The Known Unknowns: Missing Pieces in in vivo Models of Fragile X Syndrome

Fragile X Syndrome (FXS) is a rare disease and the leading monogenic cause of Autism Spectrum Disorders (ASD). It is caused by the silencing of the Fragile X mental retardation (FMR1) gene and the subsequent reduction or loss of fragile X mental retardation protein (FMRP). The clinical effects seen in FXS patients are several and highly variable making it difficult to model them in a single model or even one organism. Furthermore, several human behaviours can be measured only through surrogate endpoints in animals. Therefore, it has been challenging to develop in vivo models of FXS for drug discovery.

This review endeavours to consolidate the information on all available in vivo models for FXS specifically with a focus on their suitability for drug development, with the objective of identifying gaps and potential solutions. To do so, we have summarised the major clinical characteristics and possible mechanisms underlying clinical phenotypes associated with FXS and other disorders that arise from abnormalities in the FMR1 locus, such as fragile-X associated tremor/ataxia syndrome (FXTAS), fragile-X-associated neuropsychiatric disorders (FXAND) including ASD and fragile x-associated primary ovarian insufficiency (FXPOI). We then connect clinical features to phenotypes observed in available in vivo FXS models where possible, covering a wide range of organisms from primates, rats, mice, zebrafish, fruit fly and zebra finches. For each model organism, we list the technology of model creation, phenotypes/assays, mechanistic basis of disease manifestation and specific advantages or disadvantages of the model in the context of drug discovery.
Finally, we have highlighted the missing pieces in FXS modelling and propose strategies to address them, considering aspects of modelling spectrum disorders, repeat expansion and silencing, new functions of FMRP and identification of efficacious treatments. b) Specific connective tissue organogenesis c) Pathways of behaviours like anxiety, depression, irritability, others.
Thus, modelling FXS and improving the models can have great value to the field of neurosciences. In the present review, we discuss clinical presentations and possible mechanisms, various in vivo models and their advantages and disadvantages, and propose the next set of models and methods that can be used to plug in the missing pieces.

Clinical characteristics of Fragile X Syndrome and the suggested mechanisms
Clinical characteristics that are most commonly noticed in patients with FXS and the suggested mechanisms have been presented below in Table 1.

In vivo models of Fragile X Syndrome
Various in vivo models of FXS have been described in Table 2. We have also provided the potential advantages and disadvantages of each of these models.

Modelling spectrum disorders
Fragile X syndrome is the leading genetic cause of autism, and because of the implication of a single causal gene, animal models of FXS are numerous 26 . The characteristic clinical manifestation of FXS involves some or all of the following symptoms: long face, macrocephaly, prominent ears, prominent jaw, flat feet, joint hypermobility, macroorchidism (clinical); attention-deficit hyperactivity disorder (ADHD), anxiety autism spectrum disorder (ASD) (psychological), developmental intellectual disability, language deficits (developmental) and strabismus, recurrent otitis, gastrointestinal complaints, obesity and seizures (less prevalent) 44 . While the various models capture a subset of these phenotypes (see table 2 above), no single model has been able to mimic the spectrum of symptoms and deficits seen in human FXS patients, and this has severely impacted screening and drug discovery efforts. There could be two potential reasons for this: (i) The first is that the presence, severity and manifestation of FXS symptoms varies widely even in human patients 44 , and is likely influenced strongly by the genetic background and environmental factors. Therefore, one can argue that inconsistencies are expected in the models as well.

