Introduction
Animal models have been pivotal in understanding the etiology of many human diseases and determining effective therapeutic interventions [
1]. Research using animal models has unearthed mechanistic underpinnings and identified therapeutic targets for neurological disorders arising due to dysfunction of specific cell types or brain regions, e.g., Parkinson’s disease [
2]. Rodent models for diseases caused by viral or bacterial infections including some types of cancer and acquired immune deficiency syndrome (AIDS) have also led to an understanding of the fundamentals of the mammalian immune system leading to practical advances in healthcare management [
3‐
5]. However, for complex behavioral disorders that have a more diffused pattern, with multiple and sporadic genomic loci implicated in disease development, rodent models have evoked more questions than answers, be it schizophrenia, Down syndrome, or ASD [
6‐
8].
AutDB has focused on curating and annotating ASD research for the past 10 years using a scientific annotation framework rooted in the biology of the disorder [
9]. The systematic annotation of autism-related data on a standardized platform has been an invaluable resource to researchers seeking to sort through confounding and groundbreaking findings. Towards this end, ASD-associated AutDB gene and copy number variant (CNV) datasets have been used widely by the research community to understand the genetic heterogeneity of ASD [
10‐
12].
The
Animal Model (AM) module of AutDB was created to add depth and precision to the annotation of ASD-related models. The genetic models in AM are integrated with the corresponding gene in the human gene module of AutDB, providing human genetic evidence underlying each rodent model. Our database also includes various types of environmentally induced models for autism reported in the scientific literature. Several prenatal factors, including exposure to drugs [
13], role of paternal age [
14], and maternal immune factors circulating during gestation [
15], are being studied as causative or modulatory inducers of ASD. Additionally, the complex effects of chemical exposure and drugs after birth are also undergoing scrutiny [
16‐
21]. In contrast to Mouse Genome Informatics (MGI), the paramount resource for mouse genetics for over 35 years, AutDB is a specialized resource that includes diverse types of ASD-related animal models, evidence or hypothesis-based, including inbred strains showing face validity to ASD, annotated using a shared and standardized framework. Another distinctive and unique feature of the AutDB resource is the inclusion of rescue models, in which drugs and procedural, genetic, or dietary manipulations are used in rodent ASD models in an attempt to rescue ASD-relevant phenotypes. Together, AutDB represents a comprehensive resource including genetic and non-genetic animal models relevant in ASD biology.
Using data curated in the AM module over the past 8 years, we demonstrate characteristic patterns in analyses undertaken to study genetic and environmentally induced rodent models of ASD. Analysis of trends based on rodent model findings shows that the most frequently assessed phenotypes are related to core features of human ASD such as social interactions, ultrasonic vocalization, and repetitive behavior. Additionally, neuroanatomical features like changes in dendritic architecture, observed in postmortem human studies of ASD brains [
22], are also frequently examined in rodent models along with electrophysiology conducted on acute brain slices. These phenotypes could serve as a baseline in comparative studies of ASD models as a way of depicting complex behavioral phenotypes, related to underlying neurological substrates. With a view to facilitating translational research, we highlight pharmaceutical drugs administered to several ASD models. Finally, as a case study, we present a comprehensive analysis of phenotypes studied in rodent models of Shank3, one of the leading genetic risk factors of ASD.
Discussion
AutDB is a platform designed to be a specific resource where the AM module focuses on the in-depth annotation of genetic and non-genetic ASD models, using multiple layers of standardized vocabulary encapsulated in the Phenobase. Therefore, our database encompasses models based on high-confidence ASD genes (e.g., Chd8, Shank3), environmental inducers (e.g., VPA, MIA via exposure to viruses or viral mimetics like polyI:C), and inbred strains (e.g., BTBR). Rat models of ASD were recently added to AutDB to exploit interspecies conserved biology in investigating genetic and environmental ASD risk factors, also believed to be the best approach in the success of clinical trials. Additionally, rats are used in more studies assessing the effects of inducers, thereby increasing our repertoire of inducers tested in rodent ASD models. As the exact genetic signature of ASDs still remains to be determined and is a field of intense ongoing research, we believe that including putative and established models that represent characteristics of autism will lead to a more comprehensive understanding of this complex disorder.
