Genome-wide transcriptomic and proteomic studies of Rett syndrome mouse models identify common signaling pathways and cellular functions as potential therapeutic targets

1 INTRODUCTION

Rett syndrome (Rett syndrome; MIM# 312750) is a severe X-linked neurological disorder which mostly affects females and is the second most common cause of severe intellectual disability in females after Down syndrome (Ellaway & Christodoulou, 2001). Rett syndrome is recognized as a disorder of neuronal plasticity characterized by the loss of intellectual functioning, fine and gross motor skills, communication skills, deceleration of head growth, and the development of stereotypic hand movements occurring after a period of apparently normal development (Neul et al., 2010). The disorder is mostly caused by mutations in the gene encoding methyl CpG binding protein 2 (MECP2) (Amir et al., 1999), a multifunctional protein which is highly expressed in the brain. The MECP2 gene is located on the long arm of the X chromosome at the Xq28 locus (Liyanage & Rastegar, 2014) and due to alternative splice sites, gives rise to two distinct isoforms, MeCP2e1 and MeCP2e2 (Kriaucionis & Bird, 2004; Mnatzakanian et al., 2004). MeCP2 was first purified in 1992, where it was identified as a chromatin associated nuclear protein capable of binding methylated CpG sites (Lewis et al., 1992). The protein is ubiquitously expressed but found to be most abundantly expressed in mature neurons (Kriaucionis & Bird, 2004). Initially, MeCP2 was believed to function as a transcriptional repressor, given its ability to bind methylated DNA and recruit chromatin remodeling complexes containing histone deacetylase enzymes (Jones et al., 1998; Nan et al., 1998). However, more recent studies have demonstrated a MeCP2-mediated mechanism for activation of multiple genes, including brain-derived neurotrophic factor (BDNF) and glucose transporter 3 (GLUT3) (Chahrour et al., 2008; Chen, Shin, Thamotharan, & Devaskar, 2013). MeCP2 also regulates microRNA processing (Cheng et al., 2014) and is also considered an epigenetic regulator (Mcgraw, Samaco, & Zoghbi, 2011). It is now widely accepted that expression of MeCP2 is critical for maintaining normal neuronal and synaptic function. Many studies have been conducted on murine and patient samples (e.g., blood and postmortem brain) to better understand the disorder and identify potential therapies. Following the identification of MECP2, these studies have led to the discovery of additional disease genes, such as cyclin-dependent kinase-like 5 (CDKL5), forkhead box protein G1 (FOXG1), myocyte-specific enhancer factor 2C (MEF2C), and transcription factor 4 (TCF4) (Armani et al., 2012; Evans et al., 2005; Philippe et al., 2010). Pathogenic variants in these genes, albeit to a lesser degree, are also associated with Rett syndrome, and display clinical phenotypes that overlap with this disorder.

Recent estimates recognize that around 95% of classical Rett syndrome cases and 75% of atypical Rett syndrome cases have mutations in MECP2 (Krishnaraj, Ho, & Christodoulou, 2017). To date, over 900 unique variants have been identified within the MECP2 gene with about 500 being pathogenic or likely pathogenic and the rest being benign, likely benign, or variants of unknown significance (Krishnaraj et al., 2017). Before the ground-breaking discovery in 1999 that pathogenic variants in MECP2 caused Rett syndrome, the diagnosis of Rett syndrome was based purely on clinical diagnostic criteria as no useful biomarkers of the disorder or genes had been identified. Despite this discovery, due to the complex nature of the disorder, the diagnosis of Rett syndrome still relies largely on the clinical diagnostic criteria, and the exclusion of differential diagnoses (Gold, Krishnarajy, Ellaway, & Christodoulou, 2018). In addition, effective treatments remain to be identified and treatment is still largely symptomatic. A better understanding of the underlying biological, genetic, transcriptomic and proteomic architecture of the disorder and the functional pathways would significantly improve prospects for novel targeted therapies for Rett syndrome patients.

