Approaches to Understanding Multisensory Dysfunction in Autism Spectrum Disorder

Abstract

Abnormal sensory responses are a DSM-5 symptom of autism spectrum disorder (ASD), and research findings demonstrate altered sensory processing in ASD. Beyond difficulties with processing information within single sensory domains, including both hypersensitivity and hyposensitivity, difficulties in multisensory processing are becoming a core issue of focus in ASD. These difficulties may be targeted by treatment approaches such as “sensory integration,” which is frequently applied in autism treatment but not yet based on clear evidence. Recently, psychophysical data have emerged to demonstrate multisensory deficits in some children with ASD. Unlike deficits in social communication, which are best understood in humans, sensory and multisensory changes offer a tractable marker of circuit dysfunction that is more easily translated into animal model systems to probe the underlying neurobiological mechanisms. Paralleling experimental paradigms that were previously applied in humans and larger mammals, we and others have demonstrated that multisensory function can also be examined behaviorally in rodents. Here, we review the sensory and multisensory difficulties commonly found in ASD, examining laboratory findings that relate these findings across species. Next, we discuss the known neurobiology of multisensory integration, drawing largely on experimental work in larger mammals, and extensions of these paradigms into rodents. Finally, we describe emerging investigations into multisensory processing in genetic mouse models related to autism risk. By detailing findings from humans to mice, we highlight the advantage of multisensory paradigms that can be easily translated across species, as well as the potential for rodent experimental systems to reveal opportunities for novel treatments.

Lay Summary:

Sensory and multisensory deficits are commonly found in ASD and may result in cascading effects that impact social communication. By using similar experiments to those in humans, we discuss how studies in animal models may allow an understanding of the brain mechanisms that underlie difficulties in multisensory integration, with the ultimate goal of developing new treatments.

Introduction

Autism spectrum disorder (ASD) represents a behaviorally defined cluster of complex neurodevelopmental conditions characterized by impairment in social communication as well as the presence of repetitive and restrictive behaviors [APA, 2013]. Approximately 1 in 59 children are diagnosed with ASD, with 3–4:1 higher prevalence in males than females [CDC, 2019]. Beyond the classic social and repetitive behavior symptoms, abnormal sensory responses are also highly prevalent, with estimates of up to 85% of individuals with ASD exhibiting sensory features [Iarocci & McDonald, 2006; Robertson & Baron-Cohen, 2017]. Based upon this recognition, the DSM-5 now includes sensory disturbance as a symptom within the restricted, repetitive behavior domain of ASD [APA, 2013].

There is clear evidence that sensory processing across multiple modalities (sight, sound, touch, smell, and taste) is impacted in ASD [Baum, Stevenson, & Wallace, 2015a; Robertson & Baron-Cohen, 2017]. Studies have shown that individuals with ASD can be hypersensitive or hyposensitive to a variety of sensory stimuli, with both hypo-and hypersensitivity possible even within a single individual (e.g. hypersensitive to sight and hyposensitive to touch) [Hazen, Stornelli, O’Rourke, Koesterer, & McDougle, 2014; Lane, Young, Baker, & Angley, 2010; Leekam, Nieto, Libby, Wing, & Gould, 2007]. While all sensory modalities can be impacted in ASD [Robertson & Baron-Cohen, 2017], for the scope of this review we will mainly focus on visual and auditory processing, as it has been the major focus of multisensory work.

Founded on these differences in sensory function, a variety of sensory-based therapies and interventions including the Ayres Sensory Integration intervention [Parham et al., 2011] have been applied in ASD. Sensory-based therapies take a variety of approaches; including the active implementation and engagement of sensory stimuli with play behavior with the goal of improving adaptive behaviors, along with interventions focused on activating the arousal system with more passive inclusion of sensory information in the child’s normal routine, which may ameliorate some of the observed atypical responses to sensory stimuli (i.e. hyper- or hypo-sensitivity) [Ayres, 1972; Case-Smith, Weaver, & Fristad, 2015]. In addition to the Ayres Sensory Integration intervention, studies have evaluated these sensory-based therapies with mixed results, largely concluding there is little efficacy of these treatments or at least more evidence is still needed [Case-Smith et al., 2015; Lang et al., 2012; Schaaf, Dumont, Arbesman, & May-Benson, 2018; Watling & Hauer, 2015]. In order to apply novel sensory-based interventions, it is important to understand the underlying neural mechanisms responsible for the atypical sensory responses consistently observed in ASD. Toward this end, investigators have recently begun to harness studies on the neuronal, systems, and behavioral levels to evaluate performance not only in response to stimuli within a single sensory modality, but also when stimuli from multiple modalities are present [Baum et al., 2015a].

Introduction to Multisensory Processing

Multisensory processing represents the merging of sensory information from different modalities [Murray & Wallace, 2011]. We are constantly bombarded in our daily lives by a variety of sensory stimuli yet are largely able to integrate this information seamlessly in order to generate a coherent perception of the world. This has led to numerous studies investigating how the brain is capable of combining information from the various senses into a unified percept or gestalt [Calvert, Spence, & Stein, 2004].

