZS54 / A0A8C4TUK8 / A0A3S2PRM8 / A0A3Q3N8E0 / A0A3Q3FSG1 / A0A672RDQ6 where the start lies downstream from LSmAD, or A0A2K5CKD3 where that start occurs downstream from LSmAD but the PAM2 motif is selectively excluded, or A0A7J8EWQ7 / A0A7J7R7F2 where the start lies just upstream from PAM2, or A0A7J7S175 / A0A7J8BWL6 / A0A5F5Y0Q1 / A0A212D9N5 / A0A5N4C9F7 / A0A5N4C9H8 / A0A401TLV8 where the start locates downstream from PAM2. A schematic depiction of the human ATXN2 protein sequence with potential methionine start codons, isoform candidates with UniProt-UniParc database entries, and the encoding exon structure is illustrated in Figure 4.

Commercial antibodies were often raised against full-length overexpressed Ataxin-2. However, in those cases where the companies used a fragment of Ataxin-2 as immunogen and revealed the epitope residues, a bias towards the more abundantly expressed epitopes downstream from LSmAD is obvious. Few commercial antibodies exist against the expanded polyQ domain, against IDR1 sequences, against the region across LSm or LSmAD, and against the sequences between LSm and LSmAD, while most targeted antibodies detect immunogens downstream from LSmAD (Figure S1 depicts the relevant immunogens relative to the exon structure of isoforms according to GTex database data). Reporting bias selects figures that represent the 120-130 kDa isoforms, while bands with smaller molecular weight are usually interpreted as unspecific cross-reactivity or are ignored. Thus, there is an important unmet need to study the C-terminal isoforms of ATXN2.

ATXN2 was reported to homo-dimerize or multimerize, and it is important to consider the fact that LSm-containing proteins assemble in homo-hexameric or homo-heptameric rings around RNA. Thus, these expression data would be compatible with a scenario where Ataxin-2 isoforms integrate into assemblies that are dominated by intrinsically disordered domains that promote RNP granule formation during microtubular transport. One sixth of the components would bear the LSm head that can be activated to bind a specific single-stranded AU-rich RNA sequence directly, and then mediate the association of a helicase to linearize the RNA, presumably in a stimulus-dependent manner at the end of transport, e.g., in a synapse before stimulus-dependent translation.

Given that the polyQ-encoding exon accounts for only 3% of maximum expression in the spinal cord, it seems plausible that there is a long temporal delay before polyQ-triggered accumulation and aggregation of N-terminal ATXN2 sequences become prominent in postmitotic neuronal tissue. In a mouse mutant with an ATXN2-Q100 expansion KnockIn, the levels of phosphopeptides from LSm to C-term in non-neuronal tissue are around 10% compared to WT [130], confirming that the expansion can lead to partial loss-of-function effects. However, in postmitotic neural tissue (the spinal cord) at 14 months of age selectively a phosphopeptide close to the polyQ-repeat before the LSm domain accumulates >110-fold relative to WT, and a phosphopeptide in the isoform that starts 160 residues even further upstream at the N-terminus accumulates >30-fold [131]. This suggests that either alternatively spliced expression of small isoforms or partial degradation via proteolysis is employed by cells, and the disease mechanisms in SCA2 and ALS do not necessarily concern the full-length protein. ATXN2-fragment-specific disease mechanisms also should be taken into account.

2.6. Most N-Terminal Start Codon with Subsequent Fragment Appears in Armadillo only for ATXN2

As mentioned above, the analysis of the ATXN2 start codons in humans and closely related primates shows at least 3 different translation starts across the N-terminus for isoforms in Homo sapiens, and 4 different isoforms in Pongo abelii as well as Macaca nemestrina (Table S1a). The most N-terminal alternatively spliced sequence starts with M(X)R(PX)S/T(XX)A… and is clearly conserved with quite constant length from primates until artiodactyla, and according to one database entry, even in armadillos (Table S1b), suggesting that it has functional relevance. This N-terminal extension is not observed for ATXN2L, so its function is specific to ATXN2. It starts with a sequence that has a strong likelihood to act as mitochondrial targeting signal (score of 0.9937 in MitoProt v1.101, with predicted cleavage site at residue 144; score of 0.75 at TPpred3 for residues 1-38), and then contains three proline-rich motifs (PRMs) at amino acid residues 55-71, 117-123, 142-148 in human ATXN2 (Uni-Prot Q99700).

Sequences with significant homology (BlastP score <5e-07) to this fragment are not found in ATXN2 orthologs among paenungulata and marsupialia, but instead are contained in the SKI family transcriptional corepressor 1 (SKOR1), Actin nucleation-promoting factor (WAS), and Splicing Factor 3B subunit (SF3B4) from Vombatus ursinus, Phasolarctos cinereus, and Monodelphis domestica. In humans, close homologies exist with sequences within Transcription initiation factor TFIID subunit 4 (TAF4, 1.1e-9) and the Splicing factor, proline-and glutamine-rich (SFPQ, 7.8e-8), but also with Proline-rich protein 36 (PRR36, 4.6e-8), Formin-2 (FMN2, 2.1e-7), and Actin-binding protein WASF2 (8.5e-7).

Structural prediction software in InterPro characterizes this sequence as an intrinsically disordered region (IDR) with low complexity and proline-rich stretches, and previous literature has named all sequences upstream from the LSm domain as IDR1 of ATXN2 [62]. Thus, the above observations are in agreement with a genome-wide survey that concluded Pro-rich sequences to be involved in actin/cytoskeletal-associated functions, RNA splicing/turnover, DNA binding/transcription, and cell signaling [132,133]. At the N-terminus, this proline-rich sequence may influence the folding of nascent ATXN2. Its presence in recombinant full-length ATXN2 constructs for in vitro experiments, as well as the absence of such N-terminal sequences from ATXN2-null organisms, may have contributed to reports that ATXN2 is involved in actin-plastin pathways and in actin-endocytosis pathways [52,53,57,58,134]. However, the effects of Ataxin-2 on mitochondrial biogenesis, mitophagy, and oxidative stress responses cannot be attributed only to this IDR1, given that such an impact is already documented in the yeast ortholog Pbp1 [67,68,69,70,135,136,137,138].

2.7. The Usual Start Codon in Human ATXN2 / ATXN2L Is Followed by polyQ and a Repeat-Rich Fragment, Which Elongates Since Yeast / Insects

The second alternatively spliced exon in ATXN2 encodes the highly conserved translation start sequence MS(L/M)K(P)n(Q)n(P)n to produce an additional N-terminal fragment of up to 50-120 aa that contains several repeats, namely polyQ, polyPro, polyAla, polyGly and polySer, with one or several prolines usually at each side of the polyQ repeat. Within this sequence downstream from the polyQ repeat and before the polySer repeat, 17 amino acid residues enable N-terminal proteolysis of mammalian ATXN2 [139]. The entry A0A2K5X824 in Macaca nemestrina suggests that an alternative start sequence MGPHHVAEAP(A)_nT(A)n can also be used, producing a fragment that contains a polyA repeat instead of the polyQ repeat. This fragment immediately before the LSm domain start does not show a relatively constant length, in contrast to the previous paragraph, but instead shows a massive extension across phylogenesis. A close inspection of the sequences compiled in Table S1d reveals the repeated appearance of polyQ repeats in testudines, again in monotremata, in proboscidea, and in primates. Instead of an N-terminal mammalian polyQ-repeat (MSLKPQQQQQQQQQQQQQQQQPPAAA) that is found in primate monkey (XP_054298120.1), birds and reptiles show polyAla (MSLKQAAAAAQAA in Struthio camelus, see entry XP_068766831.1), which seems to fulfil an analogous function [140] (Figure 5a). Both repeat types are known for their pathogenic expansion in patients [141,142]. PolyAla and polyGly repeats are found at each side of the fragment. In the center, a polySer repeat without or with interruptions has appeared with increasing size since monotremata. Around the first 100 amino acid residues of the human sequence are predicted to have an intrinsically disordered structure. The fact that this fragment extends in length since yeast / insects together with its poor conservation of individual amino acid residues may suggest that several repeats function as spacers between domains that interact, e.g., during autoregulation.

The N-terminal fragment before the LSm domain in ATXN2L, in comparison, has only doubled in length since Danio rerio, and contains only short repeats without much variation. It starts since Xenopus with the conserved sequence M(L/S)(K/M)(P/Q/K)(Q/P)(P/Q). About the first fifty amino acids of human ATXN2L are predicted to show an intrinsically disordered structure. The total length of IDR1 sequences upstream from the LSm domain reaches 250 aa residues in human ATXN2L.

In contrast to previous assertions that long polyQ domains in ATXN2 are exclusively found in primates, this datamining effort observed ATXN2 orthologs with a Q>5 domain also in flying lemur, pika, rodents such as hamster / beaver / kangaroo rat, bats, carnivores, horses, unpaired and paired ungulates, armadillo, sloth, aardvark, duck-billed platypus, alligator, and leatherback sea turtle.

