Polyglutamine (polyQ) diseases are a family of nine neurodegenerative disorders caused by a cytosine-adenine-guanine (CAG) trinucleotide repeat expansion in the coding region of one of the polyQ disease-associated genes. PolyQ diseases include Huntington’s disease (HD), six spinocerebellar ataxias (SCAs) dentatorubral-pallidoluysian atrophy (DRPLA), and spinobulbar muscular atrophy (SBMA), each with its own causative gene: HTT, ATXN1, ATXN2, ATXN3, CACNA1A, ATXN7, TBP, ATN1, and AR, respectively (Williams and Paulson, 2008). With the exception of SBMA, all polyQ diseases are inherited in an autosomal dominant manner (Williams and Paulson, 2008). The CAG repeat region is translated into a stretch of glutamine amino acids, also referred to as the polyQ tract (Van De Warrenburg et al., 2002; Jana et al., 2005). In HD patients, the polyQ expansion in the huntingtin protein causes neurodegeneration that affects the striatum most severely and results in cognitive, psychiatric as well as motor disturbances and gait abnormalities (Rubinsztein, 2002; Paulsen et al., 2014). It is thought that the expansion of the polyQ tract causes the protein to misfold and aggregate, and therefore loses its normal function. Hence, these mutant proteins also become toxic components as they trigger the misfolding of other proteins (Arrasate and Finkbeiner, 2012).

In HD and SCAs, longer CAG repeat lengths in the causative polyQ genes are associated with an earlier age-at-onset (AAO) (Hmida-Ben Brahim et al., 2014; Tezenas et al., 2014). In HD, up to 75% of the variability in AAO can be explained by the HTT CAG repeat length (Hmida-Ben Brahim et al., 2014), while in SCA1, SCA2, SCA3, SCA6, and SCA7, the CAG repeat in the causative gene explains between 32 and 80% of the AAO variability (Tezenas et al., 2014). There is also evidence suggesting that, in addition to the expanded CAG repeat length in the causative gene, the normal CAG repeat lengths in other non-causative polyQ genes affect the AAO in HD and SCA’s (Pulst et al., 2005; De Castilhos et al., 2014; Hmida-Ben Brahim et al., 2014; Tezenas et al., 2014; Chen et al., 2016; Stuitje et al., 2017). For example, clinical observations show that the AAO in HD is not only affected by the expanded CAG repeat length in HTT, but also by the normal CAG repeat length in ATN1 and ATXN1 negatively affecting the AAO (Hmida-Ben Brahim et al., 2014). The observation that the number of CAG repeats in multiple polyQ genes can affect the AAO in HD or SCAs suggests that polyQ gene products are functionally interacting.

In addition to the suggested genetic associations based on shared effects on AAO, similar mechanisms likely contribute to polyQ disease pathogenesis. These mechanisms include misfolding of the disease protein, deleterious protein interactions, transcriptional dysregulation, mitochondrial dysfunction, aberrant neuronal signaling, cellular protein homeostasis impairment, and RNA toxicity (Williams and Paulson, 2008). Despite the similarities, these mechanisms seem to affect specific brain regions depending on the particular polyQ disease. To get a better understanding of how the CAG expansion in HTT affects the brain, multiple studies used genome-wide expression analysis of post-mortem samples collected from different brain regions. By comparing the brain region-specific expression profiles of HD patients and healthy controls, the highest number of differentially expressed genes was found in the caudate nucleus and to a lesser extent in the cerebellum (Hodges et al., 2006). This demonstrates that differential gene expression in the HD brain has a regional pattern corresponding to the known pattern of neuropathology. A more recent study used longitudinal RNA-sequencing expression data of HD knock-in mice with increasing CAG repeat lengths, and using weighted gene co-expression network analysis (WGCNA), they studied length-dependence of molecular networks in HD-relevant brain regions (Langfelder et al., 2016). They found one striatal module which showed strong patterns of downregulation that were both CAG length- and age-dependent.

