Twenty years of discoveries emerging from mouse models of autism

More than 100 single gene mutations and copy number variants convey risk for autism spectrum disorder. To understand the extent to which each mutation contributes to the trajectory of individual symptoms of autism, molecular genetics laboratories have introduced analogous mutations into the genomes of laboratory mice and other species. Over the past twenty years, behavioral neuroscientists discovered the consequences of mutations in many risk genes for autism in animal models, using assays with face validity to the diagnostic and associated behavioral symptoms of people with autism. Identified behavioral phenotypes complement electrophysiological, neuroanatomical, and biochemical outcome measures in mutant mouse models of autism. This review describes the history of phenotyping assays in genetic mouse models, to evaluate social and repetitive behaviors relevant to the primary diagnostic criteria for autism. Robust phenotypes are currently employed in translational in- vestigations to discover effective therapeutic interventions, representing the future direction of an intensely challenging research field.


Introduction
Identified by Leo Kanner in 1943(Kanner, 1943, and first listed in the Diagnostic andStatistical Manual in 1980 (American Psychiatric Association, 1980), autism entered the awareness of the general public only gradually. Rainman, the iconic 1988 movie in which Dustin Hoffman famously portrayed an adult with autism spectrum disorder, along with growing media coverage such as the 2002 Time magazine cover story "Inside the World of Autism" (Nash, 2002), increased awareness among parents, pediatricians, and teachers that autism is a distinct neurodevelopmental disorder, rather than a form of schizophrenia or "mental retardation" as previously assumed. Standardized, quantitative, clinical diagnosis of autism spectrum disorder became possible in 1989 with the landmark Autism Diagnostic Observational Schedule (ADOS) (Lord et al., 1989). Prevalence of autism spectrum disorder is estimated at 1 in 54 children in the United States (1.8%), and at approximately 1% averaged across 26 countries (Anon, 2021;Fombonne et al., 2021). These estimates include many people with co-morbid neurodevelopmental disorders such as Fragile X syndrome.
Diagnostic symptoms of autism focus on deficits in social interaction, social communication, restrictive repetitive behaviors, and unusual sensory reactivity (American Psychiatric Association, 2013). Behavioral interventions remain the standard of care, including Applied Behavioral Analysis and the Early Start Denver Model (Rogers et al., 2021). No pharmacological treatments for the diagnostic symptoms of autism have yet been approved by the U.S. Food and Drug Administration, although two antipsychotics are FDA-approved for treating an associated symptom, irritability (Abbeduto and Sahin, 2021). Many other classes of drugs are routinely prescribed for associated issues such as seizures, sleep disruption, anxiety, and attention deficit hyperactivity syndrome.
Worldwide research into the causes of autism have revealed a multitude of strong genetic substrates. Susan Folstein and the late Sir Michael Rutter published the first twin studies which documented significantly higher concordance for autism among monozygotic twins than for dizygotic twins and siblings (Folstein and Rutter, 1977;Castelbaum et al., 2020). In addition to mutations inherited across generations, the large preponderance of genetic alterations detected in people with autism, including single nucleotide polymorphisms and copy number variants, appear to be rare de novo mutations (Constantino, 2021;Searles Quick et al., 2021). Environmental risk factors, including older paternal age (Lyall et al., 2017) and immune dysfunctions (Ramirez-Celis et al., 2021), appear to contribute a small percentage of the risk for susceptibility to autism spectrum disorder.
Given the high prevalence of autism spectrum disorder, there is a pressing unmet medical need to understand its causes and to develop effective treatments. To these ends, the scientific community has been generating animal models of autism with mutations in genes implicated in the etiology of autism spectrum disorder (Ey et al., 2011;Kazdoba et al., 2016b). Mouse models investigated to date have informed our understanding of the roles of developmental and synaptic genes in E-mail address: crawley@ucdavis.edu.

