These sensitive periods of elevated plasticity are times of opportunity but also of great vulnerability for the developing brain. As many have experienced, it is easier to learn a new language, musical instrument, or sport as a child rather than in adulthood. On the other hand, early disruption of proper sensory or social experiences will result in miswired circuits that will respond suboptimally to normal experiences in the future. The devastating effects of early deprivation are scientifically documented [3, 4]. Studies of socially and emotionally deprived children raised in Romanian orphanages have demonstrated that the neglected children exhibit severe developmental delay, mental retardation, and neuropsychiatric symptoms [4]. Orphans need to be placed with caring foster families away from orphanages before two years of age in order to develop cognitive, social, and intellectual skills. Neglected children are not able to recover normal function even if they are later placed in similar foster homes. Comparable effects are seen for the development of the primary senses as well. Conductive hearing loss often associated with childhood ear infections can produce long-lasting deficits in auditory perceptual acuity if not treated before the age of seven [5–7]. Similarly, if a child's binocular vision is compromised by strabismus or cataract and is not corrected before the age of eight, loss of acuity in that eye, or amblyopia, is permanent and irreversible [8, 9]. If corrected promptly, restoration of normal binocular vision is possible.

Why the brain is able to recover function early in life, but loses this ability with maturity? What are the mechanisms underlying experience-dependent circuit refinement in early development? Can we recreate the plasticity of the immature brain later in life and eventually recover proper function? It turns out that a very precise balance of cortical excitatory and inhibitory (E/I) neurotransmission is required for critical period plasticity [10, 11]. Studies in the rodent visual system have shown that, in particular, the level of the inhibitory neurotransmitter GABA and the maturation of specific inhibitory circuits are crucial [11, 12]. Since critical periods are so tightly regulated, this makes them vulnerable if the E/I balance is tipped in either direction without compensatory homeostatic correction. Recent research has indicated that neurodevelopmental disorders like autism may result from disruption of this balance early in life. This could be due to a combination of genetic or environmental insults that compromise excitatory or inhibitory components at the genetic, molecular, synaptic, or circuit level. Depending on the location and severity of imbalance, a spectrum of phenotypes could result, as is true for autism. Thus, it is attractive to hypothesize that autism may result from disruption of the expression and/or timing of critical periods across brain regions.

Autism is generally diagnosed within the first three years of life, during this time of intense experience-dependent circuit refinement. The diagnostic behavioral symptoms of autism are abnormal socialization and communication, and repetitive behaviors [13]. Many studies have focused on addressing how the autistic brain perceives relevant information, like face-processing, language, and theory of mind. These data have been valuable in beginning to understand how the higher-order processing centers of the brain differ in autism. However, it is important to realize that these areas rely on integration of inputs from lower cortical regions, building off a reliable and accurate representation of the world generated by primary sensory areas. Critical period disruption resulting in a slight degradation in the quality of any or all of these senses would compromise the ability to successfully execute behaviors relying on this information, creating severe deficits. Indeed, sensory deficits have been reported in autistic individuals, indicating possible improper primary sensory perception [14].

Autism is called a “spectrum disorder” because of the extraordinary heterogeneity of intellectual ability, associated symptoms, and possible etiology. Though there is clearly a genetic basis to autism, the majority of cases have unknown causes [15, 16] Autism is comorbid with a number of other diseases, including Rett, Fragile X, and Angelman Syndrome. These diseases have known genetic causes and have been well modeled via genetic modification of animals, thereby providing valuable tools to dissect the molecular changes underlying autism. Despite these advances, there is still no cure [17, 18].