Transcriptomic and proteomic studies have been successful in identifying new candidate genes and proteins, providing a better understanding of the underlying pathophysiology, identifying perturbations in functional pathways, and providing avenues for novel areas of research. In addition to the data generated for a specific research question, there lies a wealth of additional untapped information potentially available within these “omic” studies. This then raises the question as to whether we can learn anything new from previously conducted studies through comparing existing data sets and ultimately extracting valuable information by conducting meta-analysis. A notable example of utilizing omic data within the Rett syndrome research field is the recent study by Shovlin and Tropea (2018) that delineated transcriptomic data from human Rett syndrome patient samples, to present trends in differential gene expression for investigating broader research questions.

A compelling approach to unravel novel molecular insights into the role of MeCP2 and the pathophysiology of Rett syndrome is genome-wide expression (transcriptomic) profiling and proteomic profiling. The objectives of transcriptomic profiling, whether it is using microarrays or RNA sequencing (RNA-Seq), and proteomic profiling, are to quantitate and compare gene and protein expression profiles of two or more samples to detect differentially expressed genes and proteins. In addition to single genes and proteins, identifying dysregulated and overrepresented functional pathways, biological networks, and cellular functions may also reveal novel biological insights that contribute to the disease process.

In this review, we focus on published transcriptomic and proteomic studies that were performed in mouse models of Rett syndrome, highlight the main findings and discuss where future research efforts in this field should be directed. The aim of this review is to provide a summary of all the transcriptomic and proteomic studies conducted in mouse models of Rett syndrome studies, to highlight common dysregulated genes and perturbed functional pathways, biological networks, and cellular functions salient to Rett syndrome.

2 STUDY CHARACTERISTICS

A comprehensive list of all the studies are outlined in Table 1. Studies were conducted on a range of tissue samples, primarily in the brain given the predominant neurological nature of Rett syndrome. Transcriptomic studies were conducted on various Mecp2-deficient mouse models. Most studies were conducted in males, with two exclusively in females (Bedogni et al., 2016; Zhao et al., 2017), and one study using both genders (Johnson et al., 2017). The studies were mostly carried out using knockout mice (null), with a small proportion in Mecp2-knock in mice (Mecp2^R270X, Mecp2^G273X, Mecp2³⁰⁸, Mecp2^T158A-Tavi, and Mecp2^R106W-Tavi) (Baker et al., 2013; Delepine et al., 2015; Gabel et al., 2015; Johnson et al., 2017), and Mecp2 overexpressing mice (Tg) (Ben-Shachar, Chahrour, Thaller, Shaw, & Zoghbi, 2009; Chahrour et al., 2008; Chen et al., 2015; Samaco et al., 2012).

In an attempt to identify region specific gene expression, studies were conducted on individual brain regions and cells isolated from these regions. These included the cortex (Kishi et al., 2016; Pacheco et al., 2017; Tudor, Akbarian, Chen, & Jaenisch, 2002; Urdinguio et al., 2008), visual cortex (Gabel et al., 2015; Renthal et al., 2018), the cerebellum (Ben-Shachar et al., 2009; Gabel et al., 2015; Jordan, Li, Kwan, & Francke, 2007; Mellen, Ayata, & Heintz, 2017; Mellen, Ayata, Dewell, Kriaucionis, & Heintz, 2012; Sanfeliu, Hokamp, Gill, & Tropea, 2019; Urdinguio et al., 2008), whole brain (Guo et al., 2014; Kriaucionis et al., 2006; Nuber et al., 2005), the whole hippocampus (Baker et al., 2013; Tudor et al., 2002), and the amygdala (Samaco et al., 2012). Studies were also conducted on cell sources originating from distinct regions of the brain including forebrain (Tudor et al., 2002), hypothalamus (Chahrour et al., 2008; Chen et al., 2015), midbrain (Urdinguio et al., 2008), and striatum (Zhao, Goffin, Johnson, & Zhou, 2013), granule cells of the dentate gyrus hippocampal region (Smrt et al., 2007), embryonic cortical neurons (Bedogni et al., 2016; Vacca et al., 2016), and excitatory and inhibitory cortical neurons (Johnson et al., 2017). Two studies used a combination of four distinct neuronal cell populations including thick tufted pyramidal neurons from the motor cortex, fast-spiking parvalbumin-positive interneurons from the motor cortex, noradrenergic locus coeruleus neurons, and cerebellar Purkinje cells (Ehrhart et al., 2016; Sugino et al., 2014). Other studies used astrocytes (Delepine et al., 2015; Yasui et al., 2013) and microglia and peritoneal macrophages (Cronk et al., 2015), while others used cardiovascular progenitor cells (Hara et al., 2015) and cortical callosal projection neurons (Kishi et al., 2016). Studies using cell sources not originating from brain regions include the use of fibroblasts (Orlic-Milacic et al., 2014), skeletal muscle (Gold et al., 2014), and blood (Sanfeliu et al., 2019). Sample sizes of the studies varied greatly with the highest sample size being 205 (Sanfeliu et al., 2019) and the others varying between 1 and 100 samples. The age of mice used in the studies varied between embryonic, juvenile, and mature with the youngest being E15 (Bedogni et al., 2016; Vacca et al., 2016) and the eldest 25 weeks (Tudor et al., 2002).

