Partners in crime: bidirectional transcription in unstable microsatellite disease

Antisense regulatory issues

Recent evaluations of genome-wide transcription have demonstrated that antisense transcripts can be detected for many mammalian genes, including the wild-type alleles associated with microsatellite expansion diseases [Table 1 (14,17–20)]. In most cases, the major sense strand is protein coding and is generally present at a higher copy level than the corresponding antisense transcript(s) which is noncoding and only partially overlaps the sense transcriptional unit, often in the promoter and 3′ terminal regions. Antisense transcription has been suggested to regulate gene expression at multiple levels, including (i) transcriptional interference of the sense strand due to potential collisions between RNA polymerase II holoenzyme complexes transcribing opposite strands; (ii) modulation of gene expression by recruitment of various chromatin remodeling complexes which mediate histone modifications and DNA methylation; (iii) masking splice and polyadenylation sites of pre-mRNAs transcribed from sense strands leading to alternative splice, and poly(A), site selection; (iv) duplex formation between sense and antisense transcripts that induces RNA editing and modulates RNA interference, stability and translation (21–23). The expansion of microsatellite repeats could disrupt these regulatory steps by altering the relative levels of sense versus antisense transcripts or by affecting transcriptional start or termination sites.

In the following sections, we focus on three neurological disorders where bidirectional transcription has been implicated in disease pathogenesis and progression although antisense RNAs have been reported at other microsatellite expansion loci (2,13). First, we review progress on spinocerebellar ataxia type 8 (SCA8) where both a polyQ protein produced from the sense strand and CUG^exp antisense RNAs have been suggested to play significant roles in pathogenesis. In contrast to SCA8, fragile X syndrome (FRAXA) and fragile X-associated tremor/ataxia syndrome (FXTAS) are two diseases caused by variations in a CGG•CCG expansion at the FMR1/ASFMR1,FMR4 locus. While loss of FMR1 expression clearly underlies FRAXA, potential roles for the antisense ASFMR1 and FMR4 transcripts in FXTAS will be reviewed. Finally, we evaluate the curious molecular etiology of HD-like 2 (HDL2), where the current evidence indicates that this disease is associated with a CTG expansion in the JPH3 gene and CUG^exp RNA toxicity. Given the clinical similarity of HDL2 to HD and the overwhelming evidence for a primary role of polyQ toxicity in the latter disease, it is surprising that HDL2 CAG^exp antisense transcripts have not been detected in the brain.

Prior to beginning our discussion, it is important to note that any potential pathological impacts of antisense transcription likely depend on several parameters, including the genomic organization of the disease locus, spatial and temporal expression patterns of the affected genes, repeat sequence and repeat length together with location in the gene(s). In addition, the expression of neighboring genes may also be altered which could be particularly relevant to disease loci where gene density is high, such as the DMPK gene in DM1 (24–27).

SCA8: a toxic cocktail of polyQ protein and CUG^exp RNA?

For the majority of unstable microsatellite disease genes, it is unknown if antisense RNAs are transcribed through the expanded repeat regions (14,20). An exception is SCA8, a dominant SCA with reduced penetrance, where there is evidence for bidirectional transcription that results in the production of toxic noncoding RNAs from the antisense transcriptional unit (ATXN8OS) and a nearly pure polyQ protein, ataxin 8, from ATXN8 (18). Unlike other forms of spinocerebellar ataxia that contain CAG^exp in the coding region of the sense strand, SCA8 was initially thought to be an exclusively RNA-mediated disease, similar to DM1, because the only detectable RNAs were noncoding ATXN8OS transcripts containing a CUG^exp near the 3′ end (28). As mentioned previously, RNA toxicity in the neuromuscular disease DM1 is caused by transcription of a noncoding CTG^exp in the DMPK gene. Transcription results in CUG^exp RNAs that effectively trap the MBNL1 splicing factors and alter alternative exon use during RNA processing to favor the production of fetal isoforms (12,13). These CUG^exp RNA–MBNL1 complexes accumulate in nuclear RNA foci that may be the more toxic entity or simply a protective response (29).