Investigations in both humans and animals have shown that the combination of information from multiple senses can result in behavioral enhancements, such as increased accuracies and reduced reaction times, when compared to behavioral performances under unisensory (i.e. visual, auditory, or tactile alone) conditions [Stevenson et al., 2014]. For example, human psychophysical studies have shown that the added visual information of seeing a speaker’s lip movements can facilitate speech intelligibility in a noisy environment [Sumby & Pollack, 1954]. In addition to this work in the behavioral and perceptual realms, a great deal of work has now characterized the neural circuits and processes that underlie multisensory integration [Stein & Stanford, 2008]. Collectively, this work has shown that multisensory inputs converge at many sites in the nervous system [Meredith, 2002; Wallace, Meredith, & Stein, 1992; Wallace, Ramachandran, & Stein, 2004] and that this convergence frequently results in dramatic changes in neuronal responses when stimuli from multiple modalities are presented collectively [Meredith & Stein, 1983].

Atypical Multisensory Processing in Autism Spectrum Disorder

Based on the wealth of evidence highlighting disturbances across a number of sensory systems in ASD, there has been keen interest in studying the integration of information across these different sensory modalities [Beker, Foxe, & Molholm, 2018; Cuppini et al., 2017; Feldman et al., 2018]. Such multisensory integration is fundamental in allowing us to successfully interpret communicative and social signals. Numerous neural and behavioral studies have now demonstrated atypical multisensory processing in individuals with ASD [Baum et al., 2015a; Foss-Feig et al., 2010; Iarocci & McDonald, 2006; Kwakye, Foss-Feig, Cascio, Stone, & Wallace, 2011; Marco, Hinkley, Hill, & Nagarajan, 2011; Russo et al., 2010; Stevenson, Segers, Ferber, Barense, & Wallace, 2014; Wallace & Stevenson, 2014]. Evidence of decreased multisensory gain or benefit has been reported utilizing both simple and complex stimuli [Collignon et al., 2013; Mongillo et al., 2008; Stevenson et al., 2014; Stevenson et al., 2014]. Further, these multisensory deficits have been correlated with symptom severity [Brandwein et al., 2015; Yaguchi & Hidaka, 2018], and the development of multisensory function and its neural underpinnings may be atypical in ASD as well [Beker et al., 2018; Brandwein et al., 2012; Cuppini et al., 2017; Foxe et al., 2015; Stevenson et al., 2013; Taylor, Isaac, & Milne, 2010].

One common method to measure multisensory integration is with the use of cross-modal illusions. The McGurk effect is a well-known audiovisual speech illusion that has been evaluated in children, adults, as well as in individuals with neurodevelopmental disorders [Mcgurk & Macdonald, 1976; van Wassenhove, Grant, & Poeppel, 2007; Woynaroski et al., 2013]. In this illusion, a speaker typically mouths the phoneme “ga” while viewers simultaneously hear an auditory “ba.” Under these conflicting audiovisual conditions, individuals tend to perceive an entirely new speech percept, typically reporting that the speaker uttered the phoneme “da” or “tha.” Under these illusory conditions, it is believed that subjects are integrating the visual “ga” with the auditory “ba” to create the novel percept [Mcgurk & Macdonald, 1976]. Brain imaging studies have demonstrated that the superior temporal sulcus (STS) is integral for the perception of the McGurk effect [Nath & Beauchamp, 2012; Nath, Fava, & Beauchamp, 2011], and when this brain region is disrupted by transcranial magnetic stimulation (TMS) in typically developing individuals, illusory percepts decrease dramatically [Beauchamp, Nath, & Pasalar, 2010].

A variety of studies has also evaluated the McGurk effect in clinical populations, including in those with ASD. Overall, studies have demonstrated that individuals with ASD tend to perceive the McGurk effect less frequently in comparison to TD individuals [Irwin, Tornatore, Brancazio, & Whalen, 2011; Mongillo et al., 2008; Stevenson, Siemann, et al., 2013; Stevenson, Siemann, Schneider, et al., 2014; Taylor et al., 2010], with the majority of the studies describing this as a decrease in the strength or magnitude of multisensory integration. In addition, studies have evaluated and found differences in the perception of the McGurk effect across development in individuals with ASD [Stevenson, Siemann, et al., 2013; Taylor et al., 2010]. Also, studies, utilizing various psychophysical behavioral measures, have determined that the development of multisensory integration might be delayed in these individuals [Beker et al., 2018; Brandwein et al., 2012; Foxe et al., 2015; Stevenson, Siemann, et al., 2013; Taylor et al., 2010], along with evidence of sex-dependent differences in audiovisual speech development in both typically developed and ASD populations [Ross, Del Bene, Molholm, Frey, & Foxe, 2015]. Interestingly, neuroimaging studies have found both structural and functional differences in the STS of those with ASD, and this brain region, which has been shown to be important for the McGurk illusion, has also been established as a neural hub for multisensory integration and social processing [Beauchamp, Argall, Bodurka, Duyn, & Martin, 2004; Boddaert et al., 2004; Calvert, Campbell, & Brammer, 2000; Gervais et al., 2004; Redcay, 2008; Stevenson & James, 2009; Zilbovicius et al., 2006].

