|Home | About | Journals | Submit | Contact Us | Français|
Autism is a heterogeneous neurodevelopmental disorder of unknown aetiology that affects 1 in 100–150 individuals. Diagnosis is based on three categories of behavioural criteria: abnormal social interactions, communication deficits and repetitive behaviours. Strong evidence for a genetic basis has prompted the development of mouse models with targeted mutations in candidate genes for autism. As the diagnostic criteria for autism are behavioural, phenotyping these mouse models requires behavioural assays with high relevance to each category of the diagnostic symptoms. Behavioural neuroscientists are generating a comprehensive set of assays for social interaction, communication and repetitive behaviours to test hypotheses about the causes of austism. Robust phenotypes in mouse models hold great promise as translational tools for discovering effective treatments for components of autism spectrum disorders.
Autism is a complex neurodevelopmental disorder with extraordinarily high heritability. Concordance between monozygotic twins reaches 90% for autism spectrum disorders (ASDs), as compared with less than 10% for dizygotic twins and siblings, and approximately 0.6–1.0% occurrence in the general population, along with a 4:1 male:female ratio1–5. The number of reported cases of autism has risen rapidly over the past decade, largely due to better diagnostic instruments and public awareness, although environmental causes and gene–environment interactions are also under investigation6,7. Considerable efforts are now focused on understanding the genetic causes of autism (see `Further Information') and using the genetic findings to select rational targets for effective treatments. Large international consortia are conducting linkage analyses to identify chromosomal loci and association and whole-genome scans to discover candidate genes. Rare variants in candidate genes have been reported, both de novo and familial, as well as copy number variants and epigenetic factors8–10. Strong evidence indicates that functionally interrelated mechanisms underlie the disorder. Synaptic development genes implicated in autism include neurexins, neuroligins, shanks, reelin, integrins, cadherins and contactins. However, each candidate gene mutation occurs in only a few individuals with autism1,3,9–17. Signalling, transcription, methylation and neurotrophic genes implicated in ASDs include phosphatase and tensin homologue (PTEN), MET, engrailed 2 (EN2), methyl-CpG-binding protein 2 (MECP2), fragile X mental retardation 1 (FMR1), tuberous sclerosis 2 (TSC2), calcium channel, voltage-dependent, L type, alpha 1C (CACNA1C), ubiquitin ligase E3A (UBE3A), Ca2+-dependent activator protein for secretion 2 (CADPS2) and brain-derived neurotrophic factor (BDNF)1,10,18–25. Neurotransmission genes, including the serotonin transporter, oxytocin and vasopressin receptors and GABA (γ-aminobutyric acid) receptor subunit β3, have been repeatedly associated with autism or highly implicated in social and affiliative behaviours impaired in autism1,26,27. Copy number variants include chromosomal duplications at 15q11–13 and 17p11.2 and deletions at 16p11.2 and 22q13.3 (Refs 8,10,15,28–32).
One compelling approach to test hypotheses about the many candidate genes for autism is to generate analogous mutations in the mouse genome and evaluate the mutant line for phenotypes analogous to the symptoms of autism33–35. Effective animal models should incorporate face validity (strong analogies to the endophenotypes of the human syndrome), construct validity (the same biological dysfunction that causes the human disease, such as a gene mutation or anatomical abnormality) and predictive validity (analogous response to treatments that prevent or reverse symptoms in the human disease)36,37. Mouse models have been generated with chromosomal deletions and with knockout and humanized knock-in mutations in many of the candidate genes detected in subsets of individuals with ASDs1,25,29,38–68. Mouse models with construct validity are being used to evaluate hypotheses about both genetic and environmental causes of autism, including single gene polymorphisms, copy number variants, epigenetic modifications, environmental toxins, prenatal infections, immune dysfunctions and mitochondrial abnormalities22,63,69–77. Hypotheses about multiple risk genes and gene–environment interactions are tested in mouse models that incorporate construct validity for two or more hypothesized causes, using the same behavioural assays as read-outs. Naturally occurring phenotypic differences among inbred mouse strains have been successfully utilized to identify model systems with high face validity and cost efficiency78–91. Phenotypes with strong face validity provide ideal translational tools for evidence-based treatment discovery22,63,73–77,88,92. TABLE 1 presents examples of genetic mouse models displaying behavioural phenotypes that are relevant to the three diagnostic criteria for autism25,29,38–41,43–68,78–87,89–91.