The proteomic studies identified were conducted on three different Mecp2 mouse models: Mecp2-null, Mecp2³⁰⁸, and Mecp2^Jae mice (Cortelazzo et al., 2017; Matarazzo & Ronnett, 2004; Pacheco et al., 2017). The ages of mice ranged from 2 weeks to 12-months old. Two studies were conducted on male mice (Matarazzo & Ronnett, 2004; Pacheco et al., 2017) with one study on female mice (Cortelazzo et al., 2017). Two studies were conducted on brain regions (Matarazzo & Ronnett, 2004; Pacheco et al., 2017) and one on plasma (Cortelazzo et al., 2017).

3 ANALYSIS PLATFORMS USED

Oligonucleotide microarrays were the most extensively used method for the transcriptomic studies, especially the Affymetrix GeneChip and Mouse Exon subfamilies of microarrays. Other platforms such as the Affymetrix Mu11k and MGU74A arrays, Affymetrix U430 arrays, ADDER, cDNA microarrays, Illumina, and Agilent were also used. Interestingly, in experiments carried out after 2012, next generation sequencing technologies started becoming a more predominant platform of choice. Data analysis approaches varied in stringency between experiments, with cut-off values ranging from p < .05 to .005 and fold change (fc) from >1.2 (<0.67) to >2 (<0.5). These different methodologies are of great consequence to the interpretation of data as subtle changes caused by Mecp2 deficiency may be reported to be important in the less stringent tests which do not reflect the real function of Mecp2 (Raman et al., 2018). The opposite can also happen, as too few genes may be reported in more stringent tests, hence “loosing” critical data. One would expect to find a larger list of potential dysregulated genes in the less stringent tests (p < .05; fc > 1.2) than the more stringent ones (p < .005; fc > 2), however this is not always the case which may be a reflection of the technical and design differences between the studies such as sample size, statistical cut-offs, mouse models, statistical power, ages, sex, and anatomical regions studied.

The three proteomic studies used slightly different protein analysis and quantification methods, with Matarazzo and Ronnett (2004) using 2D gel electrophoresis and Q-T of mass spectrometry. Cortelazzo et al. (2017) used 2D and MALDI-ToF/ToF, while Pacheco et al. (2017) used an in-gel tryptic digest and LC-MS/MS.