A common feature of autistic individuals is atypical behavioral responses to sensory stimulation and reports of hyper- or hyposensitivity to sensory stimulation in multiple domains [14, 21]. There are many accounts of disruption of primary sensory processing in autism [22–25], and there is a growing body of evidence that tests these reports in a controlled laboratory setting. A recent meta-analysis of 14 parent-report studies on sensory-modulation suggests that autistic individuals exhibit significantly more sensory symptoms than control groups, particularly between the ages of six and nine [26]. Interestingly, most studies have concluded that several sensory processing are more commonly disrupted in autism than in other developmental disorders; these symptoms lessen with age; their severity correlates with the severity of social impairment [27]. We will touch on a few examples of altered sensory processing in the auditory, somatosensory, visual, and multisensory integration areas from the human autism literature.

Many studies of sensory phenotypes in autism have focused on the auditory system because of the language deficits characteristically observed in patients. There do appear to be lower-level cortical auditory processing abnormalities as measured by electroencephalograms (EEG) and magnoenetcephalography (MEG) in multiple studies, but the nature of these differences is variable and depends on the specifics of the individual studies [14]. For example, while some studies have found that autistic subjects have increased latency in cortical response to tones [28, 29], others observed a decreased latency in cortical response [30–32]. These contrasting results may reflect the wide spectrum of autism phenotypes, the limited number of tested subjects, their age, or different experimental paradigms used.

Abnormal somatosensory experiences are commonly reported in autistic individuals [33]. One psychophysical study by Tommerdahl et al. [34] tested the ability to spatially discriminate two vibrotactile stimuli applied to the skin of the hand in a small group of autistic subjects. After a priming stimulus, subsequent spatial discrimination in that same area of skin improved for controls but not for autistic subjects. The authors suggest that this may reflect a deficit in cortical inhibition of neighboring minicolumns, though this claim was not directly tested. Psychophysical studies rely on the behavioral report of the subject, and therefore may be complicated by behavioral impairments in autistic subjects. Several somatosensory studies have measured brain activity instead as a means to evaluate sensory processing. Miyazaki et al. [35] found abnormal short-latency somatosensory evoked potentials (S-SEPs) in response to median nerve stimulation in about half of the autistic patients they tested. However, it must be noted that there were no concurrent controls tested in this study, and instead autistic S-SEPs were compared to S-SEPs from controls in a previous study. Another group used MEG to map the cortical representation of the hand and face regions of high-functioning autistic and control subjects [36]. Brain activity was recorded in response to physical stimulation of the skin. Interestingly, the autistic subjects had a spatially distorted cortical representation of the hand and face compared with controls. Overall, abnormal somatosensory processing may play a large role in the avoidance of affective contact that contributes to the social and communication abnormalities in autism [33].

Due to the social phenotype of autism, one of the better-studied visual impairments of autism is face processing [27]. However, an interesting hypothesis is that lower-order visual deficits would consequently impair higher-order visual processing of faces. A study by Vlamings et al. [37] recorded VEPs in autistic and control children while they looked at two types of stimuli—simple horizontal gratings or faces with neutral or fearful expressions. The gratings or faces were composed of either high or low spatial frequency lines. Autistic children, in contrast to controls, had an enhanced VEP response to high spatial frequencies and performed better at facial expression categorization when the faces were high-pass filtered. Nonautistic children generally use low spatial frequency information to categorize the emotions of facial expressions. This difference seen in autistic children is in agreement with previous findings that autistic perception is more detail-oriented [38–41]. This study suggests that abnormal primary visual processing could also contribute to social and communicative deficits in autism.

In addition to atypical unisensory experiences in autism, growing evidence points towards abnormalities in multisensory processing, which is the integration of information from different senses into one perceptual experience [14]. Deficits in multisensory integration (MSI) fit with a popular theory of the autistic brain, in which there is excessive local connectivity within one brain region but long-range hyperconnectivity between brain regions [19, 42, 43]. As for unisensory modality processing, MSI seems to be disrupted in a variety of ways depending on the study, including enhanced, decreased, or altered in some fashion. A recent study used high-density electrical mapping of the cortex with EEG to measure MSI in response to audio-somatosensory stimuli [44]. Vibrotactile stimulation and tones were presented to the passive subject either separately or in combination, and differences in event-related potentials (ERPs) between uni- and multimodal stimulation were measured. Overall, the autism group showed less MSI than controls. In contrast, a psychophysical study investigated audio-visual integration, as this has direct relevance to speech perception, and found an extended temporal window for MSI in the autism group [45].