Designing mouse behavioural tasks that are relevant to human mental disorders presents a daunting challenge. Symptoms may be uniquely human and are often inherently variable. Autism diagnosis is currently based on purely behavioural criteria, as no consistent biological markers have yet been identified2,93–98. Until now, DSM-IV99, the diagnostic manual of the American Psychiatric Association, and ICD-10100, the diagnostic manual of the World Health Organization, have required the presence of core elements in three specific categories: abnormal reciprocal social interactions, which include reduced interest in peers and difficulty maintaining social interaction, and failure to use eye gaze and facial expressions to communicate efficiently; impaired communication, which generally presents as language delays, deficits in language comprehension and response to voices, stereotyped or literal use of words and phrases, poor pragmatics (knowing how and when to use language) and lack of prosody, resulting in monotone or exaggerated speech patterns; and repetitive behaviours, which include motor stereotypies, repetitive use of objects, compulsions and rituals, insistence on sameness, upset to change and unusual or very narrow restricted interests. Proposed DSM-V revisions may merge the first two criteria into a more general social-communication factor that includes lack of social reciprocity and deficits in nonverbal and verbal communication, beginning in early childhood.
Based on extensive advice generously contributed by autism clinical experts, behavioural neuroscientists are engaged in generating new mouse behavioural tasks and in refining existing paradigms from the behavioural neuroscience literature that maximize face validity to each of the core symptoms. Here, we review the tests that have proven most useful, along with the essential control measures, for the triad of diagnostic features of autism. Neuroanatomical, biochemical, electrophysiological and genetic similarities between mice and humans support the use of mouse models to further our understanding of biological mechanisms underlying the behavioural manifestations of autism. Similar responses to pharmacological treatments in mice and humans encourage the use of well-validated mouse models in the discovery of effective therapeutics for ASD.
Mus musculus is a social species that engages in high levels of reciprocal social interactions, communal nesting, sexual and parenting behaviours, territorial scent marking and aggressive behaviours101–105. A variety of social assays have been described in the behavioural neuroscience literature34,37. The examples described below were designed to maximize relevance to the types of social deficits that are specific to autism.
Fine-grained measures of interactions between pairs or groups of juvenile or adult mice placed together in standard cages or specialized arenas provide the most detailed insights into reciprocal social interactions. Parameters routinely evaluated include nose-to-nose sniffing, nose-to-anogenital sniffing, following, pushing past each other with physical contact, crawling over and under each other with physical contact, chasing, mounting and wrestling78,81,90,103,106. Parameters are scored from videotapes by investigators, using data sheets or event-recording software. Automated videotracking systems have also been used to score social interactions between two mice41,107. The experimental design, including the specific parameters scored, session duration, time of day, prior social isolation, environmental enrichment and pair composition by age, sex and strain, is optimized to meet the goals of the experiment. Repeated testing of the same mice is usually possible; this allows researchers to evaluate trajectories across the neurodevelopmental stages of pup, juvenile, young adult and older adult. FIG. 1 and Supplementary information S1 (movie) illustrate reciprocal social interactions in mice.
Simpler, automated measures of direct social approach offer more standardized, higher-throughput assays, although fewer details of reciprocal interactions are captured. We developed an automated three-chambered social approach task, which scores time spent in a side chamber with a novel mouse versus time spent in a side chamber with a non-social novel object, an inverted wire pencil cup44,81,90,108,109. Sociability is defined as the subject mice spending more time in the chamber containing the novel target mouse than in the chamber containing the inanimate novel object. The wire cup serves as the novel object on one side and as the container control for novel object plus novel mouse on the other side. With the target novel mouse contained, the social approach is initiated by the subject mouse only. The widely spaced wire bars of the container permit olfactory, visual, auditory and some tactile contact while preventing aggressive and sexual interactions, thus ensuring a pure measure of simple interest in approaching and remaining in physical proximity to another.