One of the critical factors in the integration of information across the different senses is time, in that only when these stimuli are sufficiently close together in time are they perceptually integrated. Specifically, there appears to be a window of time within which events specified by the different sensory modalities are perceptually “bound” [Hillock, Powers, & Wallace, 2011]. This multisensory temporal binding window (TBW) makes a great deal of ethological sense, since sensory stimuli presented in close temporal proximity are highly likely to be associated with the same event. To measure the TBW, auditory and visual stimuli are presented at various asynchronies relative to one another (including at simultaneity), and participants are asked to respond if they perceived these sensory signals as synchronous or not [Stevenson, Ghose, Fister, et al., 2014]. Typically developing participants tend to report high proportions of simultaneity for stimuli separated by about 300 ms, depending on the complexity of stimuli [Stevenson & Wallace, 2013]. However, it has been shown that the TBW can be dependent on participant task demands [Mégevand, Molholm, Nayak, & Foxe, 2013], audiovisual processing may be due to attentional switching [Shaw et al., 2020], and these capabilities may be developmentally different for typically developed individuals compared to individuals with ASD [Crosse, Foxe, & Molholm, 2019]. These points highlight the fact that task design, motivation, and attentional components are all highly relevant when selecting and utilizing psychophysical paradigms to investigate multisensory processing, especially in clinical populations such as ASD. Lastly, the width of the TBW has been shown to relate back to the perceptual illusions described above, specifically in that individuals who have wider TBWs tend to report the McGurk illusion less frequently [Stevenson, Zemtsov, & Wallace, 2012]. Such a finding suggests an important relationship between the TBW and the strength or magnitude of the multisensory binding process.

Emerging evidence suggests that the multisensory temporal binding window may be abnormally wide in individuals with ASD [Baum et al., 2015a; Baum, Stevenson, & Wallace, 2015b; Foss-Feig et al., 2010; Kwakye et al., 2011; Stevenson, Siemann, Woynaroski, et al., 2014; Wallace & Stevenson, 2014], resulting in degraded perceptual representations. For example, Stevenson and colleagues measured the TBW in both TD children and individuals with ASD across three levels of stimulus complexity: simple visual flashes and auditory beeps, more complex tool stimuli (i.e. a hammer hitting a nail) and a speaker uttering speech syllables. Temporal binding windows increased in width as the stimulus complexity increased for both groups, with the widest windows being observed for speech stimuli. Significant differences in the TBW widths were observed between the groups under the most complex conditions (i.e. speech), where individuals with ASD demonstrated wider TBWs compared to TD individuals. Interestingly, no significant differences in TBWs were observed between the groups for simple flashes and beeps and tool stimuli. This suggests that multisensory temporal processing may be typical in ASD for more simplistic audiovisual stimuli, however may become atypical as stimulus complexity increases. An alternative interpretation is that by utilizing a speaker uttering syllables, there may be intrinsic differences between the groups as there is evidence of atypical processing of faces in ASD [Dalton et al., 2005; Dawson, Webb, & McPartland, 2005]. This study also measured the McGurk effect, and individuals with ASD perceived the McGurk illusion less frequently compared to TD individuals. Importantly, it was found that for individuals with ASD, a wider TBW, regardless of the stimulus complexity, resulted in less perception of the McGurk effect. Therefore, this study hypothesized those differences in multisensory temporal processing for even low-level sensory stimuli (i.e. visual flashes and auditory beeps) in individuals with ASD could result in potential cascading effects into higher order domains such as speech or communication [Stevenson, Siemann, Schneider, et al., 2014].

In addition to the McGurk effect, the sound-induced flash illusion has also been examined in individuals with ASD. In this illusion, one visual flash is paired with multiple auditory beeps, resulting in the frequent reporting of multiple visual flashes [Shams, Kamitani, & Shimojo, 2000]. Using this illusion, it was observed that children with ASD had a temporal binding window that was almost twice as wide compared to TBWs of typically developing controls [Foss-Feig et al., 2010].

Importantly, studies have now specifically demonstrated relationships between multisensory temporal processing and severity of social and language impairments in individuals with ASD [Feldman et al., 2018; Patten, Watson, & Baranek, 2014; Stevenson et al., 2018; Wallace, Woynaroski, & Stevenson, 2020; Woynaroski et al., 2013]. Collectively, the body of evidence points to altered multisensory temporal function in ASD, and these temporal alterations are likely to have cascading effects on higher-order social communicative skills that necessitate the rapid and accurate integration of multisensory information [Stevenson et al., 2016, 2018; Yaguchi & Hidaka, 2018; Zhou et al., 2018].

In sum, numerous findings, using a variety of approaches, have demonstrated the presence of multisensory and temporal processing deficits in individuals with autism spectrum disorder [Baum et al., 2015b; Brandwein et al., 2015; Collignon et al., 2013; Foss-Feig et al., 2010; Foxe et al., 2015; Irwin et al., 2011; Kwakye et al., 2011; Smith & Bennetto, 2007; Stevenson, Siemann, et al., 2013; Stevenson, Siemann, Schneider, et al., 2014; Stevenson, Siemann, Woynaroski, et al., 2014; Woynaroski et al., 2013]. In addition, there is emerging evidence that (multi)sensory function may play a critical role in the establishment and maintenance of more complex functions such as social communication [Baum et al., 2015a; Robertson & Baron-Cohen, 2017; Stevenson et al., 2018; Stevenson, Siemann, Schneider, et al., 2014; Thye, Bednarz, Herringshaw, Sartin, & Kana, 2018]. With mounting evidence of atypical multisensory integration in ASD on both the behavioral and neural levels, cross-species studies in animal models may allow for investigations to understand the underlying neural mechanisms of multisensory processing, with the goal to translate that understanding into testable hypotheses—and potential treatments—in the human population.

The Neurobiology of Multisensory Integration

The use of animal models has been integral in determining the development [Stein, Stanford, & Rowland, 2014; Wallace & Stein, 1997, 2001; Wallace & Stein, 2007], neural mechanisms [Jiang, Wallace, Jiang, Vaughan, & Stein, 2001; Wallace et al., 1992; Wallace & Stein, 2000], and brain regions and circuits responsible for multisensory integration [Driver & Noesselt, 2008; Stein, 2012]. From these studies, many of the connections and neural structures critical for multisensory processing have been determined [Meredith & Stein, 1986b; Olcese, Iurilli, & Medini, 2013; Stein, Wallace, Stanford, & Jiang, 2002; Wallace & Stein, 1994]. In addition, a body of work has begun to establish relationships between these circuits and multisensory behaviors in a variety of species ranging from mice to non-human primates [Cappe, Murray, Barone, & Rouiller, 2010; Hirokawa, Bosch, Sakata, Sakurai, & Yamamori, 2008; Olcese et al., 2013; Wallace, Meredith, & Stein, 1998].