The possibility that SCA8 is also caused by this type of CUG^exp toxicity mechanism is supported by several studies. Transgenic expression of either the wild type or an expanded human SCA8 gene in Drosophila causes progressive retinal neurodegeneration which is enhanced by muscleblind (the fly homologue of MBNL1) loss-of-function mutations (30). Moreover, CUG^exp foci are detectable in the cerebellar cortex (Purkinje cells, Bergmann glia, molecular layer interneurons) of both SCA8 patients and transgenic mice that express a human AXTN8/ATXN8OS gene with a (CTG)₁₁₆ expansion mutation (SCA8 BAC-exp) (31). These mutant mice develop a progressive neurological phenotype, striking gait disturbances and a decrease in cerebellar-cortical inhibition (18,31). While both low copy SCA8 BAC-exp and heterozygous Mbnl1^ΔE3/+ knockout mice are overtly normal, crosses between these lines result in progeny with enhanced rotarod deficits. If the toxic RNA-mediated mechanism is similar to DM1, then the reduction in cerebellar inhibition could reflect altered splicing and subsequent mis-expression of a protein involved in cerebellar inhibitory pathways. Indeed, this appears to be the case since the levels of the GABA transporter GAT4/GABT4 are elevated in both human SCA8 and mouse SCA8 BAC-exp brains and this has been linked to MBNL1-regulated alternative splicing of GAT4 pre-mRNA. These data are consistent with a model in which mutant ATXN8OS CUG^exp transcripts trap MBNL1 proteins which in turn causes mis-splicing of MBNL target RNAs that encode proteins important for GABAergic inhibition.

On the sense strand, ATXN8 transcription was initially missed in human brain samples perhaps due to CAG^exp amplification problems (18). Fortunately, ATXN8 RNAs are detectable by strand-specific RT–PCR using primers immediately downstream of the expansion in the mouse SCA8 BAC-exp model. Surprisingly, and unlike other CAG^exp expansion diseases, this RNA is predicted to encode a nearly pure polyQ protein (18). Cerebellar polyQ inclusions, detectable by monoclonal antibody (mAb) 1C2 that recognizes expanded polyQ tracts, are observed in Purkinje cells in human SCA8 and SCA8 BAC-exp mouse brains and brain stems, indicating that antisense transcription occurs through these microsatellite repeats. Thus, SCA8 is the first candidate for a microsatellite disease in which both RNA-mediated (RNA foci formation and mis-splicing) and protein-mediated (polyQ intranuclear inclusions) events have been documented (Fig. 1). Nevertheless, the pathogenic picture is incomplete. For sense transcription, it is not clear whether the levels of the ATXN8 transcription and polyQ protein, which cannot be detected by protein blot analysis, are sufficient to cause cellular dysfunction although pure polyQ proteins appear to be highly toxic, so susceptible cell populations may rapidly be eliminated (32). For antisense transcription, expression of CUG^exp RNAs leads to RNA gain-of-function effects on RNA splicing and loss of cerebellar GABAergic inhibition but the basis for the cerebellar atrophy observed in SCA8 remains an enigma since these neurodegenerative changes are not recapitulated in the SCA8 BAC-exp transgenic model. An additional complication is that another gene, KLHL1, is also transcribed in an antisense direction to the ATXN8OS gene and the 5′ ends of these genes overlap in a head-to-head configuration (33). Because antisense transcription could theoretically down-regulate expression of the opposite strand, ATXN8OS CTG^exp transcripts might decrease KLHL1 mRNA levels in SCA8. In agreement with this possibility, Klhl1 knockout mice show mild atrophy of the molecular layer in the cerebellum, a characteristic feature of SCA8 patients that is missing in the transgenic SCA8 BAC-exp model. Confirmation of this antisense loss-of-function model, which requires that KLHL1 gene expression is reduced in SCA8 brains, has not been reported.