4 COMMON DIFFERENTIALLY EXPRESSED GENES AND PATHWAYS

The most striking evidence emanating from this review is the lack of concordance in the dysregulated gene lists between the different studies. This may be attributed to a number of aspects including the fact that no two studies conducted gene expression profiling in the same mouse model, in the same tissue and at the same age. It would be expected that studies using the same, or similar sources of material would have a higher degree of concordance in their differential gene expression lists, however, this is not the case, which highlights the complexity of the disorder and how delicately Mecp2 regulates gene expression both spatially and temporally. A collation of the highlighted differentially expressed genes in these studies did reveal a few genes that showed altered expression across three or more studies which include interleukin-1 receptor-associated kinase (Irak1), Efna5, fatty acid binding protein 7 (Fabp7), FKBP prolyl isomerase 5 (Fkbp5), Plagl1, Fgf11, Homer2, Nsdhl, and Sgk1. However, as no meta-analysis have been conducted on the raw data from all, or a subset of these studies, it is difficult to determine whether these genes are an underrepresented list or indeed a true reflection of commonly dysregulated genes.

Owing to such a small number of commonly dysregulated genes among these studies, another approach to better understand the etiology of Rett syndrome is to interrogate groups of dysregulated genes that fall into distinct functional signaling pathways and biological networks. A comparison of altered pathways as opposed to individual genes may overcome the variability observed between each experiment. These variabilities may be due to biological sample differences, experimental noise, and differences in experimental and analysis approaches, or they could equally be valid discrepancies due to the tight temporal and spatial gene regulation of Mecp2. Either way, grouping differentially expressed genes that fall into the same functional molecular pathways may provide more insight into the biological pathways, providing new directions for therapeutic interventions.

Not all the studies ascribed groups of genes to specific biological networks, pathways, or cellular functions and mechanisms. However, those that did map the differential gene expression to pathways and cellular functions highlighted the diversity of the pathways, attesting to the great complexity of the disease. Several studies reported dysregulation in the nuclear factor κB (NF-κB), tumor necrosis factor (TNF), and toll-like receptor (TLR) signaling pathways. In addition, perturbations in biological mechanisms and specific cellular functions associated with neuronal migration, immune response, stress response, mitochondrial dysfunction, lipid metabolism, neuronal maturation, imprinting, and those relating to synaptic function, specifically synaptogenesis, plasticity, function, and transmission were also highlighted.

5 DYSREGULATED NF-κB, TLR, AND TNF SIGNALING PATHWAYS

Several studies reported an upregulation in the expression of the Irak1 gene (Jordan et al., 2007; Kishi et al., 2016; Urdinguio et al., 2008) implicating the NF-κB, TNF, and TLR signaling pathways. The transcription factor NF-κB controls the expression of many genes encoding cytokines involved in inflammation, TNF is a proinflammatory cytokine involved in a number of biological processes including the immune response, and the TLR signaling pathway can lead to proinflammatory cytokine genes being activated and transcribed. This inflammatory response most likely mediates neuroinflammation and may contribute to neuronal dysfunction in the brains of individuals with Rett syndrome.

This concordance in Irak1 expression was despite different brain regions and age of mice being investigated (cortical callosal projection neurons in 2-week-old Mecp2^−/Y mice (Kishi et al., 2016), cortex, midbrain, and cerebellum in Mecp2^−/Y mice aged between 6 and 10 weeks (Urdinguio et al., 2008), and cerebellum of male Mecp2^tm1.1Jae/Y and Mecp2^tm1.1Bird/Y mice (Jordan et al., 2007) and suggests the activation of the signaling pathways is a pervasive feature of Rett syndrome and not restricted to brain region or mouse age. Irak1 is a serine/threonine kinase that is activated by receptors IL1R and TLRs resulting in the translocation of NF-κB into the nucleus and the transcription of proinflammatory cytokines (Janssens & Beyaert, 2003). More recently, IRAK1 has been shown to play a key role in the NF-κB signaling pathway, a pathway which is closely associated with neuronal development and its consequent dysregulation can result in neurodegenerative diseases (Gutierrez & Davies, 2011). In addition, and pertinent to Rett syndrome, Kishi et al. (2016) reported that alleviating the abnormal NF-κB signaling pathway extends the life span of Mecp2-null mice, confirming the importance of this pathway in the disorder. As Irak1 expression is central to the functioning of this pathway, and the abnormally increased expression of Irak1 is associated with Mecp2 loss of function, significant interest is gathering in the investigation of the possible role of IRAK1 regulation in the NF-κB pathway and the pathogenesis of Rett syndrome. A dysregulated immune-inflammatory response implicating the NF-κB pathway has more recently been observed by Sanfeliu et al. (2019) in brains and blood of mice. However, although the authors did not specifically mention Irak1, the overall findings confirm the immune-inflammatory response in Rett syndrome mouse models. Finally, it must be noted that the Irak1 locus lies adjacent to the Mecp2 locus on the X chromosome and the apparent increase in expression has been proposed hold a “bystander” effect of Mecp2 deficiency.