Contents lists available at ScienceDirect
Neuroscience and Biobehavioral Reviews journal homepage: www.elsevier.com/locate/neubiorev https://doi.org/10.1016/j.neubiorev.2023.105053 causing neuroanatomical, physiological, and behavioral outcomes relevant to autism. Behavioral phenotyping of animal models is central to our understanding of human disorders in which behavioral symptoms predominate. This review highlights the history of translational evaluations of relevant social and repetitive behaviors in genetic mouse models of autism spectrum disorder over the past 20 years.

History of contributions from the laboratory of an IBNS past president
In the year 2000, when I was honored to be elected President of the International Behavioral Neuroscience Society, development of mutant mouse models of autism was just beginning. Genetic models of a few neurodevelopmental disorders existed, such as Down syndrome Ts65Dn mice (Reeves et al., 1995), Fragile X Fmr1 mice (D'Hooge et al., 1997); and ataxia telangiectasia Atm mice (Barlow et al., 1996). Always interested in biological mechanisms causing brain disorders, I had been reading the autism literature for several years. Ken Paigen, research director of The Jackson Laboratory (JAX), kindly invited me to participate in a JAX workshop sponsored by the National Institute of Mental Health (NIMH) on mouse models of autism. On December 4th, 2000, two sets of researchers assembled at the JAX Highseas Conference Center in Bar Harbor, Maine. 1 One was a group of clinical experts in the features of autism and related disorders. The other was a group of behavioral neuroscientists with expertise in mouse behavioral testing. We sat down and talked to each other. By the 4:30 pm wrap-up, we had generated an outline of existing mouse behavior assays with reasonable conceptual similarities to the core symptoms of autism (Insel, 2001).
Over the next years, through many interactive conversations with autism experts who were willing to brainstorm together, especially a strong collaboration with Joe Piven and Sheryl Moy at the University of North Carolina (UNC), I proposed a constellation of mouse behavioral assays with conceptual relevance to (a) the core diagnostic symptoms of autism, (b) associated symptoms, and (c) essential control measures (Crawley, 2004). Core diagnostic symptoms include unusual social interactions, impaired social communication, repetitive behaviors, and sensory hyper-and hypo-reactivity. Associated symptoms include intellectual disabilities, anxiety, attention deficit hyperactivity disorder, sleep disruption, and motor dysfunctions. Essential control measures to avoid overinterpretation of artifacts include physical health, ability to smell, hear, and see, and general exploratory locomotion. The framework of this behavioral phenotyping strategy has been adapted by a remarkable number of laboratories working with mouse models of autism spectrum disorder.
Scientists build on each other's creations. While giving a seminar at North Carolina State University, I was fortunate to tour the laboratory of Professor John Vandenbergh, where postdoc Miles Dean demonstrated simple equipment for evaluating sexual preferences in voles. Clinical experts had shown me that children with autism tend to play with one favorite toy rather than join other children in games. Reflecting on these descriptions of low social approach in children with autism, I envisioned adapting the standard rodent mate preference testing apparatus to identify low social approach in mice. Working with biomedical engineer George Dold in the Research Services Branch of the NIMH Intramural Research Program in Bethesda, Sheryl Moy, Jessica Nadler, and I invented an automated three-chambered apparatus that measures the normal tendency of mice to spend more time with a novel mouse than with a novel object, representing our definition of sociability, and to detect the absence of sociability in a mouse model of autism Nadler et al., 2004). Face validity is thus to the tendency of children with autism to spend more time with an inanimate toy, rather than approaching and playing with another child.
Controls were purposely built into the apparatus to measure general exploratory activity (number of entries, center time), and adequate sensory ability to detect social olfactory cues (preference for social novelty). To properly design the conditions of the 3-chambered social approach assay, our team evaluated factors such as the time course of social interactions , exploration of a completely empty wire cage versus a wire cage containing a novel object and the salience of visual versus olfactory cues in eliciting exploration (Ryan et al., 2008), age and sex of the subject mice ; strain of the target mouse inside the wire cup Yang et al., 2012a), and phase of the circadian light cycle (Yang et al., 2007. These studies provided evidence that the first 10 min in the apparatus captured the majority of social approach, that sociability was similar across adult ages for both sexes in both the light and dark phases of the circadian cycle, that an empty wire cage elicited as much exploration as a wire cage containing an odorless novel object, and that the strain of the target mouse did not affect social approach by the subject mouse unless the target mouse displayed aggressive behaviors. In our lab's experience, Ns of 12-15 mice per genotype, sex, and/or treatment, applied consistently across groups, generally yielded sufficient power for comparisons of genotype, sex, and/or treatment group. In our lab's experience, the sociability choice, i.e. more time spent with the novel target mouse than time spent with the novel target object, was a reproducible measure of sociability within a group, in re-tests of a cohort, and in independent replication cohorts of the same strain, genotype, or treatment. However, the absolute value of the number of seconds spent with the novel mouse varied considerably across cohorts of the same strain, genotype or treatment (e.g. B6 and DBA in Figure 4 and B6 in Figure 10, Moy et al., 2004;En2 in Figure 3, Breilmaier et al., 2012; vasopressin receptor Avpr1b in Figure 3, Yang et al., 2007;vehicle groups in Figures 1 and 2, Silverman et al., 2013). The 3-chambered social approach test is therefore robust and replicable as a choice assay, providing a simple yes-or-no, all-or-none, qualitative detection of statistically significant sociability or lack of sociability within a group of mice.
Comparison of the number of seconds spent with the novel mouse across genotypes, or across treatment groups, is beyond the sensitivity of the assay, because three-chambered social approach measures a choice. Some confusion has arisen about how data are correctly interpreted. In response to questions raised, additional information is offered here.