One of the best-studied models for this work has been the cat [Meredith & Stein, 1983]. Work in this model served to establish three main principles of multisensory integration (spatial, temporal, and inverse effectiveness) dependent upon the physical characteristics of the paired stimuli [Meredith, Nemitz, & Stein, 1987; Meredith & Stein, 1986a, 1986b]. The spatial and temporal principles specify that stimuli from different sensory modalities presented in close spatial and/or temporal proximity result in greater neural responses and behavioral benefits compared to those presented in a disparate manner [Meredith et al., 1987; Meredith & Stein, 1986a]. In addition, the principle of inverse effectiveness states that as the effectiveness (i.e. loudness, brightness) of the unisensory stimuli decreases, the resultant behavioral multisensory gain or benefit increases when stimuli are combined across sensory modalities [Meredith & Stein, 1986b].

Two of the traditional means to measure and evaluate the gains conferred by multisensory combinations are the interactive index (ii) and the mean statistical contrast (msc) [Stein & Stanford, 2008]. The interactive index compares multisensory responses to the largest or most effective unisensory response [Meredith & Stein, 1983, 1986a, 1986b]. The interactive index can be either positive or negative depending on whether the stimulus combination results in a gain (enhancement) or diminishment (depression) in response [Stein & Meredith, 1993].

Multisensory processing can be further evaluated by the mean statistical contrast. This compares the responses to multisensory stimuli to the combined responses from the individual sensory modalities [Perrault, Vaughan, Stein, & Wallace, 2003, 2005; Stanford, Quessy, & Stein, 2005]. Three types of responses (i.e. subadditive, additive, or superadditive) can be observed as a result of this evaluation.

In the cat, two of the best-studied structures from a multisensory perspective are the deep layers of a subcortical structure, the superior colliculus (SC) [Meredith et al., 1987; Meredith & Stein, 1986a, 1986b], and the anterior ectosylvian sulcal (AES) cortex [Wallace et al., 1992; Wallace, Carriere, Perrault, Vaughan, & Stein, 2006]. Intriguingly, it is the projections from the AES to the SC that gates the integration seen in SC neurons [Wallace, Meredith, & Stein, 1993; Wallace & Stein, 1994]. Pharmacological [Wilkinson, Meredith, & Stein, 1996] and deactivation studies [Jiang et al., 2001; Jiang, Jiang, & Stein, 2002; Wallace & Stein, 1994] show that the AES is critical for this gating, in addition to being integral for the behavioral enhancements observed under multisensory conditions.

Beyond studies in the cat; neural, circuit, and behavioral-based studies in the non-human primate have been crucial in understanding the underlying mechanisms of multisensory processing [Cappe et al., 2010; Cappe & Barone, 2005; Maier, Neuhoff, Logothetis, & Ghazanfar, 2004; Rockland & Ojima, 2003]. Striking similarities have been shown in both the development [Wallace, 2004], and neural response properties [Wallace & Stein, 2001] of multisensory integration in the cat and monkey [Wallace & Stein, 1996]. As in the cat, work in the non-human primate has shown the superior colliculus to be a major subcortical hub for multisensory integration, and neurons adhere to the established principles of multisensory integration [i.e. space, time, and effectiveness) (Wallace, Wilkinson, & Stein, 1996]. These similarities in development and neural response properties, between cat and non-human primate models, suggest that multisensory integration may be highly conserved across larger animal species [Table 1] [Stein & Stanford, 2008].

Table 1.

Comparing Cortical, Subcortical, and Behavioral Multisensory Findings Across Species