One of the puzzling features of SCA8 is the low penetrance associated with this disease and reports that some unaffected individuals harbor ATXN8/ATXN8OS alleles with as many as 1000 CTG repeats (34–36). Perhaps bidirectional transcription through repeats leads to silencing of the genes at this locus in some individuals significantly lowering the levels of toxic RNAs and proteins. However, there are no reports of heterochromatin formation at this locus. Are ATXN8 and ATXN8OS coexpressed in the same cells and, if they are, what are their relative expression levels? Laser microdissection of specific cell types and strand-specific RT–PCR could be used to address this important question. Moreover, additional cell and animal model systems need to be developed to determine the relative toxicity thresholds of rCUG^exp RNA and polyQ.

Bidirectional transcription in FXTAS

In DM1, the most severe form of the disease is present at birth and is caused by very large CTG^exp (usually >1000 repeats) mutations while shorter expansions are associated with the adult-onset degenerative disease. A similar scenario occurs for CGG^exp mutations in the 5′-UTR of the FMR1 gene (37). FMR1 expansions cause two distinct diseases depending upon repeat length (38). In the normal population, FMR1 alleles contain 6–52 CGG repeats and larger expansions (>200 repeats) cause FMR1 transcriptional silencing and the neurodevelopmental disorder fragile X syndrome. Although longer repeats (55–200) were originally considered to be pre-mutations, these intermediate expansions are now known to cause a late adult-onset disorder, fragile X tremor/ataxia syndrome (FXTAS) (39). Clinical manifestations of FXTAS include cerebellar degeneration with loss of Purkinje cells, spongiosis of the deep cerebellar white matter and axonal swelling. The primary neuropathological hallmark is eosinophilic ubiquitin-positive intranuclear inclusions disseminated through the brain and spinal column.

Several lines of evidence suggest that RNA-mediated protein sequestration events similar to those observed in DM1 and SCA8 may be important for FXTAS pathogenesis. FMR1 mRNA levels increase 2–10-fold in FXTAS and these RNAs can be detected in the ubiquitin-positive intranuclear inclusions that have been reported to also contain an array of proteins, including lamin A, Purα, hnRNPA2/B1, MBNL1, HSP40 and the 20S proteasome complex (40,41). In support of the idea that some of these proteins are effectively trapped by rCGG^exp inclusions, overexpression of hnRNP A2/B1 or Purα suppresses neurodegeneration in a Drosophila (CGG)₉₀-EGFP transgenic model and Purα knockout mice develop tremors and seizures by 4 weeks of age, suggesting an essential role of Purα in brain development and function (42–44). FXTAS-associated CGG^exp repeats are also pathogenic independent of gene context. Mouse Fmr1 (CGG)₉₈ and (CGG)_∼120 knockin models have been reported which reproduce several features of FXTAS (45,46). Recently, transgenic mice have been generated in which expression of a (CGG)₉₀ repeat upstream of either Fmr1 or EGFP is driven by the Purkinje cell-specific L7 promoter (47). In contrast to control lines expressing the corresponding transgenes minus CGG^exp, both Fmr1- and EGFP-driven (CGG)₉₀ lines develop ubiquinated intranuclear inclusions, Purkinje cell axonal swelling and increased neuronal loss. Interestingly, hnRNP A2/B1 and Purα do not colocalize with these CGG^exp inclusions.

Antisense transcription may also play a role in FXTAS pathogenesis (19). A promoter located in the intron 2 of FMR1 drives expression of the antisense ASFMR1 gene through the CCG repeat region. ASFMR1 transcript levels increase in blood cells from FXTAS patients and the ASFMR1 transcript is spliced, polyadenylated, transported to cytoplasm and may be translated into a polyproline protein using an initiation codon upstream of the CUG^exp although there is no evidence that this occurs (19). Multiple alternatively spliced forms of ASFMR1 RNA exist and the CGG^exp mutation alters the splicing pattern. Another noncoding gene, FMR4, initiates upstream of FMR1 start site (48). These two genes appear to share a bidirectional promoter since both transcriptional units are silenced in fragile X, but upregulated in FXTAS, patient leukocytes. Of course, these antisense repeats could also fold into RNA hairpins that sequester rCCG-binding proteins.