Problems with comparing number of seconds across groups
For mice tested in a choice assay, the absolute number of seconds that a subject mouse spends with one choice will necessarily contain inherent variability caused by levels of general exploratory locomotion. For example, Mouse A may display low exploration, spending only 10 s total time in both side chambers, in which 8 s are spent in the chamber with the novel target mouse, across the ten minute test session. Mouse B may display high exploration, spending 400 s total time in both side chambers, but spend only 8 sec in the chamber with the novel target mouse, across the ten minute test session. Relying on the absolute value of 8 sonds would lead to the false conclusion that Mouse A and Mouse B display identical sociability. In fact, a total of 10 s spent in both side chambers represents a sampling that is too small to be a reliable indicator of sociability for Mouse A. At the other end of the spectrum, Mouse B shows a clear choice, representing low sociability and probable avoidance of the novel mouse. In another scenario, Mouse C is also highly active, but spends 200 s in the chamber with the novel target mouse and 300 s in the chamber with the novel target object. 200 s with the novel mouse is a high number. If used alone, the 200 s score would lead to an interpretation that the subject has high sociability. However, Mouse C spent 300 s with the novel object. Therefore, Mouse C chose to spend more time with the novel object than with the novel mouse, correctly interpreted as absence of sociability.

Problems with using a percentage, ratio, or index
Many laboratories employ a percentage, ratio, or index to calculate a single number that reflects the choice of a novel target mouse versus a novel target object. These calculations are logical attempts to conduct statistical comparisons across genotypes and/or across treatment groups. However, derived statistics can lead to incorrect interpretations. Using the example above, Mouse A with low general exploration spent only 10 s in both side chambers during the 10 min session, 8 of which are in the chamber containing the novel mouse, and 2 in the chamber containing the novel object. Calculations would yield a percentage of 80%, a ratio of 8/10, or an index of 0.8. These values would be interpreted as high sociability. Again, the total of 10 s spent in both side chambers represents a sampling that is too small to be a reliable indicator of sociability.
Measuring total entries into the two side chambers, and scoring time spent in the center chamber, will help to identify low exploratory activity. Using these two control parameters can avoid incorrect interpretations.
Measuring time spent sniffing the novel target mouse and time spent sniffing the novel target object provides more accurate measures of sociability than time spent in each side chamber. This is because sniffing is scored when the subject mouse is close to the target, usually within 2 cm, with its nose facing the target. In contrast, time spent in a chamber only measures general proximity. Sometimes a subject mouse spends most of its time sitting in a corner within a side chamber, rather than approaching the target. Sniff time quantification is available with many available automated systems.