	Human	Non-Human primate	Cat	Rat	Mouse
Cortical	STS: [Beauchamp et al., 2004; Calvert et al., 2000; Nath et al., 2011; Nath & Beauchamp, 2012; Powers 3rd, Hevey, & Wallace, 2012; Stevenson & James, 2009; Stevenson, VanDerKlok, Pisoni, & James, 2011] Parietal cortex: [Calvert, Hansen, Iversen, & Brammer, 2001; Driver & Noesselt, 2008; Miller & D’esposito, 2005] PFC: [Calvert et al., 2001; Miller & D’esposito, 2005] Unisensory cortices: [Calvert et al., 1997; Foxe & Schroeder, 2005; Lakatos, Chen, O’Connell, Mills, & Schroeder, 2007; Martuzzi et al., 2007; Molholm et al., 2002; Watkins, Shams, Tanaka, Haynes, & Rees, 2006]	STS: [Bruce, Desimone, & Gross, 1981; Ghazanfar, Chandrasekaran, & Logothetis, 2008; Ghazanfar & Schroeder, 2006; Gu, Angelaki, & DeAngelis, 2008] Parietal cortex: [Avillac, Hamed, & Duhamel, 2007; Cohen, 2009; Schlack, Sterbing-D’Angelo, Hartung, Hoffmann, & Bremmer, 2005] PFC: [Fuster, Bodner, & Kroger, 2000; Romanski, Averbeck, & Diltz, 2005; Romanski & Goldman-Rakic, 2002; Sugihara, Diltz, Averbeck, & Romanski, 2006] Unisensory cortices: [Foxe & Schroeder, 2005; Ghazanfar, Maier, Hoffman, & Logothetis, 2005; Lakatos et al., 2007; Rockland & Ojima, 2003; Wang, Celebrini, Trotter, & Barone, 2008]	AES: [Jiang et al., 2001; Wallace et al., 1992, 1993; Wallace et al., 2006; Wallace & Stein, 1994; Wilkinson et al., 1996] Unisensory cortices: [Allman & Meredith, 2007]	Parietal cortex: [Brett-Green, Fifkova, Larue, Winer, & Barth, 2003; Lippert, Takagaki, Kayser, & Ohl, 2013; Menzel & Barth, 2005; Raposo, Kaufman, & Churchland, 2014] PFC: [Lipton, Alvarez, & Eichenbaum, 1999; Reid, Jacklin, & Winters, 2014] Insula: [Rodgers, Benison, Klein, & Barth, 2008] V2L: [Hirokawa et al., 2008; Schormans et al., 2017; Schormans & Allman, 2019; Schormans, Typlt, & Allman, 2017] Unisensory cortices and border regions: [Barth, Goldberg, Brett, & Di, 1995; Bieler, Sieben, Schildt, Röder, & Hanganu-Opatz, 2017; Maruyama & Komai, 2018; Sieben, Röder, & Hanganu-Opatz, 2013; Wallace et al., 2004; Xu, Sun, Zhou, Zhang, & Yu, 2014]	Parietal cortex: [Kuroki et al., 2018] PFC: [Sharma & Bandyopadhyay, 2019] V2L: [Meijer et al., 2020] Insula: [Gogolla, Takesian, Feng, Fagiolini, & Hensch, 2014] Unisensory cortices and border regions: [Cohen, Rothschild, & Mizrahi, 2011; Deneux et al., 2019; Knöpfel et al., 2019; McClure Jr & Polack, 2019; Meijer, Montijn, Pennartz, & Lansink, 2017; Morrill & Hasenstaub, 2018; Olcese et al., 2013]
Subcortical	SC: [Olive, Tempelmann, Berthoz, & Heinze, 2015]	SC: [Frens & Van Opstal, 1998; Wallace et al., 1996; Wallace & Stein, 1996; Wallace & Stein, 2000]	SC: [Meredith & Stein, 1986a; Meredith & Stein, 1986b]	SC: [Gharaei, Arabzadeh, & Solomon, 2018; Hirokawa et al., 2011; Lau, Manno, Dong, Chan, & Wu, 2018; May, 2006; Sparks & Hartwich-Young, 1989]	SC: [Doykos, Gilmer, Person, & Felsen, 2020; Drager & Hubel, 1975a; Drager & Hubel, 1975b]
Behavioral	AV detection: [Calvert et al., 2004; Frassinetti, Bolognini, & Làdavas, 2002; Miller, 1982; Raab, 1962] AV discrimination: [Raposo, Sheppard, Schrater, & Churchland, 2012; Sheppard, Raposo, & Churchland, 2013] SJ and TOJ tasks: [Colonius & Diederich, 2004; Stevenson, Ghose, Fister, et al., 2014; Stevenson, Segers, Ferber, et al., 2014; Stevenson, Siemann, et al., 2013; Stevenson, Siemann, Schneider, et al., 2014; Stevenson, Siemann, Woynaroski, et al., 2014; Stevenson, Wilson, Powers, & Wallace, 2013]	AV detection: [Cappe et al., 2010; Miller, Ulrich, & Lamarre, 2001; Wang et al., 2008] SJ and TOJ tasks: [Frens & Van Opstal, 1998; Kayser, Petkov, & Logothetis, 2008]	AV detection: [Burnett, Henkel, Stein, & Wallace, 2002; Burnett, Stein, Chaponis, & Wallace, 2004; Jiang et al., 2002; Stein, Meredith, Huneycutt, & McDade, 1989; Wallace et al., 1998]	AV detection: [Carandini & Churchland, 2013; Gleiss & Kayser, 2012; Hirokawa et al., 2008; Sakata, Yamamori, & Sakurai, 2004] AV discrimination: [Raposo et al., 2012, 2014; Sheppard et al., 2013] SJ and TOJ tasks: [Schormans & Allman, 2018; Schormans, Scott, et al., 2017; Schormans, Typlt, & Allman, 2017]	AV detection: [Meijer, Pie, Dolman, Pennartz, & Lansink, 2018; Meijer et al., 2020; (Siemann et al., 2015]

Note. This is by far not an exhaustive list; there are more examples of studies that can be found under each category for each species, and there are other cortical, subcortical, and behavioral findings that could be listed for each species, which are not highlighted. The purpose of this table is meant only to begin to highlight some of the neuronal and behavioral similarities across these animal species.

Abbreviations: AES: anterior ectosylvian sulcal cortex; AV: audiovisual; PFC: prefrontal cortex; SC: superior colliculus; SJ and TOJ: simultaneity judgment and temporal order judgment; STS: superior temporal sulcus; V2L: lateral portion of V2 (secondary visual cortex).