In human ATXN2 (UniProt Q99700), the InterPro sequence analysis shows the polyQ repeat at residues 166-188 to have a high alpha-fold confidence, and the aa residues 156-202 (YGPLTMSLKPQQQQQQQQQQQQQQQQQQQQQQQPPPAAANVRKPGGS), especially the residues downstream from the polyQ repeat, to display significant homology with the beta-sandwich domain in Sec23/24. The Sec23/24 heterodimer regulates the trafficking of membrane proteins, phospholipids, sphingolipids, and sterols via COPII-coated vesicles from the ER to the Golgi apparatus and to elongating cell protrusions [143]. This homology is not detected by the InterPro algorithm in other species. A fly dATX2 isoform indeed colocalizes with Sec23, but this was reported for isoform C, which starts from the LSm domain, similar to isoform A, but no Sec23 colocalization was found for isoform B, which starts 60 aa before the LSm domain begins [144]. While it remains unclear if Sec23 is relevant in this context, human ATXN2 indeed associates with the ER, disruption of Ataxin-2 in flies and nematodes disrupts ER dynamics, and the polyQ-expansion in human and mouse ATXN2, as well as the deletion of mouse ATXN2, leads to disturbed sterol and sphingolipid homeostasis [51,54,55,134,144,145,146,147]. To what degree N-terminal sequences of Ataxin-2 contribute to these effects remains to be elucidated.

The pathogenic expansion of the polyQ repeat within this translation start sequence to sizes ≥33Q is the monogenic cause of autosomal dominant SCA2, and at sizes 31<(Q)n<33 contributes in a polygenic manner to the risk of sporadic ALS. It is interesting to note here that despite this long-term neurotoxicity of >31Q stretches, long polyQ repeats appear among ATXN2 orthologs initially towards the C-terminus already in single-cell organisms like red algae (UniProt A0A2V3IVX2), amoebozoa (Q55DE7), dinoflagellates (A0A813ABV5, where an extremely long 73Q repeat is interrupted at three positions by single Met residues; or A0AA36IYW2, where a 41Q repeat is interrupted at 11 positions by single or double Ala residues), as well as insects like Hymenoptera (A0A158NW33, A0A836FT30, F4X2F6, A0A834JVP6, A0A834JR87), so they seem to play a relevant role for the functions of Ataxin-2, but are not documented for ATXN2L orthologs so far. Smaller polyQ repeat sizes 5<(Q)n≤31 are frequent among Ataxin-2 homologs towards the C-terminus (e.g., in chlorophytes such as A0A835VQM0, A0A9W6BKV0, nematodes such as MCP9264076.1, or insects such as Q8SWR8).

After this repeat-rich fragment but before the LSm domain, a connecting sequence is encoded by exons 2 and 3A in ATXN2 (for ATXN2L encoded by exon 2 and the first nine residues from exon 3), which appears in isoforms initiated from the first or second translation start codon and should therefore be regulated with exon 1B. However, their respective exons are separated genomically from exon 1, clustering instead with the subsequent exons that encode the LSm domain, possibly suggesting that this sequence may be regulated as an LSm modifier.

2.8. The Role of Proline Flanking Residues and Interrupting Residues for the polyQ Repeat

The observation that the pathogenic N-terminal polyQ repeat of ATXN2-IDR1 is adjacent to, or framed by, polyPro repeats or Pro residues (Figure 5b) is also documented for the pathogenic polyQ expansions in Huntington’s disease [148]. Whereas PRMs are thought to require the presence of an Arg residue for their interactions with actin and endocytosis apparatus, the polyPro residues or Pro residues at the flanks or within polyQ repeats may play different roles. Pro residues are thought to modulate the poorly solvent polyQ stretch towards an α-helix secondary structure [149], and to mediate interactions e.g., with SH3 domains [150,151] and with membrane lipid bilayers [152]. The presence of polyPro helices in the intrinsically disordered N-terminal domain of human CPEB3 was shown to modulate solubility, acting as an amyloid breaker [153]. Importantly, proline- or glutamine-rich activation domains are a common feature of transcription factors [154], so in the case of cytoplasmic Ataxin-2 orthologs, it is plausible that these sequences influence the binding to RNA.

Prolines not only serve a role in flanking the polyQ repeat. They are also frequently used to interrupt the polyQ domain. The interspersion of prolines or polyPro sequences within the polyQ repeat is observed not only in primates and some rodentia, but also in lagomorpha, pholidota, artiodactyla, marsupiala, and squamata (Table S1c). Detailed studies of Ataxin-2 orthologs with amino acid residues interrupting its polyQ domain indicate that Pro >>> His >> Ala > Met = Arg = Ser = Gly are most used to interrupt polyQ stretches. The flanking residue on the left side is most frequently Pro >> His > Met = Lys = Gly = Val = Ala = Arg, while the flanking residue on the right side is more often Pro >>> Met = Arg = His > Ala = Asp = Glu = Gly. Overall, the amino acid residues that flank and interrupt the polyQ domain are non-random. Flanking (poly)Pro or interrupting Pro was the most frequent feature throughout the evolution of Ataxin-2 orthologs.

2.9. Ancient Start Codon Preceding LSm Domain as an Optional Third Start in Human ATXN2

The optional third translation start immediately upstream from LSm was in very frequent use during evolution until reptiles and is still documented in primates. The first of two methionine residues that may serve as a start codon in exon 3B was lost in muridae and cricetidae among rodentia, together with insectivora (VRMVHILTSVV… instead of MRMVHILTSVV…), but lagomorpha, dermoptera, and primates still have this first methionine (see Table S1e). Here, an isoform with abundant expression in spinocerebellar tissue can be produced that corresponds to Ataxin-2 from LSm until the C-terminus, without the appended N-terminus. This very ancient translation start codon produces an isoform with all core ATXN2 sequences (e.g., in XP_067991570.1). Deletion experiments in Drosophila melanogaster indicated that the LSm domain stimulates mRNA translation. At the same time, it antagonizes the assembly of ribonucleoprotein granules [83]. Via RNA binding, the LSm domain influences the assembly and stability of microtubules in Drosophila melanogaster neurites [63].

The crucial residues, with high conservation rate in the LSm domain of ATXN2 from mammals to insects, versus ATXN2L from mammals to amphibians, were compiled in Figure 2a/b. The LSm sequence variation in Ataxin-2 orthologs from nematodes to single-cell organisms is extensive, with insertions at various sites, so that multiple sequence alignments have difficulty defining core amino acids. It is worth noting again that the LSm sequence has limited usefulness in identifying Ataxin-2 orthologs, while the consensus LSmAD across evolution (and IDR2 / IDR3 within organism phyla), are much more powerful. Indeed, BlastP searches with the latter consensus sequences among fungi identify more than half of Ataxin-2 orthologs that have no recognized LSm domain, suggesting that the algorithms employed by InterPro and Pfam to identify LSm-folds need to be optimized. It is unlikely that the LSm-fold acts as a commodity rather than the necessity for Ataxin-2 functions, given that interference with this LSm domain is embryonically lethal.

The LSm domain, as an ancient RNA-binding motif with oligo(U)-specificity [155,156], is known to derive from Hfq protein in Bacteria / Archaea, where it assembles in homoheptameric rings and performs RNA chaperone function [157]. Hfq-binding RNAs were experimentally identified in Salmonella typhimurium bacteria, share a similar structure composed of three stem-loops, are enriched in the interacting sequence motif 5′-AAYAAYAA-3′, and show Hfq-association at AU-rich single-strand sequences adjacent to the loops, or the polyU end of sRNAs [158,159,160,161,162,163,164,165,166,167,168]. Hfq interacts with regulatory short RNAs, modulates RNA decay, acts in ribosome biogenesis, and influences mRNA translation, showing localizations in the cytoplasm, in the nucleoid, and at the bacterial cell membrane [169,170,171,172]. Importantly, the loss of Hfq causes increased membrane permeability, more sensitivity to oxidative stress, and activates envelope stress responses via the sigma transcription factor RpoS for envelope stress [169,173,174,175,176,177,178], leading to the activation of outer membrane proteins, enzymes, and transporters, which are involved in amino acid uptake and biosynthesis, sugar uptake and metabolism, and cell energetics [169,177,179].

Hfq was adapted in eukaryotic nuclei as Sm modules in the heteroheptameric ring of splice proteins, and in eukaryotic cytoplasm in more than a dozen proteins implicated in RNA processing or decay, translation control, the telomerase RNA biogenesis, cell signaling, and gametogenesis [180,181,182,183,184,185,186,187,188,189,190,191]. Outside of ring structures, the LSm domain exists e.g., in LSM11 as a factor associated with the methylosome [192] or LSM12 as a possible methyltransferase and/or NAADP receptor [189,193]. Thus, the LSm-fold sequence consensus is not specific for Ataxin-2 orthologs. However, the LSm domain still associates preferentially with AU-rich sequences, usually in the 3’ untranslated region of mRNAs [194].

2.10. RNA Processing Factors May Bind to the LSm-LSmAD Region, or Be Added to N/C-Termini of Ataxin-2 Orthologs, in Dependence on the Biochemical Needs of Different Ecological Niches

The region including the LSm and the LSmAD sequence contains around 250 residues, and in mammalian ATXN2, it was observed to mediate binding to the RNA helicase DDX6 [64,195], and to repress the illegitimate RAN translation of the CAG-repeat [196]. It is therefore interesting that Coptotermes formosanus termites show a N-terminal stable addition of a DEAD box RNA helicase domain (DDX52 homologous) to their Ataxin-2 ortholog (UniProt entry A0A6L2P9Q7), Trichogramma brassicae wasps have a C-terminal addition of a DEAD box helicase (A0A6H5ISX2), and Coemansia sp. fungi have a C-terminal addition of an RNA recognition motif (A0A9W7Z4B9) (see Table S3). These observations support the concept that eukaryotic organisms have optimized multi-domain proteins in order to concentrate all cofactors and substrates of a pathway in one place, thus maximizing throughput efficiency.