Statistical analyses
Several approaches appear to be appropriate for determining statistical significance of the choice between time with the novel mouse versus time with the novel object, as well as time sniffing the novel mouse versus time sniffing the novel object (Yang et al., 2011). An ANOVA with group as a between subject factor and mouse/object as a within subject factor is the most stringent. A robust post-hoc test following a significant one-way ANOVA can identify which choice comparisons are significant and which are not. Paired t-tests are appropriate for determining whether the choice within a single group was significant or not significant. We recognize that multiple t-tests can produce Type 1 errors. In our experience, it is best to design experiments with a small number of groups to avoid Type 1 errors. For example, an experimental design for a drug study could evaluate vehicle and two doses, in the two genotypes of greatest interest, = 6 groups.
Given the above issues, we consider the 3-chambered social approach task to be primarily a first indicator of sociability or lack of sociability within a group of mice. Its value lies in simple automation and high throughput. Publications claiming genotype differences or treatment effects based solely on time in the chamber with the novel mouse are therefore suspect, as the data may have been incorrectly interpreted. More quantitative comparisons across genotypes or across treatment groups require other social tests with greater sensitivity, such as reciprocal social interactions between two freely moving unfamiliar mice. Replication of results in a second experiment using a new cohort of mice is ultimately the best approach for confirming the strength of a finding.

BTBR inbred strain
As part of a collaborative genetics project with David Threadgill, Jessica Nadler, Terry Magnuson, Sheryl Moy and Joe Piven, we employed the 3-chambered sociability task to test 20 diverse strains of mice available from The Jackson Laboratory . Having visited an intriguing poster about the BTBR T+tf/J 2 inbred strain at a Society for Neuroscience meeting by friend and future collaborator Valerie Bolivar (Bolivar et al., 2007), I encouraged the UNC team to include the BTBR inbred strain in the project. BTBR proved to display absence of sociability in the 3-chambered test, low reciprocal social interactions in juvenile and adult dyads, impaired social transmission of food preference, and high levels of repetitive self-grooming at juvenile and adult ages (McFarlane et al., 2008). Social deficits and repetitive behaviors in BTBR cohorts were consistent, robust, and well-replicated within our lab at NIMH (Yang et al., 2009Scattoni et al., 2011;Silverman et al., 2012Silverman et al., , 2015Babineau et al., 2013;Bales et al., 2014;Kazdoba et al., 2016b). Applying the constellation battery (Crawley, 2004), we did not detect sensory or motor dysfunctions in BTBR mice that could have produced artifacts that might have compromised interpretations of social deficits and repetitive behaviors (McFarlane et al., 2008;O'Connor et al., 2021). Many other labs have since confirmed a variety of social abnormalities and repetitive behaviors in the BTBR mice (Pobbe et al., 2011;Chadman, 2011;Gould et al., 2011;Schwartzer et al., 2013;Han et al., 2014;Amodeo et al., 2016Amodeo et al., , 2021Meyza and Blanchard, 2017;Yoshimura et al., 2017;Steinmetz et al., 2018;De Simone et al., 2020;O'Connor et al., 2021). BTBR became our gold standard for comparing phenotypes in other proposed mouse models of autism, and for evaluating new assays relevant to the diagnostic symptoms of autism. Robustness of its behavioral phenotypes facilitated the use of BTBR in therapeutic discovery.