At the time of data freeze for this article (March 2018), AutDB included rodent models based on 258 genes linked to ASD, linked to our
Human Gene module, providing details of all rare and common variants in these genes identified in affected individuals. In contrast, mouse models based on only around 60 genes have been linked to ASD in MGI (August 2018). In addition to providing ASD-specific genetic relevance to the animal models, a number of features distinguish our annotation model from MGI. First, our integrative approach includes genetic and non-genetic models of ASD within a single platform. Second, a standardized phenotypic repository (Phenobase) structured according to the diagnostic symptoms of ASD guides the annotation of all animal models in AutDB. Third, we make dedicated provisions for all the confounding details that can give rise to contradictory data and uphold robust findings. AutDB provides distinct information on the experimental paradigms used to assess a phenotype and relevant experimental details specific to studies, information that is not available in MGI. One of the primary confounders is the experimental paradigm used to assess a phenotype, e.g., anxiety can be assessed in as many as five to six direct testing protocols, including open-field test, light-dark exploration, elevated plus maze, and novelty-induced hypophagia [
31], while being reported as auxiliary observations from several other protocols like Morris water maze and social behavior testing. [
32,
33]. As noted in our results, anxiety is one of the most frequently assessed phenotypes in ASD models as well. Additionally, deficiency in olfaction, vision, or perception of pain also leads to confounding measures as these underlie the sensorimotor processes necessary to complete or perform most social or learning behavioral tasks. Olfaction and pain or nociception are the most frequently annotated auxiliary measures in our database, with 116 and 122 entries by the freeze date, still among the 35 most frequent phenoterms. Olfaction is observed to be normal in over 90% cases, whereas pain or nociception has been found to be normal in 68%, with clear instances of increased or decreased perception usually taken into account for behavioral data interpretation by authors. For researchers, AutDB provides the opportunity to compare these observations and many other sensory phenotypes in all the models based on particular genes, whether or not they have been tested in every study. Our dedicated annotation of phenotypes that are indicated as “No Change” allows researchers in bioinformatics and wet labs to eliminate known confounding sources of possible multifactorial phenotypes.
In our unique rescue model dataset within the AM, we highlight test outcomes from experimental drugs, FDA-approved drugs, behavioral interventions, and transplantations of remedial cells or microbes on rodent models of ASD. We also curate genetic manipulations that have been used to rescue ASD-related phenotypes. Genetic-, behavioral-, or transplantation-based rescue paradigms play an important role in understanding the contribution of different factors for the development of phenotypes in ASD, even if not all of them can be translated into human interventions. We aim to curate new technologies as their use in ASD research is increasing; however, caveats in their application emerge retrospectively, e.g., clozapine-N-oxide (CNO) used to activate designer receptors exclusively activated by designer drugs (DREADDs) is reverse metabolized clozapine (interestingly a drug tested in ASD models) in rodents and is not pharmacologically inert [
34]. Therefore, it is of great interest to the scientific community to combine the time-tested standard tasks and sophisticated frontier technology to unveil interesting relationships between different types of behavior.
The rescue model dataset of AutDB offers a comprehensive resource for translational studies. For example, genetic rescues on Shank3 models of ASD (see Additional file
1: Table S6) indicate the crucial role of normal Shank3 expression through adulthood, indicating the importance of these Shank3 mutant lines in testing remedial approaches for Shank3 mutations during postnatal development and adulthood [
35]. A faster and cost-effective route for new clinical therapy in ASD is the use of existing pharmaceutical agents or “repurposing” of drugs already approved by FDA for other conditions. As seen in Fig.
4, several known drugs have been tested in genetic models, including rapamycin, fluoxetine, clozapine, and risperidone. Rescue treatment paradigms and effects, annotated discretely for rescue models in AutDB with a new set of controlled vocabulary distinct from parent ASD models, are as important as the drugs used for rescue as dosage and length of treatment can significantly change the outcome in rodents and people. This is evident in the Shank3 rescue models where a drug, p-cofilin, tailored to act on a deficit specifically manifested in Shank3 mutants; dendritic spine formation displayed differences in outcome in low (0.15 picomol/g) versus high (15 picomol/g) doses. Some of the drugs tested in Shank3 models have also been tested in other mouse models of ASD; CDPPB has been used in rescue paradigms to treat Shank2 mutants leading to the normalization of social interaction [
36].
Translation from preclinical to clinical studies requires rigorous testing of a large number of paradigms, with changes in dosage and route of administration requiring. While not reported as often as favorable effects, the presence of adverse effects of drug treatments is noted in AutDB and could play an important role towards clinical drug trials. We believe refractory phenotypes should be reported whenever testing is conducted, so that the community can benefit from the knowledge and reduce unnecessary waste of resources. From the Shank3 dataset itself, it is surprising that the assessments for perseverative self-grooming and social approach or interaction were not reported after several treatments (Table
1). Overcoming this “positive data bias” is the basis for our comprehensive annotation of the “No Change” phenotype which indicates the absence of ASD-related phenotype or no statistical difference from control measures. A single model for ASD will likely not emerge from rodent studies for this genetically and clinically heterogeneous disorder. Similarly, it is unlikely that there will be a single drug prescribed for ASD therapy. As presented here, a number of the most frequent ASD-related phenotypes, like social behavior and anxiety, are assessed in Shank3 models. Additionally, based on the scaffolding function of the Shank3 at the synapse, other phenotypes like synaptic plasticity, glutamate neuroreceptor levels, and more detailed neurophysiology are also reported. This is where the advantage of rodent models lies in deciphering the constellation of phenotypes that group together and that has been the driving force for conducting the trend analyses presented here. For example, a recent study using optogenetics indicates that spatial learning can have direct effects on social behavior [
37] and it is likely that future research from different fields will enhance the existing knowledge on ASD biology with new insights. Our database is primed to curate new types of paradigms and knit together cumulative observations with new discoveries. Distinctive patterns are yet to be found in ASD research; however, it is our aim to provide the platform that facilitates their detection by standardized, unbiased annotation of observations extracted from ASD scientific literature.