Introduction
Fragile X Syndrome (FXS) is one of the most studied monogenic neurological syndromes over the past few decades. It is caused by silencing of the Fragile X mental retardation (FMR1) gene and the subsequent reduction or loss of FMR protein (FMRP). CGG repeat expansion in the 5'UTR of the FMR1 gene, followed by hypermethylation of the region is the basis of the observed silencing in FXS 1 . This event occurs sequentially in successive generations, beginning with small repeat expansion  causing pre-mutation in one generation with a toxic gain of function in the mRNA, followed by further expansion (>200) to full mutation in subsequent generations with complete silencing of the gene 2 . FMRP is an RNA binding protein 3 , a well-known regulator of translation, and is known to interact with well over 800mRNAs in the adult neuron 4 . Even though several of the mRNA targets have not been fully characterised, it can be said that, FMRP loss has a cascading effect on several pathways which result in the observed clinical features 5,6 . Several model organisms have been used to model this disease, as discussed extensively in the following tables. The major challenges in modelling FXS are listed here (and substantiated in the rest of the review article): a) The clinical presentation of the disease in humans is highly variable with different individuals showing different sets of clinical phenotypes.
b) Similar to other neurological diseases, the biochemical and pathological profiling of FXS in real time is practically impossible. This makes it extremely difficult to decipher the human pathobiology making it necessary to have in vivo models.
c) The human nervous system is the most evolved and mimicking it in lower mammals and other vertebrates is done only by using surrogate endpoints especially for cognitive and social behaviours.
d) Animal models show high variability in measurable phenotypes.
Therefore, modelling a complex neurological syndrome like FXS is a continuing process and will require a set of organisms to model all the clinical characteristics, to study various pathobiological aspects and to screen potential drug candidates.
Modelling most diseases and syndromes is necessary in disciplines of disease biology and drug discovery; however, given FXS's monogenic aetiology, these models can be potentially studied for understanding several other aspects including (but not limited to): Loss of FMRP, altered translation and synaptic plasticity [22] Clear evidence of deficits in hippocampus-dependent spatial learning and memory (unlike murine models)  Dysregulation of excitatory and inhibitory neurotransmission by: a) mGluR1 mGluR5 (glutamatergic) enhanced signal transduction b) Deficits in GABA signalling Indirect glutamatergic mechanisms that modulate mGluR: a) Dysregulation of N-methyl-D-aspartate receptor (NMDAR) b) Altered expression, trafficking, and functions of Alpha-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid receptors (AMPARs) [5]- [8] Fragile-X associated tremor/ataxia syndrome (FXTAS) Excess mRNA from FMR1 premutation may lead to FXTAS in the progeny through the following mechanisms: a) Toxic FMR polyG protein production b) Formation of ubiquitin-positive inclusion bodies through protein & RNA sequestration c) DNA damage due to R-loop formation Its predicted that CCG repeats may lead to sequestration of specific RNA-binding proteins such as lamin A/C, Pura, hnRNP, Sam68, and drosha, thus affecting some of the normal cell function in both FXTAS and FXPOI.

Insufficiency (FXPOI)
Repeat-associated non-AUG (RAN)-translated poly-glycine species (FMR polyG) leading to intracellular inclusions bodies affecting ovarian function. There is an overlap of mechanisms suggested for FXTAS and FXPOI.
[15], [16] Other an adult neuron, for example), and consequently a number of signalling pathways are likely to be altered, and differentially so at various stages during development 45 . The phenotypes observed are a result of these cumulative changes. Even though FMRP per se is fairly well conserved in all of the models, varying degrees of conservation across the target proteins, pathways and differentiation paradigms are likely to cause inconsistencies in phenotypes observed in the models. Restricting discovery programs to one model organism, or picking a single genetic background in a model, is unlikely to be beneficial, since heterogeneity is a feature of the disease, not the model. Therefore, we would like to propose that diversification, both in terms of model organisms and genetic backgrounds may be key to identifying robust and efficacious therapeutic interventions. Focussing on simpler genetic models like Zebrafish and Drosophila, may allow one to conduct high throughput screens in a genetically diverse population, and generate statistically significant results.