Our photocell-equipped apparatus uses infrared beams embedded in the partitions between compartments40,44,81,83,90,91,109–111. As the subject mouse moves between the three compartments, beam-breaks are recorded by the software and converted to time the mouse spends in each compartment and number of entries into each compartment. To provide a corroborative and more specific measure of social investigation during the test session, an observer scores time spent sniffing the novel mouse and time spent sniffing the novel object from session videotapes or in real time. The number of entries between compartments provides an independent measure of general exploratory locomotion. Mice can be tested more than once in this task — for example, at different ages to follow developmental trajectories. Videotracking software systems have been successfully used with the three-chambered apparatus, as well as observer scoring from videotapes43,48,77,107. FIG. 2 and Supplementary information S2 (movie) illustrate the automated social approach test in mice.
Another simple test of sociability uses a standard cage divided in half by a perforated partition made of clear plastic61,112 or wire104,105. The subject mouse is able to see, hear and smell the target mouse through the holes in the plastic or wire divider, but physical interactions are blocked. Time spent at the partition represents the amount of interest in the social partner. Different social partners can be sequentially placed in one compartment to evaluate social preference and social memory in the subject mouse.
Partner preference tests are used to evaluate components of social affiliation, social recognition and social memory. The choice between partners is measured by the amount of time spent by the subject mouse with each partner. Preference for social novelty is defined as the subject mouse spending more time in a chamber or in physical contact with a novel mouse than with a familiar mouse. Partners with different characteristics — for example, pair bonded mates, or familiar versus unfamiliar conspecifics — provide measures of social recognition. Partners can be present simultaneously44,81,82 or sequentially with time delays between presentations, to evaluate recognition memory113–116. Equipment used for social preference tasks include the three-chambered apparatus shown in FIG. 2, the partition test apparatus in which the subject mouse initiates more approaches and spends more time close to the partition adjacent to a novel mouse than to the partition adjacent to a familiar mouse56,84,87, a Y-maze113 and freely moving subject mice spending time with tethered target mice in three cages connected by tunnels116,117. Behavioural parameters during test sessions are scored from videotapes by investigators who are blind to the genotype or treatment condition, by software from photocell-equipped systems or by software from videotracking systems.
Interaction with a cagemate who has eaten a novel flavoured food will confer familiarity with the flavour, resulting in the subject mouse eating more of the now-familiar food than of a completely new food118–121. Familiarity is acquired when the observer mouse sniffs the breath, face and whiskers of the demonstrator mouse. Because face sniffing and close physical contact appear to contribute to the communication of flavour information, this task measures the tendency of the observer mice to obtain meaningful information through social interactions with the demonstrator.
How mice communicate is not yet well understood. Olfactory cues are of primary importance101,122. Vocalizations in the ultrasonic and sonic ranges, visual cues, gustatory and tactile modalities may also contribute to communication of information and to social bonding123–128. Several behavioural tasks are in routine use to evaluate the olfactory and auditory cues emitted by mice and the responses to these cues by other mice.
Mice deposit urinary steroidal pheromones that function as territorial scent marks and display high levels of interest in urinary scents from other mice. This is reflected in their tendency to explore the anogenital area of a novel mouse, investigate urinary scent marks in a cage, sniff a cotton swab soaked in urine and choose volatile urinary odours delivered by an olfactometer in an operant chamber104,105,129,130. The number of scent marks and countermarkings in close proximity to urinary olfactory cues may measure social motivation and/or olfactory communication89. Quantification methods include observer scoring of the number and duration of sniffing bouts from session videos and olfactory discrimination in operant tasks.
Mice tend to sniff a novel odour and then quickly habituate to its novelty40,120,131–134. Repeated presentation of a sequence of cotton swabs containing the same odour will result in the mouse spending less and less time sniffing the swab with each presentation (habituation), as measured by an investigator with a stopwatch. Subsequent introduction of a cotton swab saturated with a new odour will reinstate a high level of sniffing (dishabituation). The social odours on the cotton swabs are obtained from urine collected from another mouse or from swipes across the bottom of a cage of novel mice40,132. These social odours elicit considerably higher levels of sniffing than non-social odours, such as almond extract or banana flavouring132,134. The shapes of the habituation and dishabituation curves document the ability of mice to discriminate same and different non-social and social odours. The height of the peaks of the curves provide a measure of interest in the social and nonsocial odours. FIG. 3 and Supplementary information S3 (movie) illustrate olfactory habituation and dishabituation.