Another way to evaluate audio-visual integration is with the McGurk effect [46]. In the McGurk effect paradigm, an individual hears the sound of one phoneme (/ba/) while watching a muted video of a person saying another phoneme (/ga/). Due to the multimodal quality of speech perception, the sound of the voice combines with the sight of the lips moving, and the individual reports hearing a third intermediate phoneme (e.g., /da/), the perceptual product of normal multimodal integration. This phenomenon was originally reported to occur less frequently for autistic individuals [47]. More recent studies confirmed some level of disruption in the McGurk paradigms in autism subjects mainly affecting the ability to read lips [48, 49] and the comprehension of speech in the presence of background noise compared to control subjects [49]. Audiovisual speech integration is already present in infants as young as 2 months old [50] and contributes to phonetic learning [51] and language development [52]. Combined deficits in audiovisual processing may then contribute to delays in language acquisition and speech comprehension during social interactions or school settings.

Abnormal visual input to one eye during infancy results in permanent loss of visual acuity, amblyopia (Greek for dull vision), if not corrected during childhood. If perturbation of vision occurs in adulthood, the visual impairments are significantly milder or absent [64]. This observation in humans inspired the development of a simple laboratory paradigm to test the existence of a critical period in animal models. David Hubel and Torsten Wiesel began investigating OD plasticity in a series of Nobel Prize winning experiments in the 1960s [65, 66].

They found that the closure of one eye (monocular deprivation) of a kitten during a specific time window early in postnatal life results in an experience-dependent loss of visual acuity in the deprived eye despite no physical damage to the eye itself [67]. This is due to a competitive invasion by the nondeprived eye into cortical territory previously responsive to the deprived eye. A functional loss of responsiveness to the deprived eye and an increase of responsiveness to the open eye are followed first by pruning and then regrowth of dendritic spines on cortical pyramidal neurons [68, 69]. Further structural reorganization takes place in the form of shrinking thalamocortical projections (OD columns) serving the deprived eye and expansion of those serving the open eye [70].

The ocular dominance critical period is present in all mammals tested so far, from humans to mice, and the duration of plasticity is in direct correlation to lifespan and brain weight [71]. The identification of rodents as models of amblyopia has made possible a fine dissection of the mechanisms underlying critical period expression. In particular, by taking advantage of genetically modified mouse models, a specific inhibitory circuit has been identified that controls the timing of OD plasticity [11]. Fine manipulation of inhibitory transmission is difficult in vivo, because enhancing inhibition silences the brain, while reducing inhibition easily induces epilepsy. With the generation of a mouse lacking only one of the two enzymes that synthesizes GABA (GAD65), researchers were able to titrate down the level of inhibition and test its role in the OD critical period [12]. Strikingly, the visual cortex of GAD65 knockout mice remains in an immature, precritical period state throughout life. At any age, functionally enhancing GABAergic transmission with benzodiazepine treatment triggers the opening of a normal-length critical period [72]. Historically, inhibitory neurotransmission was believed to develop postnatally to progressively restrict plasticity, but these key experiments proved GABA to actually be necessary for a normal OD critical period, prompting further investigation into the role of inhibition in brain plasticity.