Reciprocal social interactions
Excellent assays for direct social interaction in freely moving mice and rats have been elaborated by many labs including ours. Reciprocal social interactions between two unfamiliar mice placed together in a test chamber are scored for parameters including following, nose-toanogenital sniffing, nose-to-nose sniffing, and allogrooming (McFarlane et al., 2008;Dhamne et al., 2017;Rhine et al., 2019;Tai et al., 2020). Evaluating each parameter individually yields the most informative data, although composite scores, or employing a single parameter such as sniffing, have been used. The inbred strain, sex, and age of the partner mouse may influence social interactions between two freely moving mice. Multiple interactions within a group have been analyzed in a visible burrow system (Pobbe et al., 2011) and with machine learning algorithms (Segalin et al., 2021). Ultrasonic vocalizations emitted during social interactions (Scattoni et al., 2011;de Chaumont et al., 2021), which are particularly prominent during adult male-female interactions  and by males sniffing female urine (Roullet et al., 2011), are measured with a specialized ultrasonic microphone and software. Play behaviors are detectable in juvenile mice (Terranova and Laviola, 2005;Chadman et al., 2008). Juvenile rats display high levels of rough-and-tumble play (Thor and Holloway, 1984;Argue and McCarthy, 2015;Reppucci et al., 2018). Social recognition and social memory are evaluated by measuring social behaviors of a subject mouse interacting with an unfamiliar social partner, followed by presentation of a new social partner. Preference for social novelty, the 2 The JAX nomenclature for the BTBR inbred strain of mice is currently BTBR T+Itpr3tf/J, continuing as catalogue stock #002282. optional fourth phase of the 3-chambered task Nadler et al., 2004), which we originally designed to confirm olfactory abilities to detect social odor cues, is often used to measure social recognition memory.
One of the most fascinating challenges is to identify social motivation in rodents. Several labs are pioneering approaches in which a social partner is the sole reward. These include rats opening a cage door to release a cagemate (Ben-Ami Bartal et al., 2011), and conditioned place preference to a chamber previously containing a social partner (Panksepp and Lahvis, 2007;Pearson et al., 2012).
These complex, sensitive, quantitative assays require careful scoring by human raters, who must remain blind to genotype and/or treatment condition, or by automated videotracking and machine learning software.