Complex vocalizations in the ultrasonic range are emitted by mice in social situations, including pups separated from the dam and nest, juvenile interactions, resident females in a resident–intruder task and males responding to female urinary pheromones56,84,87,123–128. Sensitive ultrasonic microphones, headphones and advanced software for detailed analyses of sonograms have revealed discrete categories of calls in mice56,84,87,124,126,128. Supplementary information S4 (audio) provides examples of mouse vocalizations in a social setting.
However, the intentional communicative nature of mouse vocalizations remains to be determined. Further research will be needed to understand which social situations elicit calls and how consistent those calls are during each specific social situation. Developing assays that are sensitive enough to detect subtleties of abnormal vocalizations in mice will be a challenge. Communication deficits in autism include developmental delays in the comprehension and use of expressive language, failure to respond to speech during early ages, the absence of rhythm and melodic prosody, literal use and interpretation of language, and the tendency to speak in monologues instead of interactively2,93,97,98. Although there is not a consistent vocalization endophenotype for autism during the first 2 years of life in humans (which would correspond to the pup stage in mice), ongoing studies are evaluating the relevance of juvenile and adult vocalizations in mouse models to the specific types of communication abnormalities in autism29,84,89,135.
Mice exhibit spontaneous motor stereotypies, including circling, jumping, backflips and self-grooming86,136–138. Scoring of stereotypies is conducted most reliably by an investigator observing video taped sessions or in real time. The observer records each bout of the stereotyped behaviour during a defined sampling period, using a scoresheet or an event recorder.
Sequences of behaviours may appear as normal patterns but persist for unusually long periods of time. BTBR T+tf/j mice (referred to here as BTBR) engage in extremely long episodes of repetitive self-grooming. BTBR mice may self-groom for up to 2 minutes, whereas bouts of self-grooming in standard control strains such as C57BL/6j(86) are much shorter, generally lasting between 5 and 10 seconds81,90,91,111. Repetitive behaviours are generally scored — from videotapes or in real time — by an observer with a stopwatch. Marble burying, a repetitive digging behaviour, is scored by counting the remaining unburied marbles139. FIG. 4 and Supplementary information S5 (movie) illustrate repetitive self-grooming in BTBR T+tf/j mice.
Perseverative behaviours are relatively common in mice. Reversal learning tasks measure the flexibility of the mouse to switch from an established habit to a new habit. A spatial habit is first established, for example, reinforcing entries into the left arm of a T-maze or by locating the hidden escape platform in one quadrant of a Morris water maze. The re inforcer is then moved to a new location − for example, the food reward is moved to the right arm of the T-maze or the hidden platform is moved to a different quadrant of the water maze pool44,82,140–142. A mouse model of autism is predicted to perform well on the initial acquisition but to fail on reversal owing to either increased perseveration or specific impairments in reversal learning. It may also be possible to model `upset to change' in mice. Olfactory disruptors introduced during a selective attention operant task produced a generalized disruption in performance143, illustrating an interesting response that could reflect an upset to a change.
Methods to measure restricted interests in rodents are under development. One approach capitalizes on the tendency of mice to explore all aspects of a novel environment, including exploratory locomotion in a novel open field, sniffing of novel objects and nose poking into holes in the wall or floor144. Perseverative exploration of only one of the available objects or holes, rather than the normal strategy of exploring all novel objects or holes, may be analogous to restricted interests in human subjects with autism.
Designing mouse behavioural assays with high relevance to the diagnostic symptoms of autism presents a substantial challenge for capturing reasonable face validity. Several symptoms of autism, such as the literal use of language and difficulties in interpreting irony or sarcasm, are unlikely to be successfully modelled in mice.