In addition to patient postmortem samples and mouse models, characterizing expression patterns of wild-type genes in the healthy brain can provide useful insights into the molecular and pathological changes regarding polyQ diseases. Using gene expression data from ten different brain tissues and 101 healthy individuals from the UK Brain Expression Consortium (Trabzuni et al., 2011), Bettencourt et al. analyzed the co-expression relationships of SCA genes in the healthy brain using WGCNA (Bettencourt et al., 2014). Two cerebellar modules were enriched in SCA transcripts of which one module seemed preserved across all brain regions and the other module seemed unique to the cerebellum. The study suggests that genes co-expressed with SCA genes across all brain regions give rise to more complex phenotypes (other ataxia syndromes and neurodegenerative disorders), while cerebellum-specific co-expressing genes results in pure ataxia phenotypes. While previous studies have focused on region-specific gene expression differences between patients and healthy controls (Hodges et al., 2006; Neueder and Bates, 2014), and co-expression relationships between SCA genes in the healthy brain (Bettencourt et al., 2014), there is no detailed assessment of the relationships among polyQ genes in pathologically relevant regions of the brain.

We aim to find common mechanisms through which the nine polyQ genes could interact with each other in order to understand how these interactions could affect neuropathology in polyQ disorders. Our approach consisted of interrogating healthy brain gene expression data from the Allen Human Brain Atlas (AHBA) (Hawrylycz et al., 2015) as its high spatial resolution allows localizing interactions to specific brain regions. PolyQ genes with similar expression patterns in a specific brain area suggest their co-involvement in functions specific to that brain area. Using the AHBA to relate biological functions to disease and pinpoint pathways to specific regions of the brain has been reported previously, for example through evaluation of normal activity in the brain of genes associated to migraine (Eising et al., 2016), and autism (Parikshak et al., 2013; Willsey et al., 2013; Mahfouz et al., 2015). Here, we follow a similar approach to detect functional relationships between the nine polyQ genes. We focus on a brain region consisting of several brain areas which were recently shown to be most severely affected in HD through an MRI-guided unbiased approach (Coppen et al., 2016). In addition, the analysis was repeated for seven anatomical brain structures to assess whether the relationships between polyQ genes within the HD-associated region are reflected in other anatomical brain structures (Figure 1). We analyzed gene co-expression networks within the different brain regions to identify region-specific relations among polyQ genes and their underlying functional mechanisms.

Polyglutamine diseases share a similar genetic basis and their phenotypes, such as AAO, are affected by CAG repeat variations in the normal range in polyQ genes other than the causative gene (Pulst et al., 2005; De Castilhos et al., 2014; Hmida-Ben Brahim et al., 2014; Tezenas et al., 2014; Chen et al., 2016; Stuitje et al., 2017). The causative polyQ genes are thought to modulate protein interactions through the role of CAG tracts in stabilizing protein interactions (Schaefer et al., 2012). We analyzed the relationships between nine polyQ genes based on their co-expression within different regions of the healthy human brain.

We performed co-expression analysis in a brain region previously associated with HD specific damage in the cortical and subcortical gray matter. By combining imaging data from HD mutation carriers with spatial gene expression patterns from healthy brains, we were able to study the role of interactions between polyQ genes in the neuropathology of HD. Seven anatomical brain structures were also included in the analysis to examine regional differences compared to the HD-associated region. The anatomical structures also included the striatum, the main affected structure in HD. However, co-expression patterns in the striatum may be more specific to the structure itself rather than to HD pathology.

Among the nine polyQ genes, the relationship between ATN1 and ATXN2 was most pronounced based on our three representations of gene associations: it was the only gene pair that directly co-expressed, shared the most co-expressed genes, and also showed functional overlap in the HD-associated region. The same polyQ gene pair also shared co-expressed genes, and a functional overlap with HTT, suggesting functional connectivity with the causative gene of HD. Furthermore, out of the nine polyQ genes, only ATXN2 had a significant lower expression in the HD-associated region compared to the rest of the brain. An expanded polyQ tract in the huntingtin protein may dysregulate pathways in which all three polyQ genes are involved. The function of ATXN1 may also be affected in HD as this gene also shared co-expressed genes and showed functional overlap with HTT in the HD-associated region.