There are a variety of cortical regions in the non-human primate shown to receive convergent input from multiple sensory modalities, and to demonstrate many of the classical multisensory features [Ghazanfar & Schroeder, 2006; Kajikawa, Falchier, Musacchia, Lakatos, & Schroeder, 2012]. These include the prefrontal cortex [Fuster et al., 2000; Romanski et al., 2005; Romanski & Goldman-Rakic, 2002; Sugihara et al., 2006], parietal cortex [Avillac et al., 2007; Cohen, 2009; Mullette-Gillman, Cohen, & Groh, 2005; Schlack et al., 2005] and the medial superior temporal area [Bruce et al., 1981; Ghazanfar et al., 2008; Gu et al., 2008]. Further, primary unisensory cortices such as visual and auditory cortex have been shown to be modulated by multisensory stimulation, expanding greatly the view of what constitutes a “multisensory” brain region [Foxe & Schroeder, 2005; Ghazanfar et al., 2005; Lakatos et al., 2007; Wang et al., 2008]. In addition, non-human primate work has identified both subcortical and cortical connections and circuits needed for multisensory processing [Clavagnier, Falchier, & Kennedy, 2004; Falchier et al., 2010; Rockland & Ojima, 2003; Smiley & Falchier, 2009]. Most relevant to this review are those studies that then investigate the underlying neurophysiology and multisensory behavior in non-human primates [Angelaki, Gu, & DeAngelis, 2009; Frens & Van Opstal, 1998; Gu et al., 2008; Kayser et al., 2008; Miller et al., 2001], allowing for the mapping of neuron-behavior relationships. Overall, these studies have demonstrated strikingly similar neuroanatomical and functional connections between non-human primates and humans, see Table 1 [Ghazanfar & Schroeder, 2006; Kajikawa et al., 2012], and has provided critical insights into the neural substrates needed for multisensory processing in an animal model that more closely resembles the neural architecture found in humans.

While the original work that evaluated the underlying neurobiology and development of multisensory processing was initially performed in animal models [Murray & Wallace, 2011; Stein et al., 2014; Stein & Stanford, 2008], numerous studies have identified the necessary brain structures, and behavioral/perceptual benefits of multisensory integration in humans as well (Table 1) [Calvert et al., 1997; Calvert et al., 2001; Colonius & Diederich, 2004; Driver & Noesselt, 2008; Foxe & Schroeder, 2005; Frassinetti et al., 2002; Lakatos et al., 2007; Martuzzi et al., 2007; Miller, 1982; Miller & D’esposito, 2005; Olive et al., 2015; Raab, 1962; Stein, 2012; Stevenson, Ghose, Fister, et al., 2014; Watkins et al., 2006]. Emerging work in humans now focuses on evaluating multisensory processing as it relates to clinical disorders where sensory processing is known to be impacted [Baum et al., 2015a; Wallace & Stevenson, 2014]. Although classical studies in larger animal models have been instrumental in building our understanding of the neural and behavioral bases of multisensory processing, it must be acknowledged that these species are somewhat limited for modeling and studying disease. In contrast, the mouse and, more recently, the rat allow facile creation of genetic models of disease risk [Nestler & Hyman, 2010]. Critically, one key use of these animal models is the ability to back translate relevant clinical findings from the human to ask questions focused on the mechanisms into how (multi)sensory systems become atypical. Rodents may offer a translational bridge to evaluate the underlying mechanisms for multisensory function in genetic models.

Multisensory Studies in Rats

In most rodent multisensory behavioral studies, rats have been used to assess multisensory function based on their ability to complete complex operant tasks [Gleiss & Kayser, 2012; Sakata et al., 2004; Sheppard et al., 2013]. Based on the relative ease of stimulus presentation, and the relevance to the human population, audiovisual stimuli have been most frequently used to assess multisensory function behaviorally [Carandini & Churchland, 2013; Sakata et al., 2004], despite the fact that rodents tend to have fairly poor vision and more specialized somatosensory and olfactory systems [Ihara, Yoshikawa, & Touhara, 2013; Petersen, 2014]. While these studies have differed on the variety of stimuli, specific stimulus features, and task, they have all successfully demonstrated multisensory processing in rats, finding similar effects to those found in larger animal species (Table 1) [Carandini & Churchland, 2013; Gleiss & Kayser, 2012; Hirokawa et al., 2008; Raposo et al., 2012; Raposo et al., 2014; Sakata et al., 2004].

Beyond behavioral studies, work has identified neural structures important for multisensory processing in rats [Barth et al., 1995; Brett-Green et al., 2003; Hirokawa et al., 2008; Menzel & Barth, 2005; Sieben et al., 2013; Tees, 1999], including specific border regions between primary sensory cortices (visual, auditory, somatosensory) that contain a large number of multisensory neurons and demonstrate similar multisensory characteristics to those found in larger animal models [Schormans & Allman, 2019; Schormans, Typlt, & Allman, 2017; Wallace et al., 2004; Xu et al., 2014]. In addition, studies have demonstrated that seemingly irrelevant unisensory stimuli (i.e. auditory) can modulate responses in primary sensory cortices (i.e. visual) [Bieler et al., 2017; Maruyama & Komai, 2018], which could be a result of connectivity between primary sensory cortices [Stehberg, Dang, & Frostig, 2014]. In addition to primary unisensory cortices and their overlapping border regions, studies have identified the cerebellum [Ishikawa, Shimuta, & Hausser, 2015], insula [Rodgers et al., 2008], parietal cortex [Brett-Green et al., 2003; Lippert et al., 2013; Menzel & Barth, 2005], prefrontal cortex [Lipton et al., 1999; Reid et al., 2014], perirhinal cortex [Jacklin, Cloke, Potvin, Garrett, & Winters, 2016], and the lateral entorhinal cortex [Doan, Lagartos-Donate, Nilssen, Ohara, & Witter, 2019], as hubs for multisensory integration in rats. In addition to these cortical structures, studies in rats have also investigated the role of the superior colliculus (SC) in multisensory function on both the neural and behavioral level [Gharaei et al., 2018; Hirokawa et al., 2011; Lau et al., 2018; May, 2006; Sparks & Hartwich-Young, 1989]. Importantly, many of these subcortical and cortical findings are similar to those that have been observed in larger animal models (Table 1) [May, 2006; Wallace & Stein, 1996]. While similarities in responses to multisensory stimuli have been observed across animal species in similar neural structures, it is important to note that there are large inter-species differences in the functional roles of these regions. It is important to recognize that while there may be structural similarities between species (i.e. regions that were once thought to respond exclusively to unisensory stimuli, yet can respond under multisensory conditions) this does not necessarily mean that these structures function in exactly the same context as in larger animals and humans. Rodent studies allow us to gain insight into the cell-type specificity, genetic underpinnings, and offer an initial view into the brain structures that may be implicated. These findings are necessary as they can then be applied to larger animal models, which share more analogous structural and functional brain regions to humans, with the ultimate goal that these insights would lead to more clinically relevant applications.