Studies also reported a significant dysregulation in genes enriched in the TNF-mediated signaling pathway. Interestingly, these studies conducted by Vacca et al. (2016) and Cronk et al. (2015) used different cells of the brain, with Vacca et al. (2016) testing embryonic cortical cells from male Mecp2^−/Y mice and Cronk et al. (2015) testing microglia and peritoneal macrophages from Mecp2^Tm1.1Bird mice, suggesting again a pervasive response.

6 DYSREGULATED BIOLOGICAL NETWORKS AND CELLULAR FUNCTIONS

7 DISCUSSION

The main aim of this review was to provide a summary of all the studies and highlight the reported dysregulated genes and functional pathways that are found to be perturbed in Rett syndrome mouse models. This data can be further analyzed, using meta-analysis for the purpose of identifying new gene targets that may lead to therapeutics for the treatment of Rett syndrome or for unraveling novel biological insights that contribute to our understanding of the underlying pathophysiology of the disorder.

With the advent of new technologies such as RNA-seq, it is now possible to rapidly and effectively sequence and quantitate the gene expression of transcriptomes of interest in a way that has not been achieved before. However, despite this technology, the complete molecular picture for Rett syndrome is still unclear and by combining multiple omics studies, there is the potential to identify new pathways, reveal multigene interactions and establish the basis for disease progression.

Mouse models have been shown to recapitulate some of the features of the phenotype observed in patients (Calfa, Percy, & Pozzo-Miller, 2011) and have provided a wealth of information on the underlying pathophysiology of the disorder. In addition to reproducing the genetic architecture of Rett syndrome in humans, a number of the biochemical abnormalities are also well represented in these mouse models (Deng et al., 2007). Despite considerable research in identifying the role of MECP2 and other Rett syndrome-associated genes (Ariani et al., 2008; Weaving et al., 2004) in the pathophysiology of the disease, there is still a lack of clarity in the factors leading to disease progression.

There has been considerable success in utilizing mouse models to develop therapeutic strategies for Rett syndrome. A number of promising clinical trials including insulin-like growth factor, sarizotan, trofinetide, glatiramer acetate, dextromethorphan, desipramine, fingolimod, triheptanoin, ketamine, and lovastatin have been produced from murine preclinical studies (Gold et al., 2018). However, to date, no therapeutic strategy has been translated into the clinic. This may not be due to the lack of efficacy of these treatments, but rather to the confounding and complex nature of the disorder and a general lack of a deep understanding of the functions of MeCP2. Patients display a broad range of phenotypic features, each manifesting at varying time points throughout development. Therefore, further scrutiny of the mouse model omic studies may uncover more information that could support future clinical trials. Although care must be taken when extrapolating the preclinical results to potential human clinical trials, studies in mouse models are critical for identifying convergent molecular pathways in Rett syndrome.