Inhibitory interneurons account for nearly 20% of cortical neurons and exhibit heterogeneous morphological and physiological characteristics [73]. Included in this large variety of inhibitory interneurons is a specific subset of GABAergic neurons that expresses the calcium-binding protein parvalbumin. Fast-spiking parvalbumin-positive basket cells (PV-cells) regulate critical period timing and plasticity [11, 74]. PV-cells develop with a late postnatal time course in anticipation of critical period onset across brain regions [75, 76]. In the visual cortex, PV-cells mature in an experience-dependent manner, and dark-rearing delays their maturation as well as critical period expression [77, 78]. On the other hand, overexpression of brain-derived neurotrophic factor (BDNF) promotes the maturation of PV-cells and speeds up the onset of the OD critical period [77, 79]. Moreover, Di Cristo et al. [80] have shown that premature cortical removal of polysialic acid (PSA), a carbohydrate polymer presented by the neural cell adhesion molecule (NCAM), results in a precocious maturation of perisomatic innervation of pyramidal cells by PV-cells, enhanced inhibitory synaptic transmission, and an earlier onset of OD plasticity. Recent results indicate that PV-cell maturation is surprisingly regulated by the Otx2 homeoprotein, an essential morphogen for embryonic head formation [78]. Otx2 is stimulated by visual experience to pass from the retina to visual cortex and selectively into PV-cells, thereby promoting their maturation and consequently activating OD critical period onset in the visual cortex.

PV-cells receive direct thalamic input and also connect to each other in large networks across brain regions by chemical synapses and gap junctions [81, 82]. Moreover, PV-cells form numerous synapses onto the somata of pyramidal cells, which in turn enrich these sites with GABA_A receptors containing the α1-subunit [11, 70, 74, 78, 83]. This makes PV-cells perfectly situated to detect changes in sensory input, to regulate the spiking of excitatory pyramidal cells, and to synchronize brain regions [84–86]. Manipulations that disrupt this specific circuit will disrupt the OD critical period [87]. Recent studies have made much progress regarding the origin and fate determination of cortical interneurons [88]. In particular, progenitors of PV-cells derive from the medial ganglionic eminence with a relatively late birth date, and their differentiation and migration into specific cortical layers can be regulated by homeoproteins like Lhx6 [88, 89], or excitatory projection neurons [90]. Although the closure of the OD critical period is tightly regulated, transplanting immature GABAergic cells into the visual cortex can reallow OD plasticity later in life [91]. This second sensitive period only emerges once the newly transplanted GABAergic cells reach a critical maturation stage of connectivity. This further supports a key role of inhibition in the timing of experience-dependent circuit refinement.

Once the critical period is initiated, plasticity is only possible for a set length of time, and then the critical period closes [92]. Several functional and structural brakes on plasticity have been identified in recent years [93]. Disruption of these brakes in the adult brain allows critical periods to reopen and neuronal circuits to be reshaped. In the case of OD plasticity, this means that monocular deprivation in adulthood would induce a shift in responsiveness to the nondeprived eye and cause a loss of acuity in the deprived hemisphere. Interestingly these brakes share a common theme of regulating E/I balance, and particularly the GABAergic system. Locally reducing inhibition in adulthood restores plasticity in visual cortical circuits [94, 95]. Treatment with the antidepressant drug fluoxetine also reopens plasticity, potentially by altering inhibitory transmission and increasing BDNF levels [96, 97]. Finally, knocking out lynx1, an endogenous prototoxin that promotes desensitization of the nicotinic acetylcholine receptor (nAchR), extends the critical period into adulthood [98]. Lynx1 likely modulates E/I balance because treatment with diazepam in lynx1 knockout mice abolishes adult plasticity by restoring this balance to normal adult levels.

Structural factors also restrict remodeling of circuits with the closure of critical periods. For example, PV-cells become increasingly enwrapped in perineuronal nets (PNN) of extracellular matrix with the progression of the critical period, and enzymatic removal of these nets or disruption of their formation restores plasticity in adulthood [78, 99, 100]. In addition, the maturation of myelination throughout the layers of the visual cortex, as measured by myelin basic protein (MBP) levels, increases as the critical period closes [101]. Myelin signaling through Nogo receptors (NgRs) limits plasticity in adulthood, and genetic or pharmacological disruption of this receptor allows persistent OD plasticity later in life [101, 102].