In analogy, further RNA processing functions that complement Ataxin-2 roles may have been integrated into its multi-domain structure, to ensure the maximal efficiency needed for different environmental stresses. Observations of various domains added at the N- or C-terminus of Ataxin-2 orthologs were therefore compiled in Table S3.

A C-terminal addition of a sequence with homology to HhH-GPD domains, like NTH1, occurred in Rhodotorula sp. CCFEE 5036 fungus (A0A4U0VMY8). HhH-GPD, like NTH1, recognizes and repairs 8-oxoguanine and other oxidative damage in DNA [197,198].

N-terminal addition of a tetratricopeptide domain plus a SAM-dependent 23S rRNA methyltransferase domain in Paramicrosporidium saccamoebae fungus (A0A2H9TMQ1), only a N-terminal 23S rRNA methyltransferase domain in Actinomortierella ambigua fungus (KAF9164688.1), or C-terminal addition of a 28S rRNA (cytosine-C5)-methyltransferase domain in Pseudozyma flocculosa fungus (A0A5C3F1S9), and a C-terminal SAM-dependent RNA (cytosine-C5)-methyltransferase domain in Wallemia ichthyophaga fungus (A0A4T0IZQ6) to the Ataxin-2 sequence provide further support for the relevance of this pathway. Tetratricopeptide repeat contacts bind RNA and assemble the splice apparatus [199], while eukaryotic 5-methylcytosine (m⁵C) RNA methyltransferases are crucial for responses to oxidative stress [200,201].

Similarly, the N-terminal addition of Ribosomal protein L9 N-terminal and C-terminal domains to the Ataxin-2 sequence in Coemansia interrupta fungus (A0A9W8LDP4), Coemansia spiralis fungus (A0A9W8GBK0), Coemansia biformis fungus (A0A9W8CWC1), and Coemansia pectinata fungus (A0A9W8H1X6) underlines this concept.

A ribonuclease inhibitor LRR repeat was added to the Ataxin-2 ortholog of Rhizoctonia solani fungi (A0A074SQD8) and of Ceratobasidium sp. fungi (A0A8H8MCZ2). The ribonuclease inhibitor serves to constrain the cytotoxicity of RNase A [202].

Regulation of the RNA polymerase was achieved by the addition of a CBF-B/NF-YA transcription factor domain in Rhododendron giersonianum shrub (A0AAV6I6Y5), or by the addition of an ACID domain of the MED25 complex in Certonardoa semiregularis sea stars (UPI001FC18B4E) to their Ataxin-2 orthologs. The MED25 complex regulates RNA polymerase via transcription modulators such as MYC [203].

The chimeric C-terminal addition of a RIO2 kinase domain in the bony fish Anabarilius grahami to the Ataxin-2 ortholog (ROL51429.1) points at rRNA processing, given that yeast RIO2 is required for the final endonucleolytic cleavage of 20S pre-rRNA at site D in the cytoplasm, converting it into the mature 18S rRNA [204]. The human ortholog RIOK2 also modulates neurotoxicity in mitochondrial complex I deficiency. In addition, it modulates ageing via telomere shortening [205,206]. It is important to note here that the RIOK2 gene is localized next to the ATXN2 gene since bony fish.

As further evidence that Ataxin-2 acts in RNA surveillance pathways, an N-terminal addition of a BRAP2 signature with an RNA-binding RRM motif, followed by a C3H2C3-type RING-H2 finger and a UBP-type zinc finger, occurred in the Ataxin-2 ortholog from Pituophis catenifer snake (XP_070807217.1). BRAP2 regulates cell survival via the antioxidant transcription factor NRF2 (SKN-1 in nematodes) and via the DNA repair factor BRCA1 [207,208,209]. Again it is noteworthy that the BRAP gene is the next neighbor of the ATXN2 gene upstream from its translation start on chromosome 12q in the human genome.

Only a HECT-type E3 ubiquitin transferase domain was added to the N-terminus of Ataxin-2 in Furculomyces boomerangus fungus (A0A2T9YUP5), implicating this ortholog in protein degradation pathways.

Despite current concepts that Ataxin-2 family members impact RNA processing mostly for mRNA transcript turnover with a poly(A)-tail via the PAM2 motif, the chimeric extra domains added to Ataxin-2 are clearly enriched for rRNA processing functions like methylation, rather than for mRNA decay versus repair management. Thus, this protein family appears to contribute to adaptations of the translation machinery after oxidative stress, with the LSm domain providing direct binding to RNA, while the variation in LSm sequences between orthologs and paralogs may relate to differences in the target sequence.

However, the structured Ataxin-2 sequences from LSm to LSmAD are not only important for RNA processing and quality control. Observations in Drosophila melanogaster show that Ataxin-2, but not its intrinsically disordered sequences, is crucial for the pathfinding of neurites in vivo [210], and, interestingly, these roles of Ataxin-2 were also fine-tuned by chimeric extra domains in some species.

2.11. LSmAD Became the Hallmark of Ataxin-2 in Rhodophytes and Protists

The LSmAD sequence of ATXN2 in Homo sapiens is encoded by exons 7 and 8, with genomic clustering suggesting that exons 6 and 9 are co-regulated with LSmAD.

The LSmAD sequence is unique and is the hallmark for the Ataxin-2 protein family, in contrast to the LSm domain that also exists in the LSM1-16 protein family for each species, and in contrast to the PAM2 motif that also occurs in PAIP1/2, LARP4, eRF3/GSPT1/2, TTC3, USP10, PAN3, GW182, Tob1/2 and in other factors across the diverse species [211,212]. Since the presence of an LSmAD sequence defines the Ataxin-2 family, it is very unfortunate that there are no insights into the LSmAD functions or into the molecular interaction partners of LSmAD. Regarding the interaction of LSm with target mRNAs, the LSmAD sequence has an enhancing role in Drosophila melanogaster brains [213], and indeed it was shown to modulate LSm conformation [77]. Interestingly, it was also reported that LSmAD mediates the localization of ATXN2 to the Golgi apparatus [50]. Luckily, LSmAD is highly conserved in sequence and length across evolution, so it is uniquely helpful to identify orthologs of Ataxin-2. Using BlastP to search for homologs of yeast PBP1-LSmAD (FGVKSTFDEHLYTTKINKDDPNYSKRLQEAERIAKEESQGTSGNIHIAE-DRGIIIDDSGLDEEDLYSGVDRR) among single-cell organisms, nothing significant was detected in bacteria and archaea. The LSmAD sequence was also not detected in ciliate / flagellate parasites except Trypanosoma and Plasmodium (both adapted to oxidative stress environments like blood cells), and already in rhodophytes and chlorophytes. However, among eukaryotic protists, plants, and animals, many orthologs were found with high significance (BlastP score <1e-10) and compiled in Table S1f. Thus, the LSmAD sequence appeared only in species with multi-domain proteins involved in stress responses that evolved to counteract oxidative damage due to the endosymbiosis of mitochondria and chloroplasts, and due to exposure to sunlight.

Regarding crucial amino acids within this domain, it was reported that LSmAD contains a clathrin-mediated trans-Golgi signal (YDS, amino acid 414-416 in human ATXN2) and an endoplasmic reticulum exit signal (ERD, amino acid 426-428 in human ATXN2) [49]. However, the experimental evidence for their functional role was never unequivocal, with both not being well conserved in ATXN2, and ERD not being well conserved in ATXN2L either, despite the LSmAD sequence showing astonishing conservation since protists (Table S1f).

In view of this strong homology among LSmAD sequences throughout evolution (see Table S1f), we used them in extensive BlastP searches to define Ataxin-2 orthologs with unusual length (>1000 residues) that contain further protein domains of known function. The findings are reported in the last paragraphs of the Results text. It is relevant to know that many LSmAD-containing factors among chlorophytes were named CTC-interacting domain (CID) proteins, which refers to the presence of a PAM2 motif and its interaction with PAB1-binding protein (the MLLE domain of poly(A)-binding proteins was previously known as CTC domain) [214].

2.12. Sequence after LSmAD, Including the PAM2 Motif, Has a Polyampholytic Intrinsically Disordered Structure and Is Modified by Alternatively Spliced Exons Ante-10 and 10 in Human ATXN2

Downstream from the LSmAD border, the Ataxin-2 orthologous proteins are predicted to be mostly intrinsically disordered, with short interruptions at irregular positions. The total length of Ataxin-2 orthologs in protozoa and fungi varies between 650 and 1200 amino acids, with great variability and only a few consistent trends from algae to mammals. Practically all of the length variability occurs among the disordered sequences, much more towards the C-terminus in IDR2 before PAM2 (Table S1g), and IDR3 after PAM2, than at the N-terminus in IDR1. Thus, it is clear that the length and properties of these intrinsically disordered sequences are carefully controlled. The IDR2 shows clearly consensus patterns that differ between vertebrates, insects, fungi, and algae. Size variation and occurrence of primitive repeats in IDR2 are considerable.

RNA surveillance occurs in stress granules, and fat storage occurs in lipid droplets, both of which exist in liquid-liquid phase separation (LLPS). Thus, the quite variable sequences with polyampholytic intrinsically disordered structure [215] immediately from LSmAD to the C-terminus (extending up to 1200 amino acids in orthologs, like XP_061722979.1 in Cydia pomonella, where an unusual positive polyelectrolyte repeat follows the LSmAD sequence, or KAJ8263487.1 in Conger conger, or KAJ1504482.1 in Coelomomyces lativittatus), may play a role in localizing Ataxin-2 in the vicinity of these cytoplasmic compartments.