Repeat expansion and silencing
In FXS, a repeat expansion in the 5' UTR of the human FMR1 gene leads to hypermethylation and silencing, which results in a drastic reduction of FMRP. Therefore, control at the level of repeat expansion or methylation are the best therapeutic avenues. In humans, the FMR1 locus naturally contains 6-55 CGG repeats; however, all of the model organisms (except the primate) contain very few repeats, if any. The mechanism of repeat expansion involves different stages in which the premutation stage (55-200 repeats) is associated with an increase in FMR1 RNA levels (toxic gain of function), which may be a prerequisite for the progression to full expansion (>200 repeats) leading to hypermethylation and subsequent silencing. Therefore, lack of a critical number of repeats at the outset, combined Potential model to study olfactory plasticity and subversion in neuronal function, and investigate the molecular link between expanded repeats and plasticity in a simplified system.
Only one published report of this model. The reproducibility of this model needs to be established.

Zebra finch, Taeniopyia guttata
No model yet Vocalization & Language development is dependent on fmr1 expression.
Mechanisms unknown [42], [43] Potential model to study the mechanisms behind speech and language pathology.
No model developed so far, feasibility unknown.
with a lack of progressive expansion could explain why the knock-in models do not show any promoter methylationdependent silencing. With a carefully chosen combination of artificial repeat insertion, control of FMR1 RNA levels and manipulation of the DNA replication or repair pathways to promote slippage, it may be possible to develop such a model. We believe that such studies can be conducted in simpler models like yeast 46 , used to derive optimal conditions which may then be moved into higher models like Zebrafish and mouse, to create true FXS models. Interventions which aim to interfere with repeat expansion, or those that revert the hypermethylation-driven silencing (small molecules [47][48][49] , or genetic interventions 50,51 ) can then be screened in such a model, and will either significantly ameliorate the disease (even a two-fold increase in FMRP level is associated with a significantly higher IQ 52 ) or better still, prevent it.

New functions of FMRP
Since FXS is primarily seen as a disease of the brain, studies have focussed on FMRP's neuronal function and neuronal phenotypes in the models. FMRP is ubiquitously expressed during early developmental stages, and it is increasingly clear that FMRP interacts with and regulates diverse cellular pathways in addition to its primary function as a translation regulator in neurons. These include regulation of RNA editing and splicing, chromatin structure, cellular differentiation kinetics, ion channel regulation and microRNA pathways 53 . Therefore, exploring the molecular mechanisms underlying the contribution of these pathways to disease progression or phenotypes could be an important new avenue of research. Traditional translation targets of FMRP may also be influenced by disruptions in these other pathways (for example, AMPA receptor and RNA editing 54 , microRNA and ion channels 55 ) and may be better rescued by novel treatments or combination therapies.