`Theory of mind', the ability of one person to intuit what another person is feeling and thinking, may not be innate to the mouse repertoire. However, two recent reports support the possibility that mice display elements of empathy. Subject mice show greater responsiveness to a painful experience after observing cagemates who have experienced a painful stimulus80,145. Subtleties of language are also unlikely to be innate to mice. However, the complexity of mouse ultrasonic vocalization patterns may contain considerable communicative information135. Quantitative measures of the reward value of social interactions are not yet in place for mice or for autistic individuals. A starting point for measuring the reward value of social interactions in mice may be the literature on rat operant chambers that measure the number of lever presses for parental access to pups146, rat operant chambers that measure the number of lever presses for adult access to sexual partners147 and a mouse conditioned place preference task for social odours84,85.
Executive functions requiring the simultaneous integration of large amounts of complex social and non-social information may be localized in the prefrontal cortex, a brain region that is not well-developed in mice. Complex cognitive abilities are evaluated in mice using cognitive tests that depend on medial frontal cortex connections, such as the intradimensional and extra-dimensional attentional set-shift task148,149. Eye gaze is difficult to track in mice, as the pupil is hard to distinguish. However, elements of eye gaze, joint attention and attentional focus might be modelled in mice using sustained attention tasks, such as the five-choice serial reaction time test150 with auditory, visual or olfactory distracters151 Some features of autism, such as the 4:1 male:female ratio and regression of social communication after one year of age, have yet to be identified in a mouse model.
Associated symptoms, which occur in subsets of autistic individuals, include seizures, anxiety, mental retardation, hyperreactivity and hyporeactivity to sensory stimuli, sleep disruption and gastrointestinal distress152–154. Analogous pheno-types would be useful additions to a mouse model that displays robust social deficits. Standardized mouse assays are available to measure seizures (observer scoring, electroencephalography (EEG) recordings), anxiety-related behaviours (elevated plus-maze, light–dark exploration), cognitive abilities (spatial learning (Morris water maze), contextual and cued fear conditioned emotional learning, shock avoidance, object recognition and operant discrimination tasks, among others), hyper-sensitivity to sensory stimuli (acoustic startle, air puff startle, hot plate) and sleep (EEG recordings, circadian running wheels, home cage monitoring systems)37,155. Standard assays of mouse developmental milestones from birth through weaning are useful for identifying phenotypes that are relevant to associated symptoms of autism during early development37,44,56,87,98,156.
A fundamental issue resides in potential artefacts caused by mouse phenotypes which are relevant to an associated symptom of autism, but which confound the interpretation of a mouse phenotype with higher relevance to a specific core symptom of autism. For instance, mice with anxiety-like traits will engage in low exploratory activity, resulting in minimal entries into the side chambers in the three-chambered sociability task, thus rendering social approach data meaningless. Identifying phenotypes relevant to associated symptoms, versus artefacts that confound the interpretation of tests relevant to diagnostic symptoms, poses an internal paradox to be parsed on a case-by-case basis.
Severe physical disabilities will cause false positives in many of the behavioural tasks described above34–37,157–159. For example, olfactory deficits will inhibit performance on social approach, social recognition, olfactory discrimination and scent marking tests. Motor dysfunctions will prevent a mouse from active exploration of test environments that require locomotion, including social chambers, T-mazes and holeboards. To rule out artefacts, each new line of mutant mice has to be evaluated on a series of measures of general health, body weight, neurological reflexes, home cage behaviours, open-field activity, rotarod performance, visual forepaw placing, acoustic startle and pain sensitivity36,37. Given the fundamental role of olfaction in mouse social behaviours, social and non-social olfactory abilities are routinely evaluated with multiple tests, including latency to locate buried food, olfactory habituation/dishabituation to non-social and social odours, and preference for social novelty44,132.
The number of mice per group (n) for behavioural experiments is considerably larger than the number of mice needed for most biological assays. Larger numbers of mice are usually necessary to compensate for the unavoidable variability in environmental factors that influence mouse behaviours, such as handling by animal caretakers, vivarium conditions, early parental care and home cage dominance hierarchies. n = 10–20 per genotype and per sex is often required to achieve sufficient statistical power when performing, for example, a Two-Way Repeated Measures Analysis of Variance (ANOVA). When a significant ANOVA is detected, a posthoc test, such as Newman–Keuls, Tukey's, Sheffe or Bonferroni–Dunn, is used to compare group means for specific differences between genotypes and/or treatment effects. If breeding, housing or testing capacity is limited, small subgroups can be generated to accumulate the needed numbers, as long as each genotype is represented in each subgroup on each day of behavioural testing, and the data from wild-type littermate controls do not differ across subgroups.