The fact that the human ATXN2 gene has the alternatively spliced exons ante-10 and 10, at the end of the LSmAD cluster of exons, represents more evidence that this ATXN2 region has importance and is regulated in dependence on functional context, like stress.

2.13. PAM2 Motif in Many Protists and Insects Lies Close to Long polyQ Stretches

PAM2, as a short linear motif, is immersed among sequences with intrinsically disordered structures (Table S1h), at about 3-4 fifths of the total protein length, in contrast to LSm and LSmAD, where InterPro pLDDT scores predict high alpha- (or beta-) fold confidence. Previous literature has named the intrinsically disordered sequence until PAM2 as IDR2, and the intrinsically disordered sequence after PAM2 as IDR3 [62]. PAM2 is crucial for the ability of ATXN2 to be incorporated into stress granules [62,216]. PAM2 sequences changed considerably during evolution, so the InterPro algorithms often do not detect them, particularly in lower species; therefore, such candidate PAM2 motifs were compiled in Table S1h. It was already reported for CID3/CID4 in land plants [214] that a part can be repeated in tandem (sekstLNPNAKEFk-LNPNAKSFtps), but among excavates, amoebozoa, and algae, there is only an imperfect core sequence with considerable variability for Ataxin-2 family members (pksksKLNPNAKpFsLNinAKeF, see Table S1h).

Within the IDRs on each side of the PAM2 motif, polyQ repeat sequences are frequent from protists until insects, reaching up to 30Q without interruption (downstream from PAM2, for A0A2V3IVX2 in the rhodophyte Gracilariopsis chorda, or downstream from LSmAD for Q55DE7 in the amoebozoon Dictyostelium discoideum), at the extreme size of 73Q with 3 Met interruptions (A0A813ABV5 in Symbiodinium necroappetens), or later with interruptions by Pro residues (F4X2F6 in Acromyrmex echinatior, A0A836FCM7 in Acromyrmex insinuator, A0A158NW33 in Atta cephalotes, A0A8J2HMF1 in Cotesia congregate, A0A834JR87 in Vespula vulgaris, A0AA40FF43 in Melipona bicolor). The 73Q domain in Symbiodinium necroappetens is translated from nucleotides (Met coding triplets in bold) CAG(CAA)₂(CAG)₂CAAATG(CAG)₂CAA(CAG)₂CAA(CAG)₂CAA(CAG)₂CAA(CAG)₄CAA(CAG)₂CAA(CAG)₆CAA(CAG)₃CAAATG(CAG)₂CAA(CAG)₂(CAA)₂(CAG)₂CAA(CAG)₂CAA(CAG)₄(CAA)₂(CAG)₃(CAA)₃(CAG)₃CAA(CAG)₂CAA(CAG)₃ATGCAG, while the 30Q domain in Gracilariopsis chorda red algae gets translated from nucleotides (CAA)₃CAG(CAA)₂(CAG)₂(CAA)₂(CAG)₂(CAA)₅(CAG)₆CAA(CAG)₂(CAA)₄, and the 29Q domain in slime mold Dictyostelium discoideum is translated from nucleotides (CAA)₂₉. An uninterrupted 45Q domain in Condylostylus longicornis is translated from (CAA)₂CAG(CAA)₇CAG(CAA)₈(CAG)₃(CAA)₄CAG(CAA)₅CAG(CAA)₃CAGCAA-(CAG)₃(CAA)₄, see NCBI entries XP_055375020.1 and XM_055519045.1. Thus, there seems to be neither a codon preference nor a likely repeat structure underlying polyQ domains.

Downstream from LSmAD, the intrinsically disordered region 2 (IDR2) of human ATXN2 starts in the sequences encoded by exon 8 and extends across PAM2 into exon 17. The PAM2 motif in human ATXN2/ATXN2L is encoded by exons 16/15, respectively, which are genomically clustered with one preceding and two subsequent exons, suggesting joint regulation and function in this area. The subsequent constitutive ATXN2 exons 22/23 still encode intrinsically disordered sequences (IDR3). Within the ATXN2 exons 16 and 17, several methionine residues exist as candidate homologs of the yeast C-terminal methionine-rich region where metabolic redox sensing occurs to control phase separation of Ataxin-2 [217,218,219], but none of them is conserved in ATXN2L or in the single-copy Ataxin-2 orthologs, although the two paralogs and these orthologs show similar condensation dynamics.

2.14. Sequence Beyond LSmAD Without an Intrinsically Disordered Structure

In the C-terminal half of Ataxin-2 orthologs, the intrinsically disordered sequences are usually interrupted at least once, in primates up to 130 amino acids beyond the PAM2 motif, for some 40-200 residues (around position 500-880 for the typical Ataxin-2 ortholog of 1000 amino acid length) (Table S1i). For human ATXN2, this interrupting ordered sequence is encoded by exons 18/19/20, for human ATXN2L by exons 17/18/19. Interestingly, for human ATXN2, the alternatively spliced exon 18B evolved inside the start of this cluster, while the alternatively spliced exon 21 evolved at the end of this cluster, both encoding ordered sequences. This regulation effort is compatible with the notion that the functional relevance of this ordered area has to be modulated upon changes of stress versus stimulus.

2.15. C-Terminus Alternatives

The C-terminal end of Ataxin-2 orthologs can contain only an intrinsically disordered sequence, or first an intrinsically disordered part and then an ordered terminal end (Table S1i). Much alternative splicing is documented for human ATXN2 at the C-terminus, with constitutive strongly expressed exons 22/23 encoding intrinsically disordered ATXN2L-homologous sequences where some isoforms stop (e.g. Q99700-4). Furthermore, the alternatively spliced exons 24A/B encode ordered ATXN2L-homologous sequences including their joint stop (Suppl. Material S5). In ATXN2 exon 24A, there is another methionine-rich sequence as a potential similarity with the yeast PBP1-C-terminal region that acts as a sensor of metabolic redox status to control the phase separation of Ataxin-2 orthologs [217,218,219], but in the human species, these residues are situated in a structured region, and only one of these methionine residues is conserved in human ATXN2L. The translation products of the alternatively spliced exon 23B (encoding a sequence known since primates) and alternatively spliced exon 25 (encoding a Q(P/A/T)(H)_1-3Q(P/Q)(P/Q)L sequence before an alternative stop, followed by a long non-coding 3’ untranslated region) are predicted to have ordered structure as well, mediating the modification of the C-terminus in periods of growth versus stress.

This C-terminal sequence, encoded by exons 22-24A in human ATXN2, shows the maximum expression among all regions of Ataxin-2 according to GTEx RNAseq quantifications in nervous tissue, as mentioned above. It shows clearly conserved residues from humans until early chordates (Table S1i). Regarding its function, it was shown that the C-terminal IDR enhances Ataxin-2 target mRNA interaction, influences the long-term habituation, ribonucleoprotein (RNP) granule formation, and neurodegeneration in Drosophila melanogaster retina [210,213,220]. Therefore, the maximal expression levels of the C-terminal exons may serve to encode Ataxin-2 isoforms that are crucial for RNP granule formation and motility during neurite transport, while the lower production of longer isoforms, including LSm-LSmAD, might serve the transition from RNP transport to active translation in stimulated synapses [221]. It may also be relevant that this C-terminus has the highest sequence homology (28.7%, p = 3.5e-7) in chordates with entry A0A8C4PYQ0, a regulator of trafficking from ER to Golgi with predicted localization to ER exit sites, thus being assumed to influence the production of phospholipids and sterols, together with protein biosynthesis and trafficking, as well as calcium homeostasis.

The definition of highly conserved residues in this C-terminal sequence among insects (Table S1i) was employed to search for related proteins among nematodes and cnidaria, finding high homology with entry A7S107 (30.9%, p = 3.8e-13), again a regulator of trafficking from ER to Golgi with predicted localization to ER exit sites that is recognized as member of the SEC31 family [222,223].

2.16. Ataxin-2 Forms Chimeric Proteins with Very-Long-Chain Fatty Acid Synthases in Various Algae, Suggesting an Evolutionary Link and a Potential Role in Wax Biosynthesis in Plants

Several other motifs of known function were integrated with the core features of Ataxin-2 orthologs in chimeric proteins of several protozoa, algae, and plant species (Table S4). Given that plants have no mobility to run from danger and damage, stress response pathways are more prominent and more streamlined in plants than in animals. These additional domains may represent meaningful efforts to optimize the efficiency of Ataxin-2 within its biochemical context, given that the analysis of mice with mutant ATXN2 also documented prominent dysregulations in these processes. The data suggest Ataxin-2, together with co-translated neighbor genes, have a role in membrane stress responses, with strong conservation across evolution from algae to mammals.

Firstly, always a very-long-chain 3-ketoacyl-CoA synthase domain, and variably a chalcone/stilbene synthase domain, and a 3-oxoacyl-[acyl-carrier-protein] synthase were added to Ataxin-2 in Chlorella sorokiniana (A0A2P6U097), Chlorella ohadii (A0AAD5DQN1), Auxenochlorella protothecoides (V5NG01), and Micractinium conductrix (A0A2P6V2R2). The very-long-chain 3-ketoacyl-CoA synthases (abbreviated as KCS), also known as very-long-chain fatty acid (VLCFA) condensing enzymes, play a crucial role together with very-long-chain fatty acid elongases in the adaptation of lipid membranes to abiotic stress, prominently by the biosynthesis of (epi)cuticular wax [224,225,226].