Investigating the genetics of autism
In the early 2000s, genes were discovered that conveyed a high risk for autism spectrum disorder. Thomas Bourgeron and co-workers at the Institut Pasteur in Paris discovered point mutations strongly associated with autism in genes coding for the X-linked synaptic cell adhesion proteins Neuroligin 3 and Neuroligin 4 (Jamain et al., 2003), and in the postsynaptic scaffolding gene SHANK3 on human chromosome 22q13 (Durand et al., 2007). Chromosomal loci, including 15q11-13 which contains GABA receptor units (Christian et al., 2008) and 16p11.2 (Fernandez et al., 2010), were linked to cases of autism and related neurodevelopmental disorders. Co-morbidities of autism with single-gene disorders such as Fragile X syndrome (Wassink et al., 2001) and tuberous sclerosis (Jeste et al., 2016)  How do researchers determine the biological and behavioral consequences of a specific genetic mutation? One standard strategy is to generate a mouse model with essentially the same mutation. Molecular genetics labs began generating lines of mice which incorporated a risk gene for autism. Transgenic, knockout, knockin, and more recently, CRISPR-generated mutant mice, quickly became available to the research community through collaborations and repositories. The field developed important guidelines. Heterozygous mating was recognized as the optimal breeding strategy, to allow comparison of phenotypes in littermates including all three genotypes: wildtype + /+ , heterozygote + /-, and null mutant -/-, thereby minimizing the effects of different home cage environments (Brody and Geyer, 2004). Comparisons across developmental ages, of males and females, and with different background strains used to breed the mutation (Crawley et al., 1997), yielded valuable information about differential expression of phenotypes within a mutant line, depending on these factors.
As one of the few labs employing behavioral assays relevant to the diagnostic and associated symptoms of autism, we were the fortunate recipients of many lines of mice that incorporated a mutation identified in people with autism. Through our collaborations with molecular genetics teams at the National Institutes of Health and throughout the world, we identified strong, weak, and absence of autism-relevant social deficits across a wide range of mouse models of autism. Table 1 provides examples of the genetic models which our lab tested over the years, from 2007 to 2020. As referenced in Table 1, social abnormalities were robust as compared to wildtype littermates in mice with mutations in the genes coding for the early brain development protein Engrailed-2 and the postsynaptic density protein Shank3B. Significant differences were detected in some aspects of social behaviors in mice with a deletion at the 16p11.2 locus and with mutations in Shank3A, Cntnap2, Fmr1 on an FVB background, Pten at older ages, vasoactive intestinal peptide, and the serotonin transporter. In contrast, scores on social tasks did not differ from wildtype litters in mice with mutations in Chd8, Fmr1 on a B6 background, the β3 subunit of the GABA-A receptor, Grin2B, Neuroligin2, Neuroligin3, Neuroligin4, oxytocin, Shank1, and vasopressin receptor Avpr1b. High levels of repetitive self-grooming were robust and highly replicable primarily in Shank3B mutant mice. We recognized that other laboratories had reported social abnormalities and repetitive behaviors in some of the mutant lines listed in Table 1, for which our laboratory did not detect significant genotype differences. These divergences in findings led to our interest in pursuing phenotypic robustness.
Replicating findings in two independent cohorts within a laboratory is the best strategy to confirm the robustness of findings. Supported by Autism Speaks, we developed the Preclinical Autism Consortium for Therapeutics (PACT). The goal of PACT was to identify hypothesisdriven pharmacological interventions using well-replicated phenotypes of genetic mouse models of autism. Experimental design focused on testing two independently bred full cohorts of each mutant line, including Shank3B, Cntnap2, Pten, Fmr1, GABARβ3, and Grin2B. Each cohort, consisting of 10 male and 10 female mutants and 10 male and 10 female wildtype littermates, was tested on two or more corroborative assays in the behavioral domains of social, repetitive, cognitive, anxietyrelated, sensory and motor phenotypes. The identical set of behavioral assays, employed for each cohort, consisted of juvenile reciprocal social interactions, adult 3-chambered social approach, male-female reciprocal social interactions and ultrasonic vocalizations, spontaneous motor stereotypies, repetitive self-grooming, marble burying, elevated plus-maze, light↔dark transitions, open field locomotion, acoustic startle, prepulse inhibition of acoustic startle, olfactory habituation/ dishabituation, hot plate nociception, novel object recognition, contextual and/or cued fear conditioning, Morris water maze hidden platform acquisition, and Morris water maze hidden platform reversal. In parallel, Mustafa Sahin's laboratory at Boston Children's Hospital tested two cohorts of the same mutant lines on EEG measures of seizures and circadian sleep profiles. Portions of these datasets were published (Dhamne et al., 2017). However, the majority of these extensive experiments, conducted between 2012 and 2018, were not published, often because behavioral results from the two cohorts differed. Further, in several cases, genotype differences were significant on some parameters of an assay but not others, e.g. anogenital sniffing but not following or vocalizing during male-female interactions.
Our PACT experience may resonate with many other behavioral neuroscientists whose initial data do not replicate robustly. The scientific enterprise benefits greatly from transparent disclosure of the strengths and weaknesses of findings. Particularly when a phenotype is selected as a preclinical outcome measure to evaluate therapeutics, it is essential that the mouse model displays a robust, highly significant, easily replicable abnormality, against which the treatment response can be quantified unequivocally.