A number of studies have utilized lesions, pharmacology, and/or cooling blockades to disrupt and evaluate multisensory circuits and function in animal models [Burnett et al., 2002; Burnett, Stein, Perrault, & Wallace, 2007; Jiang et al., 2001; Jiang, Jiang, & Stein, 2006; Wilkinson et al., 1996]. In the awake, behaving rat, Hirokawa and colleagues used pharmacology to transiently disrupt and identify a border region between the primary visual and auditory cortices, known as the lateral part of V2 or V2L, that is critical for audiovisual processing in rats [Hirokawa et al., 2008]. Allman and colleagues recently investigated the role of V2L in multisensory temporal processing [Schormans & Allman, 2018; Schormans, Scott, et al., 2017]. Remarkably, rats are capable of performing classic psychophysical simultaneity judgment (SJ) and temporal order judgment (TOJ) tasks [Schormans, Scott, et al., 2017], generating TBWs that appear to be similar to those found in the human population [Wallace & Stevenson, 2014]. Interestingly, studies in the mouse have also shown anatomical connections between primary auditory cortex (A1) and V2L, as well as primary visual cortex (V1) and V2L [Charbonneau, Laramee, Boucher, Bronchti, & Boire, 2012; Laramee, Kurotani, Rockland, Bronchti, & Boire, 2011], suggesting that V2L may be a multisensory structure conserved across rodent species.

Emerging Multisensory Studies in Mice

Over the past decade, increasing interest has focused on multisensory research in the mouse [Borges-Merjane & Trussell, 2015; Hornix, Havekes, & Kas, 2018; Olcese et al., 2013; Radvansky & Dombeck, 2018; Reig & Silberberg, 2014]. Importantly, similar findings to the rat have emerged in the mouse, where unisensory stimuli from other modalities can impact neuronal responses in primary sensory cortices [Cohen et al., 2011; Deneux et al., 2019; Knöpfel et al., 2019; McClure Jr & Polack, 2019; Meijer et al., 2017; Morrill & Hasenstaub, 2018; Zhang, Kwon, Ben-Johny, O’Connor, & Issa, 2020];results that may be supported by growing evidence that primary sensory cortices may share underlying connections with one another [Masse, Ross, Bronchti, & Boire, 2017; Morrill & Hasenstaub, 2018]. As observed in the rat (Table 1), studies have also shown that the parietal cortex [Kuroki et al., 2018; Lyamzin & Benucci, 2019; Najafi et al., 2019; Song et al., 2017], orbitofrontal cortex [Sharma & Bandyopadhyay, 2019], cerebellum [Chabrol, Arenz, Wiechert, Margrie, & DiGregorio, 2015] are neuronal hubs for multisensory integration in the mouse. Elegant work identifying subcortical (i.e. inferior colliculus - [Dillingham, Gay, Behrooz, & Gabriele, 2017], superior colliculus - [Doykos et al., 2020; Drager & Hubel, 1975a; Drager & Hubel, 1975b] to cortical connections [Lesicko, Hristova, Maigler, & Llano, 2016], cortical–cortical connections through transcranial electrical stimulation [Hishida, Kudoh, & Shibuki, 2014], and multisensory neuronal hubs in association cortices via calcium imaging [Kuroki et al., 2018] are beginning to demonstrate the underlying neural circuitry that supports multisensory processing in the mouse.

In addition to these physiological and circuit-based findings, our lab was the first to demonstrate multisensory processing on the behavioral level in mice [Siemann et al., 2015]. We found that mice from different background strains (i.e. C57 and 129 S4/S6) demonstrate increased behavioral accuracies under paired audiovisual conditions compared to auditory or visual conditions alone. Furthermore, this work in mice has begun to establish similarities in the stimulus characteristics and principles that support these behavioral benefits [Siemann et al., 2015]. Interestingly, a more recent study designed and utilized another behavioral paradigm to demonstrate inverse effectiveness in mice as well, and with this task the authors observed that stimulus detection is significantly improved under audiovisual conditions [Meijer et al., 2018]. The establishment of these behavioral paradigms set the stage for work structured to assess how molecular and circuitry manipulations impact multisensory behavior.