The co-expression patterns among polyQ genes within the HD-associated region overlapped with those observed within anatomical structures involved in HD neuropathology. In the striatum, the main structure affected in HD, ATN1 and ATXN2 were directly co-expressed and shared co-expressed genes. This suggests that this gene relationship has a more important role in HD than in other polyQ diseases. For the frontal and parietal lobes, ATN1 and ATXN2 were directly co-expressed and shared functional terms together with HTT, similar as for the patterns observed for the HD-associated region. The frontal lobe includes the primary motor cortex BA4, and is located next to the parietal lobe, which includes other motor areas. The motor cortex, involved in planning, control, and execution of voluntary movement, has been previously implicated in HD (Pfurtscheller and Berghold, 1989; Fogassi and Luppino, 2005; Nachev et al., 2008; Stinear et al., 2009). The motor cortex is also known to be directly connected to the caudate nucleus of the striatum (Muhammad et al., 2006). In a previous study, HD gene expression profiles in the caudate nucleus and motor cortex were strikingly similar (Hodges et al., 2006), suggesting that similar molecular mechanisms in different brain regions are involved in neurodegeneration. These mechanisms may include polyQ genes as we find similar co-expression patterns in the frontal lobe and striatum between HTT, ATN1, ATXN2, and ATXN1 (Figure 5).

There was a significant overlap in co-expressed genes between HTT, ATN1, and ATXN2 in all examined brain regions. It would be interesting to evaluate whether this triangular relation between HTT, ATN1, and ATXN2 is dysregulated in all polyQ diseases, especially HD, DRPLA, and SCA2. In addition, we observed strong connectivity of ATXN2 and AR with HTT in the pons that has been described to undergo atrophy in SCA2 (Ying et al., 2006). This relationship may possibly be disrupted in the pons of SCA2 patients. Altogether, these region-specific interactions demonstrate that co-expression analysis can reveal many interesting relations between genes based on their spatial information.

Accumulating evidence suggests that normal CAG repeat size variations in polyQ genes could act as genetic modifiers of AAO in different polyQ diseases. For example, the normal CAG repeat length in ATXN3, known to have a deubiquitinating function (Matos et al., 2011), has been found to have a positive effect on HD progression as it was associated with a later AAO (Stuitje et al., 2017). There is also evidence for a polyQ-length dependent interference of both the mutant and normal proteins with a range of other protein binding partners (Kaltenbach et al., 2007; Li et al., 2007; Benn et al., 2008; Lim et al., 2008; Yokoshi et al., 2014; Ashkenazi et al., 2017; Becker et al., 2017). Mutant proteins can interact either more strongly or weakly with other polyQ containing proteins and thereby inhibit there physiological function (Ashkenazi et al., 2017; Becker et al., 2017). A case in this regard is a recent study which showed that the polyQ domain of wild-type ataxin 3 enables it to interact with beclin 1, a key initiator of autophagy, allowing the deubiquitinase activity of ataxin 3 to protect beclin 1 from proteasome-mediated degradation and thereby enabling autophagy (Ashkenazi et al., 2017). They demonstrated that mutant huntingtin polyQ fragments competed with ataxin 3 for interaction with beclin 1 in a polyQ-length dependent manner, thereby inhibiting autophagy and contributing to neuronal dysfunction. This finding may explain the protective role of larger ATXN3 CAG repeat sizes in HD (Stuitje et al., 2017). In our analysis we did not find any relation in co-expression between HTT and ATXN3 in the examined regions. This might be due to the polyQ length-dependent competing between huntingtin and ataxin 3 being relevant only with pathologically expanded polyQ tracts, which were not present in AHBA. Furthermore, atrophin 1 and the androgen receptor have been found to bind to beclin 1 in a polyQ length-dependent way. Among the polyQ genes and examined brain regions in our study, we found that BECN1 is only co-expressed with ATN1 in the pons. Several putative interactions with HTT reported in previous studies were also supported by the co-expression analysis in our study. In the HD-associated region, ATN1 and ATXN2 shared co-expression partners with HTT and both genes were also functionally related to HTT. For each of the genes it was shown before that variations in the normal CAG repeat length, together with the expanded CAG repeat in HTT, affect the AAO in HD (Hmida-Ben Brahim et al., 2014). The co-expression patterns between HTT, ATN1, and ATXN1 suggest a functional relationship in brain regions that are involved in HD pathology.