Secondly, additional presence of a BRX domain is documented in Lactuca saligna (UniProt entries A0AA36E8E6 and A0AA36E8L8). BRX maintains brassinosteroid biosynthesis above a critical threshold [227].

Thirdly, a lecithin:cholesterol/phospholipid:diacylglycerol acyltransferase (LACT/PDAT) domain is present at the N-terminus of the Ataxin-2 ortholog in Populus alba (A0A4U5R0B4). Plant PDAT enzymes are involved in seed lipid droplet reserves by triacylglycerol formation via an acyl-CoA-independent pathway, and in membrane lipid remodeling [228,229]. Particularly in unicellular microalgae under stress conditions, the accumulation of triacylglycerol is accompanied by gradual chloroplast degradation with re-organization of membrane structures and remodeling of the lipidome [230].

Fourth, a formate/glycerate dehydrogenase catalytic domain with homology to the Escherichia coli SerA protein (crucial for serine biosynthesis) is added at the N-terminus of Ataxin-2 in Symbiodinium necroappetens (A0A813ABV5). L-serine is a precursor to sphingolipids and phospholipids [231], and formate/glycerate dehydrogenases have a role in the stress responses of plants (Symbiodinium carries a chloroplast) [232,233].

2.17. Ataxin-2 Is Chimeric with Oxysterol-Binding Proteins in Various Hexapods and with Myelin Biosynthesis Factors in Animals

This evidence from primitive plants is in good agreement with the integration of Ataxin-2 sequences into chimeric proteins among animals (Table S5).

In the arthropod-hexapod class of collembola, the Ataxin-2 domains are joined to oxysterol-binding protein sequences in Folsomia candida (A0A226E5R7), Archesella cincta (A0A1D2NII5), and Allacma fusca (A0A8J2NQ72, A0A8J2J4F7, A0A8J2NIC3). The role of sterols for plasma membrane dynamics is well known; oxysterols have an additional signaling function [234,235].

Furthermore, the addition of an N-terminal apolipophorin-III precursor domain to Ataxin-2 occurred in the wax-producing honey bee Apis cerana (A0A2A3ESF8) and in the ant Temnothorax longispinosus (A0A4S2KN76). In insects, apolipophorin-III assists in loading diacylglycerol, generated from triacylglycerol stores in fat bodies via the adipokinetic hormone, onto hemolymph lipophorin, and it also modifies the stress responses to bacteria [236,237].

A C-terminal addition of a Ser/Thr hydratase domain to Ataxin-2 is documented in Paralvinella palmiformis worms (A0AAD9K567). Such L-serine ammonia-lyases are central to the production of acetyl-CoA and lipids in other worms [238] and mediate stress responses against acid already in bacteria [239].

In Hermetia illucens flies (A0A7R8YXR3), a C-terminal C2HC-type zinc-finger transcription factor with homology to myelin transcription factor-1 (MYT1) is appended onto the Ataxin-2 ortholog. Here, it is important to know that animal myelin, sphingolipids, ceramides, and tear film meibum, similar to plant wax (in the Latin language, “cera” means wax), are produced from very-long-chain fatty acids [240]. MYT1 regulates, for example, the expression of the myelin protein PLP1 [241,242].

The Cloeon dipterum mayfly Ataxin-2 ortholog (A0A8S1C5C0, A0A8S1C908, A0A8S1C9E1) shows an N-terminal addition of a Ser/Thr protein kinase domain with homology to CAMKK. Human CAMKK controls adiposity and cholesterol homeostasis [243,244,245].

Among bony fishes, Clarias magur Ataxin-2 (KAF5904804.1) shows an N-terminal RALDH1 retinol dehydrogenase domain combined with a MAPKAP5 domain. The human RALDH1 ortholog ALDH1A1 protects against lipid peroxidation byproducts such as malonaldehyde, while MAPKAPK5 (also known as PRAK or MK5) controls cytosolic phospholipase A2 (cPLA2) and mTORC1 among other substrates, protecting against oxidative stress in Parkinson’s disease [246,247,248,249]. With special relevance for SCA2, RALDH enzymes govern the metabolism of retinoic acid, its signaling impact, and therefore hindbrain differentiation [112]. Human ALDH2, encoding a major biosynthesis enzyme for retinoic acid in liver mitochondria [250], and MAPKAPK5 are neighbor genes of ATXN2 on chromosome 12q24.

An Ataxin-2 ortholog in the naked mole rat Heterocephalus glaber (G5AKU3) shows C-terminal addition of a sesquipedalian family pleckstrin homology domain. Members of this family, such as yeast Boi1/Boi2 and human FAPP1, are required for the fusion of secretory vesicles with the plasma membrane at sites of polarised growth, acting by modulation of phosphatidylinositol 4-phosphate (PtdIns4P) levels [251,252]. Again, a sesquipedalian family member is a neighbor gene of ATXN2.

Overall, Ataxin-2 orthologs were combined with other sensors and effectors of membrane stress responses into longer multi-domain proteins, both in plants and animals.

2.18. Ataxin-2 Is Chimeric with an AMP Kinase Subunit and with a SCYL1 Ortholog in Some Fungi, Where Its Mutations May Impact Metabolic Reserves and Lipid Homeostasis

In fungi, domain additions to Ataxin-2 orthologs include (Table S6):

-

An N-terminal KA1/Ssp2_C domain in Neophaeococcomyces mojaviensis (NCBI Genbank entry KAJ9652119.1). This C-terminal domain in the serine/threonine protein kinase AMPK catalytic (alpha) subunit Ssp2 acts as a sensor to regulate cellular responses to a variety of nutritional and environmental stresses [253].

-

A longin-like domain exists at the C-terminus in Saitoella complicata fungus (UniProt entry A0A0E9NFC8). These sequences are involved in membrane dynamics regulation, occurring e.g., in GDP/GTP exchange factors for Rab7 / Rab32 / Rab38, such as the MON1/CCZ1-MC1 complex or the HPS1/HPS4-BLOC3 complex [254].

-

An ARM-like fold together with a CIK-related domain is present at the C-terminus in Aspergillus felis fungus (A0A8H6QL38). This combination of armadillo-repeats in CIK- (Catalytically Inactive Kinase) related proteins is preserved in human SCYL1-3, which are involved in the regulation of traffic between the Golgi apparatus and the ER, or play a role between the trans-Golgi network and the endosomal system, but also contribute to tRNA export from the nucleus [255,256]. This observation seems relevant since SCYL1 mutations cause progressive spinocerebellar neurodegeneration in mouse and human (SCAR21 type), and motor neuron disease with mislocalization of TDP-43 in mouse, similar to ALS and to ATXN2 mutation effects in humans [257,258,259,260,261,262]. SCYL1 mediates MTORC1 signals [263].

-

An N-terminal addition of a phospholipid transporter P-type ATPase occurred in Physocladia obscura fungus (A0AAD5T663 and A0AAD5XG15).

-

A C-terminal addition of a SEC6 domain is documented in Candida oxycetoniae fungus (A0AAI9WVV1) and in Candidozyma haemuli fungus (A0A2V1AQ46). The exocyst component SEC6 interacts with oxysterol-binding protein to regulate polarized vesicular membrane transport and growth in a sterol-dependent manner [264].

-

Further, an N-terminal cytochrome c oxidase, subunit Va/VI domain exists in Racocetra fulgida (A0A9N9AHJ9). This observation suggests that Ataxin-2 can act on the intra-mitochondrial pathway of oxidative stress generation directly.

-

A Calcineurin-like phosphoesterase domain was added in Diversispora eburnea (A0A9N8VJW7). Such a domain occurs, e.g. in nucleotidases and sphingomyelin phosphodiesterases [265].

-

A Cysteine desulfurase was appended to the N-terminus, and at the same time, a cell cycle-regulated microtubule-associated protein (TPX2) domain was added at the C-terminus in Bifiguratus adelaidae (A0A261XY39). The mitochondrial cysteine desulfurase NifS / IscS / NFS1 acts in the core iron-sulfur cluster (ISC) assembly complex, where perturbations lead to the human disorder Friedreich’s ataxia [266,267]. Cellular iron-sulfur (Fe/S) proteins are involved in lipid synthesis and steroid metabolism [268]. The TPX2 protein is an assembly factor for spindles, downstream from RanGTP [269].

Thus, Ataxin-2 orthologs (in diverse species from protists and algae to higher plants, fungi, and animals) are integrated into longer chimeric proteins that enable streamlined adaptation of lipids in membranes and fat bodies upon stress exposure, together with the production of (epi)cuticular wax or myelin from very-long-chain fatty acids, and the adaptation of chloroplast / mitochondrial homeostasis. This may suggest that mutations in Ataxin-2 orthologs play a role not only in RNA quality control but also in lipid adjustments after cell damage. Since LSmAD sequences are always adjacent to LSm domains and seem to function jointly with them, it is worth considering if the sensor of oxidative stress could be the LSm via binding to misfolded or oxidized RNA, while the Ataxin-2-specific effector of stress control could be LSmAD via modifications of lipid homeostasis.