Translational applications
A decade after my IBNS presidency, we were ready for the next challenge: therapeutic discovery. Mouse lines harboring mutations syntenic to human autism risk genes and copy number variants had become readily available (Ey et al., 2011;Ellegood and Crawley, 2015;Kazdoba et al., 2016b;Ferhat et al., 2017;Zerbi et al., 2021). Rat and non-human primate models were emerging (Bauman et al., 2010;Veeraragavan et al., 2016;Harony-Nicolas et al., 2017;Zhao et al., 2017;Berg et al., 2018;Zhou et al., 2019;Möhrle et al., 2021). Phenotyping of mouse behaviors relevant to autism had become routine in many laboratories internationally (Crawley, 2007;Ey et al., 2011;Kazdoba et al., 2016b;Ferhat et al., 2017;Clipperton-Allen and Page, 2020;Caruso et al., 2020). Robust social deficits and repetitive self-grooming were well replicated in Shank3B and BTBR mice (McFarlane et al., 2008;Pobbe et al., 2010;Scattoni et al., 2011;Gould et al., 2011;Peça et al., 2011;Han et al., 2014;Kazdoba et al., 2016a;Meyza and Blanchard, 2017;Wang et al., 2017;Dhamne et al., 2017;Rhine et al., 2019. Employing BTBR and Shank3B, we began applying our research tools to preclinical translation. Many hypothesis-driven drug targets had emerged from genetic sequencing of people with autism, and from basic research into electrophysiological and neuroanatomical abnormalities in mouse models and people with autism. An insightful overarching neuropharmacological concept, originally proposed by John Rubenstein and colleagues at the University of California San Francisco, postulates that autism and related neurodevelopmental disorders are characterized by an excitatory/inhibitory imbalance (Rubenstein and Merzenich, 2003;Rubenstein, 2010;Sohal and Rubenstein, 2019). This hypothesis emerged from several lines of evidence, including the higher incidence of seizures in autism spectrum disorder, somatosensory cortex-mediated hypersensitivity to sensory stimuli, striatal-mediated repetitive behaviors, and the role of GABA in early development of neurons and brain circuitry (Ben-Ari et al., 1994;Li and Pozzo-Miller, 2020;Salmon et al., 2020;Besag and Vasey, 2021;He et al., 2021). Compelling receptor targets for reducing glutamatergic excitation and increasing GABAergic inhibition emerged from the excitatory/inhibitory imbalance concept.
We were fortunate to be able to test several compounds from both mechanistic classes, thanks to our outstanding lab members, funding agencies, and to the generous mouse behavioral testing space built on the University of California Davis medical campus in Sacramento during my recruit from NIMH to the UC Davis MIND Institute in 2012. Reducing glutamatergic neurotransmission with a negative allosteric modulator of the mGluR5 receptor, GRN-529, reduced repetitive self-grooming in three independent cohorts of BTBR mice, and reduced stereotyped vertical jumping in C58/J mice (Silverman et al., 2016). General locomotion was unchanged by the effective doses, an important control for potential sedating effects. Remarkably, GRN-529 also improved sociability in BTBR mice, on both the 3-chambered social approach test, and on social contact parameters of reciprocal social interactions . Direct administration of GABA agonists had similarly beneficial actions. The GABA-A agonist gaboxadol decreased repetitive self-grooming in three cohorts of BTBR mice (Rhine et al., 2019). The GABA-B agonist r-baclofen reduced repetitive self-grooming and marble burying in BTBR, reduced stereotyped vertical jumping in C58/J, and reversed BTBR social sniffing deficits in the 3-chambered sociability test, at doses that were not sedating (Silverman et al., 2015). Controls for these experiments included the less active enantiomer s-baclofen and C57BL/6J mice with normal sociability and low repetitive behaviors.
R-baclofen was further tested in 16p11.2 deletion mice. Deficits in reciprocal social interaction, object location memory, and contextual aversive learning in 16p11.2 heterozygous mice were reversed by r-baclofen treatment (Stoppel et al., 2018). Taken together with other reports of r-baclofen reversal of phenotypes in mouse models of intellectual disability (Henderson et al., 2012;Qin et al., 2015), these corroborating preclinical results supported the consideration of new clinical trials with Arbaclofen in people with autism spectrum disorder.
In 2018 the European Union initiated the Autism Innovative Medicine Studies-2-Trials (EU-AIMS-CT1) at seven academic institutions to evaluate Arbaclofen in people diagnosed with autism . EU-AIMS-CT1 employs the Socialization Domain of the Vineland Adaptive Behavior Scales to test social, communication, sensory, and daily living skills, along with electrophysiological biomarkers, in subjects with autism at ages 5-17 years. It will be interesting to learn the results of this large clinical trial after patient recruitment is completed in 2023.
Pharmacological targets of other classes showed promising results in preclinical models tested by several laboratories including ours. Ampakine compounds CX1837 and CX1739 improved social sniffing in BTBR mice during 3-chambered social approach . Replicating results from our PACT studies, new cohorts of Shank3B null mutant mice again displayed high repetitive self-grooming and lower levels of reciprocal social interaction, with a robustness and replicability appropriate for preclinical evaluation of interventions (Rhine et al., 2019). The TrkB receptor agonist 7,8-DHF improved spatial learning in Shank3B mice, and reversed social sniffing deficits in BTBR (Rhine et al., 2019).
It is important to note that treatments with compounds based on many hypothesized pharmacological targets failed to have a significant effect in a variety of mouse models of autism. In our lab, negative results were obtained for several hypothesis-driven pharmacological interventions in Shank3B mice (Rhine et al., 2019). In carefully conducted studies by postdoctoral fellow Jennifer Parrott (Rhine et al., 2019), D-cycloserine, a partial agonist of the glycine site on the glutamatergic NMDA receptor, at a low dose which did not induce hyperactivity, was ineffective at reversing social deficits or reducing self-grooming in Shank3B mutant mice (Rhine et al., 2019). CX546, one of the weaker positive allosteric modulators of the glutamatergic AMPA receptor, similarly did not rescue aberrant behaviors in Shank3B mice (Rhine et al., 2019). The TrkB agonist 7,8-DHF but did not improve cognitive and motor deficits in the Ube3a mouse model of Angelman syndrome (Schultz and Crawley, 2020). The mTOR inhibitor rapamycin did not ameliorate social deficits or repetitive behavior in BTBR mice (Rhine et al., 2019). One possible explanation for many of the negative pharmacological results in mouse models of autism is that successful interventions must begin at young ages. Administering the most promising of these compounds to mice at juvenile ages, or prenatally in utero, may be necessary to understand the critical developmental time window for intervention.