In addition to reopening plasticity, disruption of these brakes also may allow recovery from early deprivation-induced loss of function, like amblyopia. In order to test this, animals are subjected to long-term monocular deprivation spanning the critical period. This results in permanent amblyopia, even if the deprived eye is reopened in adulthood and allowed to receive visual input. Significantly, some of the manipulations described above allow recovery of acuity, including enzymatic degradation of PNNs [103], disruption of NgR signaling [102], administration of fluoxetine [96], and enhanced cholinergic signaling by lynx1 knockdown or treatment with acetylcholinesterase inhibitors [98]. Treatment with drugs like fluoxetine and acetylcholinesterase inhibitors offers particularly promising therapeutic potential because they are already FDA-approved for human use. As the mechanisms behind the closure of critical periods are explored, more light will be shed on potential interventions that could reopen plasticity or reset abnormal critical periods by restoring the brain to a more juvenile-like state.

How generally might these same mechanisms apply to critical periods in other parts of the brain? Interestingly, recent evidence has shown that similar mechanisms may exist in other brain regions. For example, the maturation of PV-cells in the barrel cortex peaks during the critical period for whisker tuning [75]. Furthermore, whisker trimming exclusively during this critical period in mice results in decreased PV expression and reduced inhibitory transmission in vitro [104]. In the zebra finch, brain regions dedicated to singing exhibit progressive PNN formation around PV-cells with a time course that parallels the critical period [105]. The maturity of the song correlates with the percentage of PV-cells that are enwrapped in PNNs, and this can be manipulated with experience by altering exposure to tutor song. In rodent auditory cortex, spectrally limited noise exposure prevents the closure of the critical period for regions of auditory cortex that selectively respond to those interrupted frequencies, and PV-cell number is also reduced in those regions [106]. In the rodent, conditioned fear can be eliminated during early life but is protected from erasure in adulthood [57]. A developmental progression of PNN formation around PV-cells coincides with this switch and enzymatic degradation of PNNs allows juvenile-like fear extinction in adulthood [58, 59], similar to the reopening of OD plasticity in the adult visual cortex [99].

Though there is clearly a genetic basis to nonsyndromic autism, there is no single gene or family of genes that is exclusively implicated. Rather, it is likely the inheritance of several risk factors, perhaps in combination with an environmental or epigenetic trigger, that ultimately cause autism. This would fit with the E/I imbalance theory, where the presence of one mutation that increases excitation may not alone be sufficient to disrupt the balance whereas coinheritance of this mutation with another that decreases inhibition would be enough to prevent homeostatic correction and result in a dramatic E/I imbalance [42]. There is substantial evidence of altered inhibition in autistic patients, suggesting a lack of homeostatic correction and a resulting E/I imbalance.

In support of inhibitory disruption, studies on autistic patients demonstrate broad alterations in the GABAergic system. The levels of GABA measured in the plasma of autistic children may be elevated [122, 123] while the enzymes that synthesize GABA (GAD65 and GAD67) are significantly reduced (~50%) in postmortem autistic parietal cortex and cerebellum [124, 125]. The relevance of GABA levels measured in the plasma to the actual levels in the brain is unfortunately unclear. Multiple studies have also found both GABA_A and GABA_B receptor disruption in autistic brains [126–129]. Altered modulation of GABA_A receptors in the presence of GABA was suggested by a study that detected reduced radioactively labeled benzodiazepine binding to hippocampal GABA_A receptors [130]. On a more structural level, autistic neocortical minicolumn number was increased and width decreased, indicating abnormal cortical organization regulated by inhibitory circuitry [131, 132]. In addition, cortical projection neurons exhibited increased dendritic spine densities, providing structural evidence for changes in connectivity in autism [133]. These combined data support the notion of changes in E/I balance at the level of cells, synapses, and circuits in autism.

Over the years, mouse models of autism have been generated based on rare mutations identified in autism patients, environmental insults associated with autism, or mutations known to cause diseases that are comorbid with autism (reverse genetic approach). Existing mouse strains have also been screened for behaviors relevant to autism (forward genetic approach). In this section, several different mouse models of autism are reviewed, with their common disruption of E/I balance receiving special attention.