3. Discussion

Current knowledge of the Ataxin-2 protein family is focused on the full-length isoforms. In particular, the roles of the LSm domain and the PAM2 motif and their impact on RNA processing have been investigated, although they are not unique to Ataxin-2. In contrast, very little knowledge exists about Ataxin-2 small isoforms and about the LSmAD component, its interactors, together with its functions. This focus has persisted, although small isoforms are more strongly expressed than the full-length isoform, and although LSmAD exists only in the context of Ataxin-2, constituting its defining feature. The fact that Ataxin-2 family members frequently show chimerisms with added domains at either terminus of the protein was previously unknown.

Here, this manuscript represents a datamining effort to analyze the evolution of Ataxin-2 structural components, their expression regulation, and the functional impact of their mutations. For a systematic evaluation, every effort was made to identify a maximum number of orthologs of Ataxin-2 in the various kingdoms of life, and to compile them in Table S1a. Overall, we hope to derive functional insights from lower organisms that can be applied in patients, where Ataxin-2 can be exploited as a crucial switch to mitigate the progression of neurodegenerative diseases [21].

Altogether, the data are compatible with a novel scenario where mutations in the Ataxin-2 protein orthologs since algae affect not so much mRNA, but mainly rRNA serving the biosynthesis of secreted and membrane proteins, and strongly impact membrane resilience in periods of cell stress (main datamining and experimental evidence is summarized in Figure 6). We speculate that strong expression of small isoforms of Ataxin-2 containing C-terminal IDR2 and IDR3 may alleviate the deleterious consequences of mutations in the N-terminal IDR1, so that human exon 1B seems the preferable target for knockdown therapies to prevent neurodegenerative diseases.

3.1. Ataxin-2 Chimerisms, Probably Due to Transcriptional Readthrough, Include Domains Enriched in Functions for rRNA Processing and Membrane Resilience

As a completely novel finding, large chimeric forms of Ataxin-2 with extra domains added to the N- or C-terminus were found to occur in approximately 5% of lower species, so this observation will be discussed first, in great detail. The lists of such domains added to Ataxin-2 in diverse species contain an enrichment of rRNA processing enzymes. If the chimerisms do not represent random errors of organismal protein synthesis or technical artifacts, then Ataxin-2 might contribute to defending and adapting the translation apparatus rRNAs in periods of cell stress, and to protect all vulnerable mRNAs with AU-rich elements and rapid turnover from oxidative damage, rather than acting on specific mRNAs. LSm and PAM2 perhaps make contact with RNA in an effort to sense oxidative damage, and only then act as modifiers of translation and repair. Under such oxidative stress, lipid peroxidation and membrane damage are bound to occur in parallel, and the associated LSmAD and IDR sequences may become crucial to protect the cell. A crucial question regarding LSmAD as a characteristic Ataxin-2 feature is whether it functions mainly in an auxiliary manner to modify the RNA-binding of the adjacent Lsm domain, RNA stability, and translational control – or if it modulates membrane interactions and composition upon recognition of oxidized RNA by the LSm domain.

This novel concept takes into account the ability of eukaryotic organisms to create multi-domain proteins in order to maximize the efficiency within a metabolic pathway. Given that the order of genes within chromosomes and within genomes is dynamically reshuffled across evolution, there is a continuous sorting process if neighbor genes are well matched or should be assembled differently. It was recently shown that transcriptional readthrough across neighbor genes is frequent under stress conditions for GC-rich genes (such as Ataxin-2) [270,271], in particular upon calcium release from ER stores [272]. Also, healthy cells show this feature in genomic regions with high transcriptional activity [273,274]. Although most genes in a genome appear to have no functional connection to their neighbors, some genes indeed are clustered with functional interactors within the same locus (HLA genes, as a famous example). It is known in humans that the chromosomal locus of Ataxin-2 is part of an exceptionally large linkage disequilibrium block across 1 megabase [117,275], so there is a selective advantage in these neighbor genes to stay together. It is unclear according to what principles the positioning of new genes or duplicate gene-copies into a meaningful genomic context occurs, and how chromosomal rearrangements are adapted over time. Given that the genomic surrounding of Ataxin-2 has changed over time, we cannot know which among the observed chimeric proteins has a physiological added value. However, it is conceivable from the data above that mutations in Ataxin-2 during cellular stress may also affect the transcription of neighbor genes and the translation of their domains.

Of course, some annotated proteins in databases are based on preliminary documentation, and some might be artifacts due to technical problems. The limitations of our study mainly center on our inability to distinguish artificial protein fragmentation and chimerism from physiological short isoforms and extra-long readthrough multi-domain proteins, respectively. In the following, examples are provided that support transcriptional slippage rather than ribosomal slippage [276] to underlie the observed chimerisms: In the 2577 residues of the UniProt entry A0AAF0YF69 from Vanrija pseudolonga fungus, in the 2585 residues of A0AA48I7S8 from Cutaneotrichosporon cavernicola fungus, and in the 2553 residues of A0AAD3TXG0 from Cutaneotrichosporon spelunceum fungus, Ataxin-2 orthologous sequences have an added cytidine deaminase domain together with immunoglobulin domains that are still enclosed within a chromosome 7 whole genome shotgun sequence, according to InterPro analysis. Such recombined fusion proteins could arise, e.g. during cDNA synthesis via erroneous annealing or may be the product of transcriptional readthrough, where intron retention due to cell stress is a known crucial feature [270,277].

Chimerisms of Ataxin-2 were frequently found among lower organisms, where cell stress is more frequent than in highly developed organisms. Interestingly, a chimerism is documented in the ATXN2 ortholog of the Pituophis catenifer snake (NCBI protein entry XP_070807217.1, see Table S3), where genomic information is already available. The N-terminal addition of an extra-domain with BRAP2 can indeed be explained as transcriptional readthrough, given that BRAP2 is the closest neighbor gene upstream from ATXN2 according to the UCSC genome browser from humans to Alligator reptiles and Xenopus amphibians, with the same direction of translation. Given that practically all genomes are unknown for the species with extra-domains added to Ataxin-2, it is probable that many or all the above chimerisms represent transcriptional readthrough events. Thus, the generation of readthrough transcripts and chimeric proteins appears to be a relevant feature of Ataxin-2, which gets activated in periods of cell stress.

Chimerism of ATXN2 among highly developed organisms, where the genome sequence is known for related species, is also documented in the Heterocephalus glaber naked mole rat (entry G5AKU3, see Table S5). A sesquipedalian family domain was added to its C-terminus. Again, this can be explained as transcriptional readthrough, given that the Sesquipedalian-1 protein is encoded by the PHETA1 gene, the immediate neighbor of the ATXN2 gene at its 3’ end, with the same transcriptional orientation, as a genomic position conserved until human.

Further examples include the C-terminal addition of a RIO2 kinase domain to the ATXN2 ortholog of the bony fish Anabarilius graham (UniProt ROL51429.1), and indeed the UCSC genome browser of the bony fish Danio rerio shows the RIOK2 gene to be localized at the 3’ end of the ATXN2 orthologous gene in the same transcriptional direction. A final example is the N-terminal addition of a RALDH1 retinol dehydrogenase domain and a MAPKAP5 domain to the ATXN2 ortholog in the bony fish Clarias magur (UniProt entry KAF5904804.1), and again the UCSC genome browser of the bony fish Latimeria chalumnae confirms the position of the ALDH1A1 and MAPKAPK5 genes at the 5’ end of the ATXN2 gene, but interestingly in the opposite transcriptional direction. Overall, these examples from species with known genome sequence indicate that the extra domains added onto Ataxin-2 are transcribed and translated from neighbor genes as a general rule.

Of course, it is important to be cautious and to ignore artifacts during datamining efforts, but further formal arguments and a host of experimental evidence led us to consider the above chimerisms as relevant, or at least worthwhile to be assessed further.

Firstly, technical artifacts should be distributed randomly across databases, but the Ataxin-2 orthologs with additional domains are frequent among low organisms but very rare among higher animals and plants.

Secondly, artifacts should implicate all pathways at random in Ataxin-2 biology, but the additional domains are preferentially involved in stress responses, rRNA processing, the mobilization of fat/glycogen stores, and the modulation of lipid membranes.

Thirdly, although aberrant fusion proteins may occur in nature as pathological events that cause pathology, e.g., cancer, their conservation across multiple related species points to a selective advantage of this fusion event in that particular ecological niche.

Fourth, previous experimental characterization of Ataxin-2 mutant organisms has revealed only small selective rather than striking or broad changes of RNA repair or mRNA translation in normal tissues and growing cells (in summary of many transcriptome profiling efforts from humans to mice and flies, [55,56,194,221,278,279,280,281,282,283,284,285,286,287,288,289,290,291]. In contrast, strong anomalies of lipid (and carbohydrate) stores and membrane homeostasis were readily demonstrable upon Ataxin-2 mutation not only in lower organisms such as flies, but in a conserved manner until mammals. Thus, a host of previously reported experimental evidence supports the concept that Ataxin-2 mutations affect membrane composition.

Irrespective of whether the extra-domains reflect only transcriptional readthrough events that are relevant selectively for stressed cells, or are part of a context of physiological functions of Ataxin-2, the pathways surrounding such extra-functions are affected by Ataxin-2 mutations as a consequence of the existence of chimeric proteins. So it seems relevant in the following paragraph to provide a summary of current experimental evidence that implicates Ataxin-2 mutations in membrane stress.