In conclusion
From our lab's earliest days in pioneering one of the first videotracking systems for rodent exploratory behaviors (Crawley et al., 1982), to inventing an automated conflict test for mouse anxiety-like behaviors (Crawley, 1981), to developing the automated 3-chambered sociability choice task relevant to autism (Yang et al., 2011), to phenotyping genetic mouse models, and culminating in preclinical evaluations of pharmacological interventions by our lab and many others (Kazdoba et al., 2016b, c), this author has been extremely fortunate to complete a linear body of work that led to potential clinical treatments for the core symptoms of autism. It is a rare and wonderful thing to have one's career dreams come true. Asking intriguing questions, designing the right experiments, employing optimized methods, and learning what the data reveal, are a scientist's fundamental rewards. Contributing to the scientific enterprise through mentoring students and postdocs, maintaining leadership roles in scientific societies, journal editorial boards, consultancies and scientific advisory committees, serving on study sections, reviewing manuscripts and grantsthese are our ways of giving back. Interacting with superlative collaborators and colleagues at conferences, especially at IBNS meetings, sharing our lab's findings in publications, grant applications, seminars, lectures, and posters, and most importantly, assisting future generations through interactive mentoring: it all adds up to 50 years of joys. My sincere gratitude goes out to the IBNS leadership and the Editors of Neuroscience and Biobehavioral Reviews for the honor of their invitation to publish this scientific memoir.

Author statement
All animal studies described in this manuscript were approved by the Institutional Animal Care and Use Committees of the National Institute of Mental Health Intramural Research Program and the University of California Davis, in compliance with the NIH Guide for the Care and Use of Laboratory Animals.

Conflict of interest
The author declares no financial or ethical conflicts of interest for any of the studies described in this manuscript.

Acknowledgements
Funding: Work by the Crawley laboratory described in this review article was generously supported by grants from the National Institutes of Health R01NS085709, U54HD079125, and P50HD103526, Autism Speaks Targeted Awards #8703 and #9868, the Simons Foundation Autism Research Institute grant #204340, the University of California Davis MIND Institute, and the National Institute of Mental Health Intramural Research Program. The author expresses deep appreciation to each of the outstanding students, postdoctoral fellows, technicians, and fantastic collaborators, whose hard work contributed to the experiments described. Special thanks go to my outstanding collaborators and co-workers Sheryl Moy, Jill Silverman, and Mu Yang; to PACT collaborators Mustafa Sahin, Dan Smith, and Rob Ring; and to lab members Mike Pride, Nycole Copping and Jill Silverman who conducted the PACT experiments. The author is indebted to the multitude of extraordinary colleagues at the UC Davis MIND Institute, who contributed intellectually to the strategies pursued by our translational research program. Consistent support from the NIMH Intramural Research Program for novel methods development enabled our inventions of impactful mouse behavioral assays for use by the neuroscience research community.