Dölen et al. [207] tested OD plasticity in an Fmr1 knockout mouse model of Fragile X syndrome. They recorded chronic VEPs before and after brief (3-day) MD during the canonical critical period. Knockout mice showed significant potentiation of the ipsilateral open eye response but no depression of contralateral deprived eye response as is seen in wildtype mice. The potentiation of the open eye is usually only seen after longer periods of MD, secondary to the depression of the closed eye. Therefore, the knockout mice do show an OD shift during the critical period, but the nature of this shift is unusual. Furthermore, visual acuity before and after MD was not measured, so it is unknown if the deprived eye ever became amblyopic and if critical period plasticity really took place. In addition excitatory thalamocortical synapses in somatosensory cortex during the perinatal critical period in Fmr1 knockout mice. FMRP ablation resulted in dysregulation of glutamatergic signaling maturation. The fraction of silent synapses persisting to later developmental times was increased; there was a temporal delay in the window for synaptic plasticity, while other forms of developmental plasticity were not altered in Fmr1 knockout mice. indicating that FMRP is required for the normal developmental progression of synaptic maturation impacting the timing of the critical period for layer IV synaptic plasticity [208].

Tropea et al. [144] tested OD plasticity in adult MeCP2 heterozygous female mice that are considered an accurate model of Rett syndrome despite having less severe symptoms than MeCP2 mutant male mice. Using optical imaging of intrinsic signals, they found adult OD plasticity after 4-day MD in mutant but not wildtype mice. Plasticity was not evaluated during the normal critical period or at any other age, so it is unknown if the critical period is delayed or extended in the absence of MeCP2. Interestingly, treatment with IGF-1 peptide abolishes this aberrant plasticity, and also alleviates other symptoms of Rett syndrome in these mice.

Together these results show that critical period plasticity is abnormal in these mouse models of autism; however, the way in which critical periods are altered appears to differ depending on the particular etiology of autism. This would make sense in light of the heterogeneity that is so characteristic of autism. Therefore it is imperative to thoroughly test all aspects of critical periods to see if they are accelerated, delayed, extended, weaker, stronger, and so on (as portrayed in Figure 1). Unfortunately all of the above studies have only compared the ratio of responses between the two eyes and have not looked at the functional readout of acuity. In the future, we hope several lines of autism mouse models will be systematically analyzed, complete with an evaluation of normal visual system functional development before any sensory manipulation is performed. A detailed examination of single-cell excitability and visual spatial acuity will reveal whether the model's visual system suffers from perturbed sensory processing, which is indicative of abnormal circuit refinement in visual cortex during development. After the baseline visual function is determined, then plasticity can be tested by short- or long-term monocular deprivation performed at various ages between eye opening and adulthood. Given the possible common disruption of PV-circuits across brain regions [58, 59] and the importance of the GABAergic system to critical period regulation [70], a multilevel analysis of PV-cell maturation should also be performed. The recent identification of new pharmacological and environmental strategies to recreate the highly plastic state of GABAergic circuitry possessed by juvenile animals represents a promising therapeutic avenue for the restoration of normal function in affected neuronal circuits [93].

The variable nature of E/I imbalance and altered plasticity in autism animal models suggests that the disruption of critical periods in autism is likely heterogeneous, in some cases resulting in excessive plasticity and in others, insufficient plasticity. This could be due to disruption of the mechanisms governing either the onset or closing of critical periods Figure 1, and both could be detrimental to functioning. A brain that is too plastic at the wrong times could result in noisy and unstable processing. On the other hand, a brain that lacks plasticity early in life might remain hyper- or hypoconnected and unresponsive to environmental changes early in life. A situation could also arise where plasticity is at an optimal level in some systems and an aberrant level in other systems, which could the case in Asperger and/or Savant syndrome.