3.2. Experimental Confirmation that Mutations in Ataxin-2 Alter Membrane Resilience

In particular, the absence of the Ataxin-2 ortholog PBP1 in Saccharomyces cerevisiae yeast caused downregulation of several mitochondrial metabolism enzymes, namely citrate synthase, alpha-ketoglutarate dehydrogenase, glyoxylate reductase, succinyl-CoA ligase, and ATP synthase component ATP1 [138]. PBP1 is integrated in the phosphorylation cascade between AMPK and MTORC1 [292], sensing metabolic redox status via reversible oxidation of multiple methionines that control its condensation state [217,218,219]. Given that these methionine switches are not conserved during evolution, it is conceivable that oxidation damage is also sensed via lipid peroxidation or via oxidative RNA damage.

-

In Caenorhabditis elegans, the Ataxin-2 ortholog regulates the fat content and size of cells, acting via MTORC1 signaling [293,294].

-

In Drosophila melanogaster flies, Ataxin-2 accumulates in fat bodies together with Sec23 as a marker of ER exit sites [144]. The depletion of fly Ataxin-2 mitigates glycogen metabolism problems similar to mTORC1 repression rapamycin treatment [295].

-

The absence of ATXN2 from Mus musculus tissue results in massive accumulation of lipid droplets and glycogen, together with reduced sterol degradation in cerebellum and elevated blood cholesterol [134]. Proteome profiling of these ATXN2-null mouse livers demonstrated significant deficits of enzymes in the fatty acid beta-oxidation and malonyl-CoA / methylmalonyl-CoA pathways [296]. Again, according to unbiased global proteome and metabolome profiling efforts, the ATXN2 polyQ expansion in mouse cerebellum has indeed its main impact on the very-long-chain fatty acid elongases such as ELOVL4 (residing at the ER in a multi-protein complex), the very-long-chain fatty acids VLCFA24-26 with precursors such as acetyl-CoA and N-acetylaspartate [130], and their derivative sphingolipids and ceramides [145]. This is accompanied by changes in inositol-tris-phosphate metabolism and calcium/calmodulin-dependent kinases [297]. Receptors for the inositol 1,4,5-trisphosphate lipids were also implicated in an SCA2 mouse model [298]. In the spinal cords from two different SCA2 mouse models with ATXN2 polyQ expansion, the suppression of enzymes for cholesterol biosynthesis from acetyl-CoA and squalene, together with cholesterol and oxysterol deficits, was the main finding [147,287,299]. The restoration of brain cholesterol turnover was reported to have therapeutic value in a SCA2 mouse model [300].

-

For Homo sapiens, the brains of SCA2 patients show a reduction of C22/24-sphingomyelin and cholesterol levels [145], and the myelinated white matter of the brain is deficient [24,301,302,303,304,305,306,307], already at presymptomatic disease stages [31,308,309]. The subcutaneous fat deposits, the cheek fat body, the visceral fat, and the body weight of SCA2 patients decrease progressively, as well as the levels of testosterone as a cholesterol-derivative, in peripheral tissues [310,311,312]. In an epidemiological genome-wide association study of a Japanese population, variants of ATXN2 were found to underlie the susceptibility for dyslipidemia [313].

3.3. What Is the Role of Ataxin-2 for the Unfolded Protein Response and Retinoic Acid Signaling?

In bacteria, the RNA chaperone Hfq, as an ancestor of the LSm proteins, responds to envelope stress via the sigma transcription factor for envelope stress, RpoS. This sigma factor is regulated either at the post-transcriptional level via Hfq or at the proteolysis level via CLPP/CLPX-mediated unfolded protein responses [178]. Membrane stress responses via sigma transcription factors were not conserved in eukaryotes. Instead, stress granules and the Unfolded Protein Response (UPR) with the Sigma-1 Receptor (S1R) at the mitochondria-associated ER membrane (MAM) in interaction with the inositol 1,4,5-trisphosphate receptor (IP3R) and Inositol-Requiring Endoribonuclease 1 (IRE1) mediate the appropriate responses for ER stress in mammals [314,315,316]. For the UPR of bacterial and mitochondrial membranes, CLPP is crucial [317,318,319,320,321,322,323], and mutations in CLPP result in progressive ataxia and leukodystrophy in patients and mice [319,324,325].

The elongation to synthesize very-long-chain fatty acids occurs at the ER, and is the prerequisite for the biosynthesis of (glyco-)sphingolipids, ceramides and myelin / wax, when membranes need to be strengthened [326,327]. To integrate proteins into plasma membranes / the myelin sheath / the (epi)cuticular wax, their biosynthesis under stress conditions may require internal ribosomal entry site (IRES)-dependent translation from appropriately spliced mRNAs, then post-translational modification like glycosylation in the ER lumen, trafficking via the Golgi apparatus along microtubules to their destiny in a polarized cell, and finally their turn-over by the ER associated degradation (ERAD). Ataxin-2 localizes to all these compartments and may have evolved to modulate all these processes during stress periods, including the recruitment of lipids from cytoplasmic fat droplets and carbohydrates from glycogen granules [134], for the modification of membrane composition and complexity.

Ataxin-2 impacts not only ER membranes but also mitochondria. It was shown to control translation preferentially for mitochondrial factors [135], to modulate PINK1 levels [70,328], interact with PARK2 and sequestrate it into insoluble aggregates [68,69]. Thus, the crucial cytoplasmic factors that govern mitochondrial membranes regarding stress signaling, damage repair, and autophagic elimination are within the network of Ataxin-2 functions.

Retinoic-acid-triggered cell differentiation of mammals involves ER stress UPR via XBP1 [329,330,331,332,333]. IRE1 selectively acts on XBP1 transcript, excising 26 nucleotides [334,335,336]. This generates an XBP1 splice variant, which encodes a transcriptional activator to promote UPR target gene expression [334,335,336]. Very recently, ATXN2 was also found to selectively target XBP1 mRNA, stabilizing it and suppressing its translation, until Ataxin-2 is ubiquitinated by FBXO42 in a SKP-A/CULLIN-1 complex and gets degraded [291]. XBP1 is not only the central transcription activator in ER stress pathways via UPR, with an important role for neurodegenerative disorders [337,338,339,340], but is also crucial in physiology for secretory cell signaling to regulate lipid and insulin metabolism, as well as autophagy [341,342,343]. IRE1 and XBP1 influence the expression of MYC/MYCN and receive feedback control from them, within the pathway where retinoic acid governs differentiation, e.g., of neuroblastoma cells [344,345,346,347,348,349], with Ataxin-2 levels previously shown to depend on MYCN within this pathway [350]. During severe ER stress, regulated IRE1-dependent decay (RIDD) of mRNAs (to lessen ER load by reducing translation and post-translational ER translocation and secretion of proteins) generates single-strand mRNA fragments that lack 5′-caps or 3′-poly(A) tails, thus activating retinoic inducible gene-I (RIG-I) protein and mitochondria-dependent innate immunity [351]. The LSM domain of Ataxin-2 might recognize such RNA fragments, and Ataxin-2 might contribute to RIDD activity.

In mammals, the mutations in both ATXN2 and CLPP cause ataxia phenotypes [1,21,324,325]. ATXN2 mutations have a profound effect on IP3R1 expression [297,352], and mutations in mammalian IP3R1 also cause a phenotype of ataxia [353]. The relevance of the IRE1-XBP1 pathway among ER stress responses is well studied in ALS, but was also reported for Ataxia-Telangiectasia and Friedreich’s Ataxia [354,355,356,357,358,359,360,361]. A crucial role of Ataxin-2 orthologs for vitamin A derivatives, cholesterol, and retinoid homeostasis (as modulators of membrane strength versus stress) was evident in lower organisms, and appears conserved until mammals, given that Atxn2-KO mice show altered retinoic acid signaling [352]. The receptors for retinoic acid and related compounds, RAR-RXR-RORalpha, are in close interaction [362,363,364,365,366]. RORalpha protects neurons against oxidative stress [367,368]. Abnormal retinoic acid signaling was reported in several forms of ataxia, namely SCA1 [369,370], SCA3 [371], SCA7 [372,373], FRDA [374], A-T [375], and the staggerer mouse mutant [376,377,378,379]. Several of these ataxias show disease course modification by Ataxin-2.

3.4. Are Chimerisms, Transcriptional Readthrough, and Neighbor Genes Relevant for Ataxin-2-like?

It remains unclear to what degree the roles and regulation of ATXN2, which remained in the ancient genomic context of Ataxin-2, are still conserved for ATXN2L after gene duplication into a different genomic context. Within its locus on human chromosome 16p, the same transcriptional orientation is shared among neighbor genes EIF3C – NPIPB9 – ATXN2L – SH2B1 – ATP2A1. Chimeric proteins containing ATXN2L with extra domains from its neighbor proteins were not observed in our study, although ATXN2L expression is also activated in human neuroblastoma cells during stress periods [128].

The observation in Figure 2 that ATXN2L has acquired a distinct sequence pattern that is clearly separate from each ATXN2 paralog throughout different species, from fish to human, is in good agreement with previous findings that the loss of one paralog does not trigger the compensatory induction of the other paralog [128]. Jointly, the data indicate separate roles in a cell.