HDL2: one frame is not enough

Huntington disease-like 2 (HDL2) is another autosomal dominant and progressive neurodegenerative disease caused by a CTG•CAG expansion mutation which, like HD, results in prominent loss of striatal neurons (49–51). In contrast to HD, the HDL2 sense transcript, encoded by the Junctophilin-3 (JPH3) gene, contains a CTG^exp that is located in a 3′-UTR or in a coding region to produce either a polyalanine or polyleucine protein depending on the splicing of JPH3 exon 2A that contains three alternative 3′ splice sites upstream of the CTG repeat. HDL2 expansions are relatively short (40–59 repeats) which probably reflects the location of the repeats in both coding and noncoding regions. While a combination of protein- and RNA gain-of-function mechanisms could explain HDL2 pathogenesis, the location of the CTG^exp in a noncoding region suggests an RNA-mediated mechanism. Indeed, CUG^exp RNA foci are detectable in cortical neurons of HDL2 postmortem brain and these foci colocalize with MBNL1 (50). Mis-splicing of several MBNL1 targets, MAPT exon 2 and APP exon 7, in the HDL2 brain has also been reported.

Because of the similarities between HD and HDL2, protein gain-of-function mechanisms have also been evaluated. One obvious candidate would be a polyQ protein expressed from JPH3 antisense CAG^exp transcripts. In support of this possibility, the 1–2 µm ubiquitin-positive peri-nucleolar inclusions observed in HDL2 brains are also recognized by the mAb 1C2, which was originally elicited against the TATA-box-binding protein (TBP) that contains a polyQ tract (51). An argument against a role for JPH3 antisense transcription is that it has not been possible to detect CAG^exp transcripts from HDL2 brains using RT–PCR strategies and mAb 1C2 has been reported to cross-react with polyalanine and polyleucine, which are encoded on the sense transcript (50,51). However, another study failed to confirm that 1C2 recognizes either polyalanine or polyleucine (18) and anti-JPH3 antibodies do not detect expanded polyalanine or polyleucine proteins in the HDL2 brain (51). Another confounding factor is that TBP is detectable in HDL2 intranuclear inclusions (51). Similar to SCA8, these 1C2-positive inclusions do not generally colocalize with nuclear RNA foci in the HDL2 brain so sense or antisense transcripts may not be expressed at similar levels in the same cell populations or the combination of toxic RNA and protein leads to under-representation due to rapid cell death. Nevertheless, the issue of JPH3 antisense transcription in the brain should be re-investigated using RNA-seq technologies particularly in light of a recent asymmetric strand-specific analysis of gene expression (ASSAGE) report. Using normal blood mononuclear cells as well as several cell lines, this study demonstrated that JPH3 antisense transcripts exist and indeed are more abundantly expressed (6-fold more sequencing reads) than sense RNAs [Table 1 (14,20)].

Perspective

While contemporary studies on SCA8, FXTAS and HDL2 have revealed the potential for bidirectional transcription to play a formidable role in the pathogenesis of these diseases, many fundamental questions remain. Does repeat length influence bidirectional expression of the affected locus? Is it possible that toxic RNAs and proteins are coordinately produced from the same expansion mutation or do some cell types preferentially synthesize one over the other? Are cells that produce both toxic RNAs and proteins particularly susceptible to expansion-mediated cell death? Answers to these and many related questions require the development of a new generation of cell- and animal-based models that are designed to specifically evaluate the potential for combinatorial protein and RNA toxicity induced by bidirectional transcription.

Conflict of Interest statement. None declared.