Ubiquitin may be involved in HD pathogenesis through the relationships between HTT, ATN1, and ATXN2. The ubiquitin-proteasome system (UPS) has been linked extensively to the pathogenesis of neurodegenerative diseases, including HD and other SCAs (Williams and Paulson, 2008; Arrasate and Finkbeiner, 2012; Atkin and Paulson, 2014; Dantuma and Bott, 2014; Ortega and Lucas, 2014; Bettencourt et al., 2016). Aggregates in polyQ diseases show parts of ubiquitin and several important homeostatic proteins (Bowman et al., 2005). We found that E3 ubiquitin ligase genes UBR4 and CHIP/STUB1, and ubiquitin-activating UBA1 gene to be co-expressed with HTT, ATN1, and ATXN2. This suggests that the three polyQ genes contribute to the UPS in the HD-associated region of the healthy brain. Biochemical properties of CHIP are well-studied and this gene is known to interact with mutant polyQ proteins (including HTT, ATXN1, ATXN3, and AR) suppressing polyQ aggregation by promoting proteasomal degradation (Jana et al., 2005; Miller et al., 2005; Adachi et al., 2007; Williams and Paulson, 2008; Kuiper et al., 2017). CHIP has also been implicated in several SCAs indicating it is directly involved in disease pathology (Ronnebaum et al., 2014). The gene UB4R has a fundamental role in calcium signaling and neuronal survival and may contribute to neurodegenerative conditions (Parsons et al., 2015). This gene has been identified as a candidate locus in early onset episodic ataxia (Conroy et al., 2014). It is not known whether recognition of aberrant proteins by ubiquitin ligases is beneficial or disadvantageous (Chhangani et al., 2013). UBA1 is an important regulator of cellular protein homeostasis and contributes to the pathogenesis in SBMA and HD (Liu and Pfleger, 2013; Groen and Gillingwater, 2015). Altogether, we found genes involved in the UPS that co-expressed with multiple polyQ genes which have been associated with similar pathologies in neurodegenerative disorders. The UPS is involved in protein homeostasis and may be affected early in HD through strong regional interactions. If there are changes in huntingtin protein function due to polyQ expansion, then the interaction between ATN1 and ATXN2 is likely to be affected. The low expression of these genes in certain brain regions could indicate their increased vulnerability to dysregulations of the UPS. We suspect co-expressed polyQ genes to have a role in AAO, as they and their shared co-expression partners seem to be involved in functions that have been associated to polyQ diseases.

There are several issues that limit a one-to-one comparison between the co-expression relationships reported in the present study and phenotype-genotype associations described previously (Pulst et al., 2005; De Castilhos et al., 2014; Hmida-Ben Brahim et al., 2014; Tezenas et al., 2014; Chen et al., 2016; Stuitje et al., 2017). First, we do not expect all phenotype-genotype associations to be explained by changes in co-expression patterns. Second, there is a wide degree of heterogeneity in both the data and methods used in the genetic association studies (Pulst et al., 2005; De Castilhos et al., 2014; Hmida-Ben Brahim et al., 2014; Tezenas et al., 2014; Chen et al., 2016; Stuitje et al., 2017), limiting the possibility to combine their findings. Third, thus far, not all possible pairs of polyQ genes have been tested in the genetic association studies, highlighting the need for more comprehensive genetic association studies in larger cohorts of patients with polyQ disorders.

To validate our findings in other brain gene expression data we need a dataset with a spatial sampling resolution that allows co-expression analysis within substructures of the brain. Although the UK Brain Expression Consortium (Trabzuni et al., 2011) sampled multiple (10) brain regions per healthy individual, the expression within a substructure, e.g., the cerebellum, is still represented by a single sample. Gene expression data of HD patients may be used to observe whether co-expression patterns are altered in disease state. However, existing datasets are particularly rich in the number of individuals they sampled and not brain regions (Hodges et al., 2006). This captures variation across individuals, while in our gene co-expression networks we capture spatial variation within a brain region of interest (e.g., striatum).

We showed that polyQ genes are co-expressed in the healthy brain and that their relationships are also specific to certain brain regions including a region associated with HD. Our aim was to find co-expression patterns between polyQ genes in different brain regions. The co-expression networks are likely altered in polyQ diseases due to interaction changes, especially in brain regions associated with the disease. The fact that these findings could not be validated on other expression datasets implies the importance of follow-up studies to understand more about the mechanisms behind polyQ diseases. We show that gene expression in the healthy brain may render specific regions vulnerable to expression changes based on gene co-expression networks.

AK, NA, MR, and AM designed the study. BL provided funding and participated in the conceptualization of the study. JvdG provided the imaging data, which was processed by OD. AK performed the data analysis. AK, NA, WvR-M, MR, and AM interpreted the data and wrote the manuscript with input from all authors. AM and MR supervised the overall project. The final manuscript was read and approved by all authors.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.