Apart from their similar role for stress granules, ATXN2L deficiency in mice was recently shown to trigger multinucleated cells (perhaps due to deficits of nuclear membranes or their stability), selectively affecting nuclear envelope component SYNE2 at microtubule organizing centers (MTOC), the nuclear FMR1 interacting protein 2 (NUFIP1), and the nuclear alternative splicing pathway. It is also interesting to note that the nuclear homeobox transcription factor and pyramidal neuron marker CUX1 is downregulated in ATXN2L-null mouse brain, in view of the CUX2 gene being localized at the 3’ end of the ATXN2 gene in mouse and human, suggesting that a CUX expression enhancer acting in trans is part of the ATXN2L gene duplication segment and was affected by the mutation [59,128,129,380]. ATXN2L seems distributed more between the ER and the perinuclear space, while ATXN2 appears associated more with the ER, Golgi, endosomes, and plasma membrane. Therefore, we speculate that ATXN2L is specialized in the formation / maintenance / strengthening of nuclear membranes in eukaryotes, where microtubules are anchored, playing a less complex role that did not need to evolve much. In comparison, ATXN2 would be specialized for Golgi membranes, plasma membranes, and cell protrusions, a role that has become much more sophisticated for mammalian neurons / neurites / synapses than it was in protists.

3.4. Ataxin-2 Isoforms

Previous research mostly investigated the full-length isoforms of each species. Although several articles documented allelic splicing of Ataxin-2, and one article emphasized the existence of at least two start codons [124,126,127], any small bands in immunoblots got usually dismissed as apparent cross-reactivity. In particular, it was thought that the functions of Ataxin-2 depend on the joint presence of LSm, LsmAD, and PAM2 sequences. However, this datamining study identified many species where alternative isoforms either contain LSm and LSmAD or lack them. In addition, several database entries were encountered where the PAM2 motif was alternatively spliced out from the Ataxin-2 ortholog. Figure 4 documents six alternative isoform starts before the LSM sequence, another four starts before PAM2, and 3 additional starts after PAM2, but there is no functional understanding about the different contexts where each isoform is useful. It is clear that database entries may be contaminated by artifacts and fragments, but the vast majority of Ataxin-2 ortholog isoform entries start with a Met start codon, as expected for functional isoforms.

Together with the evidence that the C-terminal IDR3 sequences are most strongly expressed, and that multiple variations of C-terminal but not N-terminal short isoforms exist in spinocerebellar tissue, as seen in Figure S1, it is clear that the old concepts have to be revised and the short isoforms should be systematically investigated.

The smaller isoforms towards the C-terminus may be crucial for SCA2 and ALS patients. It is conceivable that the previous experiments targeting exon 1 of mouse Atxn2 [134,381] were partially compensated by increased generation of shorter isoforms, some of which would contain all core domains. Such compensation efforts would explain why mice with Atxn2-knockout or Atxn2-knockin targeting exon 1 have only mild phenotypes of obesity, insulin resistance and dyslipidemia, or age-associated neurodegeneration, respectively [5,52,53,55,69,70,81,124,130,131,134,145,147,282,284,286,296,297,352,381,382]. These observations clearly diverge from Atxn2l-knockout mice with disruption at LSm / LSmAD sequences showing mid-gestation embryonic lethality with brain cortex lamination defects [59,128,129], from the knockout of complete fly Ataxin-2 or its LSm or LSmAD domains in Drosophila melanogaster that result in lethality [210], and from the knockdown of nematode Ataxin-2 in C. elegans that also causes embryonic lethality and germline defects [60,278,383].

3.5. How Are Ataxin-2 LSm and LSmAD Domains Essential for Eukaryotic Life?

The datamining about Ataxin-2 evolution confirmed that it became relevant when eukaryotic cells had already engulfed mitochondria and plastids, apparently when eukaryotic synthesis of organellar membranes started, and when organelles were distributed across the cell geometry. Ataxin-2 is not yet documented in glaucophyte, but appears in rhodophyte and chlorophyte algae and in other primitive protists. Upon terrestrialization of plants and animals, gene duplications were conserved, suggesting that more stress responses required higher dosage. However, it is unclear whether higher gravitational forces or increased exposure to sun light, heat, and drought are responsible. The fact that flying birds (although not flying bats) lost the duplicated gene might argue in favor of the importance of mechanical stress. In consequence, the notion that Ataxin-2 also has a role for the resilience of cellular membranes and cytoskeleton would be strengthened, in addition to its role for rRNA and mRNA quality control.

The essential parts of Ataxin-2 are evidently the beta-sheet structured LSm and the alpha-helix structured LSmAD, and perhaps also the linear PAM2 sequences. On the one hand, targeted interference with these domains causes lethality already at the embryonic stage, while interference with the IDR sequences triggers mild phenotypes. On the other hand, only these three sequences and structures are strongly conserved from algae to humans, while IDRs have considerable variability.

3.6. How Should Future Investigations Be Re-Focused?

From the conclusions above, it is clear that samples from SCA2 and ALS13 patients, as well as their disease models and treatment interventions, should now be re-investigated. Urgent questions concern (i) the altered expression of neighbor genes due to the mutation or the therapeutic approach, (ii) in particular the altered expression and stability of the potentially active U7 snRNA copy at the ATXN2 gene locus (most U7 copies are encoded on chromosome 12p13), as well as the processing of replication-dependent histone mRNA, together with the dynamics of U-bodies, Cajal bodies and spliceosomes [384,385,386,387], (iii) the altered expression of individual exons together with the stability of shorter Ataxin-2 isoforms, (iv) the altered resilience and dynamics of membranes and cytoskeleton of cells, (v) the oxidative damage and altered methylation of rRNA, in these samples.

This should be complemented by the generation of new mice with targeted conditional knockout of the Atxn2 LSm domain and of the Atxn2 LSmAD sequence, to first assess if this intervention causes lethal effects, and secondly to create tools where tissue-specific interactots of LSm and LSmAD can be studied.

Importantly, the interactor molecules of the LSmAD domain, as the characteristic feature of Ataxin-2 family members, should be documented to elucidate their functions. Subsequently, point mutations and deletions should target the LSmAD domain for a precise mapping of such interactions.

Furthermore, the impact of Ataxin-2 and its mutations on the different pathways of the unfolded protein response and the mechanical resilience / composition of membranes should be explored systematically, regarding the endoplasmic reticulum, mitochondria, plasma membrane endo/exocytosis, and nuclear membrane trafficking.

Most importantly, there is a consequence of our datamining effort for the design of preventive therapies against SCA2 and ALS13. In view of the complexity of Ataxin-2 isoforms and chimerisms, it seems crucial to selectively target the poorly expressed polyQ-surrounding sequences encoded by exon 1B of human ATXN2 rather than the strongly expressed C-terminal sequences or the conserved domains where interference can trigger cell death. Selective targeting of these sequences before the LSm domain would target only a region of ATXN2 that is non-essential and has evolved only since bony fish (Figure 4). Exon 1B encodes polyQ (as CAG repeat with CAA interruptions) and polyP (via CCC or CCG tripletts), with both repeats jointly enhancing the GC content of the 5’ end of ATXN2 and thus increasing its mRNA stability but also its propensity for chimerisms in stress periods [270,388], while differentially modulating liquid-liquid phase-separation given that polyQ forms amyloid while polyP breaks it [153]. Such re-design of knockdown approaches to target exon 1B would avoid any impact on isoforms that are not full-length and would permit a partial compensation of functions by upregulated expression of smaller isoforms.

4. Materials and Methods

The publically available bioinformatics tools employed in this study comprise the BlastP algorithm at UniProt-UniParc and NCBI databases, with compilation of results in EXCEL as tables and in WORD as text files; additional homology searches via the STRING Heidelberg database; identification of additional orthologs by automated interrogation of these databases (harnessing the GenAI Claude Code tool to compile and share data via an iterative LSMAD-bootstrapping strategy, BlastP against NCBI nr database with taxonomy filters, ConservedDomain-Search validation with strict E-value thresholds, quality tier classification, and Python/BioPython implementation, followed by Suprabase data storage, and GitHub plus Docker software for data sharing); multi-protein sequence alignment via the CLUSTAL Omega algorithm; protein domain analyses by InterPro and Pfam algorithms as well as MitoProt and TPpred3 prediction tools of mitochondrial targeting sequences; structure analyses at ebi.ac.uk / rcsb.org databases and via PyMOL; expression analyses for mRNA isoforms at ENSEMBL and NCBI databases and for individual exons at GTExportal.org; genomic analyses of exon-intron structure via the UCSC genome browser; visualization of alternatively spliced isoforms with Maya software; biomedical publications from the PubMed database. Regarding the protein structure 3D modelling, the AlphaFold predicted structures of human ATXN2 (AF-Q99700-F1-v6), ATXN2L (AF-Q8WWM7-F1-v6), and G. chorda ortholog PBPIP4 (AF-A0A2V3IVX2-F1-v6) were used in structural domain alignments. Structured regions corresponding to LSm, LSmAD and polyQ domains were extracted from the main predictions. Alignment was performed on the extracted molecules using the built-in “align” function with default parameters in Open-Source PyMOL, v3.1.0 (Schrödinger, LLC). Commercially licensed graphics softwares to illustrate results included Powerpoint, Adobe Photoshop, and BioRender.

5. Conclusions

Overall, the functions of Ataxin-2 in mRNA processing have been the main focus of attention so far, but its evolutionary data and genomic context suggest that its mutations also impact rRNA stress and membrane stress. Analysis of alternatively spliced exons and documented protein isoforms suggests low expression of N-terminal sequences, which could be targeted in neuroprotective efforts by knockdown strategies and may be compensated by strong expression of C-terminal isoforms, thus minimizing adverse effects.