Materials and methods
Data curation and annotation
AutDB consists of manually curated and annotated data from published, peer-reviewed scientific literature on the basis of relevance to ASD. For curation in the Animal Model module of AutDB, articles are selected for annotation based on a preliminary assessment of the validity of the rodent model and the detailed phenotypic characterization of the model. Our database is updated quarterly with annotations from the latest published literature. Details of the annotation process are documented in a wiki web resource (
http://174.79.186.155:18000/AM_wiki/index.php/Rodent_annotation_guideline).
The validity of an animal model is based on its relevance to ASD. The articles must describe animal models that are based on evidence from association studies in humans, or, alternatively, models that display strong face validity for ASD-consistent endophenotypes. An animal model where an ASD-associated factor is manipulated to assess resulting phenotypes is an evidence-based model, while a model showing ASD-consistent endophenotypes for a factor with no association with ASD is a hypothesis-based model. We attempt to build a comprehensive dataset for rodent models based on high-confidence ASD genes and prioritize reports containing detailed phenotypic data on those. We only annotate differences from controls that are reported to be statistically significant (p value < 0.05 is the cutoff that is conventionally recorded as significant in biological studies), after careful review of figures from the main text or supplementary material. Instances where control data are either not reported or mentioned as “data not shown” are not included in AutDB annotations. Most journals have stringent policies regarding statistical testing and data analysis, so even if controlling for effect size or power for each test reported is beyond the scope of AutDB, peer review and editorial supervision are expected to take those considerations into account. Additionally, rescue models are annotated based on rescue paradigms where an agent or intervention is used to alleviate a phenotype in an ASD animal model. We categorize these rescue paradigms based on the type of agent: transplantation-based, procedural, behavioral, genetic, or pharmaceutical. We have developed annotation methods to clearly represent the treatment effects on ASD phenotypes paralleled in rodent models.
The data freeze date for all data shown in this article is March 31, 2018.
Phenobase: dynamic hierarchical phenotypic metadata
The individual phenotypes observed in ASD models are annotated using controlled vocabulary (“phenoterms”) and organized into 16 broad categories in a resource termed as the Phenobase (Additional file
1: Table S1).These categories are grouped as “core” where the comprising phenoterms closely parallel ASD core phenotypes, “auxiliary” when the parallel human ASD phenotypes are not core diagnostic features of ASD, or “physiological” for most other associated phenotypes that are routinely assessed to determine biological underpinnings of ASD.
Phenoterms, which are arranged in a hierarchical manner, are based on endophenotypes observed in rodent models. As many complex endophenotypes of rodent animals are being reported, we have added phenoterms that capture specificity while still rooted in a broader term, such that broad terms at the top of the hierarchy are separated by colons from specific terms. For example, the term “Morphology of the basal ganglia: Striatum: Caudoputamen” specifies the morphological changes to the caudoputamen, a part of the striatum, which in turn is a part of the basal ganglia. Our phenoterms are intended to capture complex as well as simple endophenotypes that are physiological, robust or quantitative, and conserved between species. It should be noted here that the phenoterms have been developed in the context of curated ASD literature; therefore, they reflect the phenotypes assessed in ASD rodent models and not the full complexity and scope of a category per se.
In addition to phenoterms, the Phenobase contains a standardized list of experimental paradigms. This list is a discrete part of the database that is used to represent the tests used to study phenotypes in different animal models. The standardization of experimental paradigms adapts a uniform nomenclature that circumvents the variety of synonyms used in literature for similar experimental setups. The Phenobase catalogs are over 450 experimental paradigms that map to the whole set of 620 phenoterms. The combined use of phenoterm and experimental paradigm provides a more comprehensive picture of observations made by authors, which allows for better comparisons of model phenotypes between different ASD models as well as between researchers.
Data analysis and visualization
Most data analysis and frequency measurements were conducted in R using the following packages: “tidyverse,” “dplyr,” and “forcats” [
38]. Graphics were developed with ggplot2 or in excel [
39].