Identifying efficacious treatments for FXS-moving the needle on the preclinical side
While molecular mechanisms of FXS are fairly well understood, and small molecules targeting at least ten different pathways are able to rescue several molecular phenotypes such as protein synthesis, synaptic plasticity and calcium regulation, none of these have translated into clinical efficacy in humans 56 . While one obvious explanation is that this is due to the complexity of, and our lack of understanding of the human system, there are alternative explanations which require due consideration. An important take-away from the clinical trials conducted so far, is that tests which measure core phenotypes like behaviour and cognition directly and not through surrogate indications, need to be developed in order to determine the true efficacy of drug treatment. However, there are many avenues for improvement on the preclinical side.
(i) First, newer models which capture the repeat expansion and methylation features should be developed, since targeting these upstream nodes will result in maximal impact (section above).
(ii) Second, varied genetic backgrounds and multiple model organisms should be employed in order to determine the robustness of the phenotypes or treatment being assessed. The molecular phenotypes are highly conserved from flies to human, and the small molecules being considered for clinical trials have been identified based on these conserved pathways. It may be prudent therefore to conduct such studies in models like Danio rerio (Zebrafish) where true "wild type" animals (wild caught) can be used (incorporating the genetic diversity present in the natural world), with as large a sample size as required to power the statistics. Such an approach may allow one to incorporate the varied baselines in the population (for example, the median increase in anxiety in wild-caught vs. lab-wild type strains 50 ) and multi-factorial influences on the neurological phenotypes during the screening process to make results more robust 57 . Techniques for simple, rapid and inexpensive model creation, such as transient knockdown of gene expression (using DNAzymes or morpholinos in zebrafish 32,37,58 , RNAi in D.melanogaster 59 , and differentiation of patient derived cell lines 60 ) will also aid in increasing the number of varied models available to assess the same phenotype and its treatment. A larger number of treatments (compounds and paradigms) can be tested in a more diverse set of assays in such models, and may drastically improve the chances of finding a drug that will translate well into humans compared to traditional approaches using the mouse model.
(iii) Third, multiple tests or assays which measure the same parameter should be employed in each study, and at least one of these should measure the same parameter as in a clinical trial (such as fMRI or EEG).
(iv) Fourth, drug screening should be conducted in models where the link between the molecular changes due to FMRP loss, to circuitry and behaviour are well-established (such as in olfactory system of D.melanogaster 61 ), and rescue at each stage should be assessed.
(v) Fifth, a number of treatment windows may need to be explored for each class of drug, coupled with longitudinal studies which measure impact over the long term, especially for treatments targeting behaviour. Given that the circuitry can be modulated only in certain windows during development, earlier treatment windows need to be preferentially identified and studied.
(vi) Finally, FMRP appears to play a role in multiple unconnected pathways, therefore genetic studies in models like flies and zebrafish could be used to better understand these new molecular functions of FMRP. Subsequently, combination treatments to address more than a single target at a time may be prioritized for screening.

Disruptive platforms and niche models
Given that decades of therapeutic research using the available models of FXS have not led to the identification of a clinically efficacious drug, the possibility that the molecular landscape and regulatory network in the human brain is not sufficiently or completely replicated in any other model, has to be considered for the next phase of therapeutic research. In such a scenario, critical and validated endophenotypes 62 may need to be used as a basis for screening directly in a "human" model. Brain organoids satisfy the need for "human origin" as well as provide the genetic, cellular and architectural features of the human brain and could therefore be a powerful platform for identification of critical endophenotypes as well as for screening 63 . Brain organoids from patient derived cells (hiPSCs) have been used to study varied diseases of the central nervous system 64 , and are likely to be relevant to FXS, where phenotypes are thought to stem from defective cross-talk between multiple cell types and altered neural circuitry. However, the current state of this technology in terms of the time, skill and expense involved, as well as the inability to sustain organoids in culture to model adultphenotypes limits its applicability to drug discovery, as on date. Yet another strategy to conduct studies on human origin brain tissue, in an in vivo setting is the clever use of transplanted FXS brain tissue (iPSCs which differentiate after transplantation or neural precursor cells (NPCs)) into the mouse brain 65 . While the chimeric setting has the same limitations as the mouse model in terms of screening for behavioural or cognitive end points, it is likely to reveal the most authentic, cell-type and microenvironment specific responses to drugs.

Conclusions and future outlook
Fragile X syndrome is a classic test case for a rare disease model, which despite the availability of numerous models and studies over the decades, has not yielded any therapeutic benefits for patients. We believe that part of the reason for this has been the use of approaches which were developed and standardized on the basis of what has worked for the more prevalent, non-monogenic diseases that have dominated the clinical research and drug development fields. In the case of drug development for rare monogenic disorders where disease manifestation is heavily influenced by the underlying genetic background and treatment needs are primarily symptomatic, radically different approaches like the ones described above may be more fruitful. Implementation of such strategies and the consequent identification of efficacious treatments may cause a paradigm shift in the way rare disease biology, modelling and drug development is practiced in the future.

Funding and Acknowledgements
Work in the authors' laboratories is supported by funding from Dr.Reddy's Institute of Life Sciences, Hyderabad, India and Department of Biotechnology (DBT) grant BT/PR28305/GET/119/272/2018 awarded to PK and AS. The authors would like to acknowledge Dr. K. Chatti and Dr. K. Parsa for valuable comments on the manuscript.