The strength of a phenotype increases when the initial findings are replicated in a second and third cohort of littermates of all genotypes. Phenotypic replications that are produced by different investigators in the same laboratory, by different investigators in different laboratories, from independent lines of mice with the same mutation generated by different laboratories using different DNA constructs, or by breeding into different genetic backgrounds, clearly strengthen the conclusiveness of phenotypes. Minor methological differences and the influence of environmental factors become trivial when findings are well replicated across these different contexts.
The entire set of behavioural tasks described above can usually be conducted in the same set of mice, as long as reasonable attention is paid to the sequence in which the tests are conducted. For example, the most stressful tasks should be performed at the end of the sequence, and with sufficient intervals between testing days159. Occasionally a task cannot be conducted owing to species issues, such as body size or activity levels, physical or procedural artefacts caused by background genes, unexpected consequences of the targeted gene mutation, or side effects of a treatment. Most of the behavioural assays can be successfully applied to normal and mutant lines of mice and rats, inbred strains of mice and rats, and some other rodent species.
The field is now poised to pursue a comprehensive characterization of behavioural traits that are relevant to the symptoms of autism in each of the candidate gene mutant lines of mice. Positive findings obtained from a mutant mouse model will reinforce interest in pursuing a gene in molecular and clinical studies. For example, a rationale for developing metabotropic glutamate receptor 5 (mGluR5) antagonists as therapeutics has been provided by studies showing that a genetic reduction of mGluR5 reversed some of the symptoms in Fmr1 mouse models of fragile X syndrome, in addition to studies showing that an mGluR5 antagonist treatment reversed Fmr1 phenotypes and repetitive self-grooming in BTBR mice74,88,160–162 (TABLE 2). Neuroanatomical, electrophysiological, neurochemical and other phenotypic characterizations can be used to test emerging hypotheses about the biological mechanisms responsible for the brain dysfunctions underlying neurodevelopmental disorders11,12,43,48,63,75,77,138,140,163. Identical phenotyping strategies can be applied to investigate putative environmental causes for autism. Interesting behavioural abnormalities have emerged from rodent models of prenatal exposure to valproic acid164, prenatal influenza infection165, immune dysfunctions71 and exposure to neurotoxins70.
Treatments for the symptoms of autism are being intensively sought163,166. Early behavioural interventions, such as applied behaviour analysis, pivotal response training, parent training, behaviour management and social skills training in groups (all of which are primarily provided through special educational programmes), are currently the only treatments that significantly improve the first and second core symptoms (unusual reciprocal social interactions and communication deficits)156,167. Medications can have significant effects on associated symptoms such as hyperactivity or mood, but have not been shown to directly affect the core features of autism. Behavioural interventions have ameliorated symptoms in several mouse models of neurodevelopmental improved locomotion and rotarod performance in Rett syndrome Mecp2 mutant mice and reduced hyperactivity in fragile X syndrome Fmr1 knockout mice168–171. social peer enrichment improved social interactions in low-sociability BTBR mice reared as juveniles with social B6 cagemates172. Preclinical successes have been reported for genetic rescues and pharmacological reversals of aberrant phenotypes in mouse models of ASD. Successful drug candidates include mGluR5 antagonists, rapamycin, BDNF and oxytocin63,73–77,88,92,114,160–162,173–178 (TABLE 2). As knowledge grows about the genetic and environmental factors that confer susceptibility for autism, mouse models with construct validity and phenotypes that are relevant to core symptoms will offer strong translational systems for discovering rational therapeutics.
M. L. Scattoni, Istituto Superiore di Sanitá, Rome, Italy, generously contributed the supplementary audio file giving examples of mouse vocalizations. We thank A. Katz, Laboratory of Behavioural Neuroscience, National Institute of Mental Health (NIMH), Bethesda, Maryland, USA, for the outstanding editing of the supplementary movies. J.N.C. is supported by the NIMH Intramural Research Program MH02179. C.L. is supported by NIMH grants R01MH81873, 1RC1MH089721 and 1R01MH089390.
Competing interests statement The authors declare no competing financial interests.