In this review, we have identified published transcriptomic and proteomic studies conducted on Rett syndrome mouse models and highlighted the signaling pathways and cellular networks that were identified to be aberrantly regulated. Although many genes have been investigated for their involvement in the pathogenesis of the disorder, a set of commonly dysregulated genes has yet to be identified across all (or most) studies. This may be achieved using meta-analysis, however, caution should be exercised and careful consideration needs to be taken with the design of the analysis. One of the confounding factors that may result in a lack of concordant dysregulated genes between studies, among others, is the heterogeneous cell populations of the brain which may dilute the effect of Mecp2-deficiency. As gene expression profiles can vary significantly between different cell types, this could result in failure to identify a common list of differentially expressed genes. In addition, the subtle changes in the differential expression of many genes could contribute to the diversity of gene expression encountered, yet may miss being detected by transcriptomic and proteomic analysis because the expression differences were too subtle. The variations in mouse models, age and sex of the mice used could also contribute to discordant results between studies. This is highlighted by the transcriptomic study by Jordan et al. (2007), who compared the Mecp2^tm1.1Jae and Mecp2^tm1.1Bird mouse models at three different disease progression stages using the same transcriptomic platform, yet they were unable to find more than a handful of commonly differentially expressed genes between the models.

8 LIMITATIONS

A major limitation of conducting transcriptomic studies of whole brains or selected brain regions is the heterogeneous cell population consisting of many different cell types including a variety of neuronal cells, microglia, astrocytes, and oligodendrocytes as well as endothelial and fibroblast cells. It is widely accepted that different cell types confer different gene expressions and thus it is expected that the gene expression profiles will differ considerably among this heterogeneous cell population. Further, isolating specific brain regions of interest, especially small regions such as the hippocampus, is technically challenging, and therefore, the differentially expressed genes may be representative of the surrounding cells and not only of that particular region. Another limitation is that some samples may be pooled to meet the large RNA requirements for expression profiling. In this case, we should consider the possible effects of inter-mouse variation, where the gene expression of individual mice are so different that they dilute out any significant values.

Cultured cells also pose a limitation in that they no longer display the inherent phenotype and thus gene expression profiles that they would in situ. Microglia are particularly sensitive, as there are no in vitro methods that recapitulates all characteristics of adult homeostatic glia (Timmerman, Burm, & Bajramovic, 2018). Data has shown that RNA transcript profiles of ex vivo microglia differ considerably from those of in vitro microglia which has, to a certain degree, been attributed to culturing in the presence of serum (Bohlen et al., 2017).

Differences in study design and utilizing different measurement platforms is a major issue, introducing platform based inconsistencies, making it difficult to conduct meaningful and accurate comparisons between studies. Other inherent problems that introduce study bias include small sample size, different statistical cut-offs (p-value and fc), genetic background of transgenic mouse models, animal environment, data quality, statistical power, different mouse models, ages, sex, and anatomical regions studied. Lastly, despite Rett syndrome being a predominantly female-based disorder, the majority of these studies used male mice. Sexual dimorphism has been recognized as a major potential confounder in mouse studies (Karp, Heller, Yaacoby, White, & Benjamini, 2017).

9 FUTURE DIRECTIONS

Despite all these studies, no meta-analysis has been conducted using the raw data in an attempt to tease out any converging genes or signaling pathways with the idea of identifying novel curative treatments for Rett syndrome. There may be a number of reasons for this, which include, but not limited to, the raw data of some of the studies not being publically accessible. Moving forward, raw data from every study should be deposited in publically accessible online repositories where meta-analysis, using standardised analysis approaches, may be conducted. As such, statistical parameters such as p-values, corrections for multiple testing, sample pooling, and in the case of RNA-seq, read-depth, read length and single/paired end reads, may be applied universally across all data sets. Furthermore combining transcriptomics with proteomics will enrich the investigations which may provide more reliable “druggable” targets.

10 CONCLUSION

The importance of omics studies in driving discoveries in human genetics, especially in a complex neurodevelopmental disorder such as Rett syndrome is undeniable. The data generated from genome-wide studies supported by proteomics will systematically build an improved understanding of the key biological aberrations associated with MECP2 deficiency. We cannot also exclude the possibility of identification of “false positives” and inconclusive “leads” while we search for answers, however some of these leads may indeed open new avenues of research, bridging the gap between the known and the unknown. As a resource to quickly and comprehensively identify transcriptomic studies conducted in mouse models in Rett syndrome, we hope this review will assist researchers working towards tackling this devastating disorder.