To this point, most recent work is now attempting to connect the underlying neural mechanisms with the associated multisensory behavior in the mouse model. It has been demonstrated that neurons in the V2L, also known as area AL, are responsive to audiovisual stimuli and critical for multisensory behavior [Meijer et al., 2020]. In this study, mice were head fixed and performed a stimulus detection task in which auditory, visual, and combined audiovisual stimuli were presented while simultaneously utilizing calcium imaging to record from neurons in layer II/III in V2L or V1. Meijer et al., found that neural responses could be elevated under audiovisual conditions in the V2L, and the neural population activity in this region corresponded with behavioral performance under multisensory conditions. This work is impactful as it identifies a novel neuronal hub, in the mouse, crucial for multisensory behavior, and also demonstrates the similarities across rodent species (i.e. V2L is implicated in both the rat and mouse), highlighting the possibility to utilize the mouse model in more complex brain-behavior based studies that may be relevant for larger animal species.

Multisensory Studies in Mouse Models Associated with ASD

Based on emerging results in wild type mice, there have been a few investigations of multisensory processing in genetically modified mice (Table 2) [Gogolla et al., 2014; Wada et al., 2019]. Gogolla and colleagues evaluated the neural mechanisms of multisensory processing in four mouse lines, including the Shank3 knockout mouse that models Phelan-McDermid syndrome, which is robustly associated with autism diagnosis. They also evaluated the Mecp2 null mouse model of Rett syndrome, the Gad65 knockout mouse model of GABA deficiency, and the BTBR T+tf/J inbred strain of mouse that demonstrate behaviors which serve as proxies for the human clinical phenotype of autism. However, it is important to note caveats such as the BTBR T+tf/J inbred mouse strain represents no known idiopathic form of ASD [Meyza & Blanchard, 2017], Rett syndrome itself is no longer considered within the autism diagnosis [APA, 2013], and the Shank3 mouse model is typically homozygous whereas in the clinical population, individuals display heterozygous mutations [Yi et al., 2016].

Table 2.

Findings of Altered Multisensory Processing in Neural and Behavioral Domains for Mouse Models Associated with ASD

Mouse model	Cortical	Subcortical	Behavioral
BTBR T+tf/J	Insula: [Gogolla et al., 2014]	N/A	N/A
Shank3 KO	Insula: [Gogolla et al., 2014]	N/A	N/A
MecP2 KO	Insula: [Gogolla et al., 2014]	N/A	N/A
Gad65 KO	Insula: [Gogolla et al., 2014]	N/A	N/A
SERT Ala56	N/A	N/A	AV detection: [Siemann et al., 2017]
Caps2 KO	N/A	N/A	Rubber tail illusion: [Wada et al., 2019]

Note. Note the recency, yet relatively limited number of studies, demonstrating the importance of these types of studies for future investigations.

The authors assessed multisensory processing, in anesthetized mice, in the insular cortex by recording neural responses utilizing in vivo imaging of intrinsic flavoprotein fluorescence to auditory (i.e. pure tones with varying frequencies and intensities to the contralateral ear) and somatosensory stimuli (i.e. air puffs to the contralateral front paw) along with the combined multisensory presentations. Multisensory responses were atypical and the magnitude of multisensory enhancement was reduced in the four selected lines in comparison to wild type animals. In addition, differences were observed in the size and densities of GAD65 puncta, and atypical miniature inhibitory postsynaptic currents in pyramidal cells from layers 2/3 of the insular cortex were found in response to auditory stimuli. Interestingly, the authors also found that the maturation of responses to auditory stimuli was atypical in BTBR mice across multiple developmental time periods. Based on these differences in inhibitory signaling and altered developmental trajectories, the authors chronically administered diazepam, a GABAA receptor positive allosteric modulator, between postnatal days 15–28. Multisensory function was then assessed during adulthood, with neural responses normalizing under multisensory conditions, compared to responses in control wild type mice, as well as a normalization of the abnormal self-grooming behaviors seen in these models [Gogolla et al., 2014]. In addition, the authors treated these mouse models with diazepam later in life, at postnatal days 48–54, and found that diazepam did not normalize electrophysiological responses to multisensory stimuli. These results suggest there are sensitive periods of development in which manipulation of GABAergic circuitry can result in the normalization of neuronal responses to multisensory stimuli. Therefore, this was the first study to demonstrate atypical (multi)sensory processing on the neuronal level in mouse models associated with ASD, demonstrated an altered developmental trajectory for multisensory function, and provides intriguing evidence suggesting that that this multisensory dysfunction could have cascading effects into core domains such as repetitive or restricted behaviors that characterize ASD.

In addition, our lab demonstrated the first evidence of abnormal multisensory behavioral responses in a genetically manipulated mouse (Table 2) [Siemann et al., 2017]. In this study, we utilized a mouse with a point mutation in the serotonin transporter (SERT) in which a glycine (Gly56) is substituted to an alanine (Ala56). In a study of multiplex ASD families, the Ala56 allele was the most common of a group of rare amino acid coding variants in the serotonin transporter [Sutcliffe et al., 2005]. The SERT Ala56 mouse shows elevated whole blood serotonin levels, paralleling the hyperserotonemia found in 25% of children with ASD, as well as abnormalities in social, communicative, and repetitive behaviors [Gabriele, Sacco, & Persico, 2014; Muller, Anacker, & Veenstra-VanderWeele, 2016]. We found that SERT Ala56 animals demonstrated a significant decrease in response accuracy to visual and auditory targets; however, the most significant behavioral deficits were observed under combined audiovisual conditions compared to wild type littermate controls. These multisensory effects were larger compared to differences between groups for either auditory or visual alone stimuli, however our results may also reflect poorer unisensory performance in SERT Ala56 mice that is further compounded when auditory and visual stimuli are combined. When examined in more detail, it was found that the behavioral gain normally seen under multisensory conditions was eliminated in SERT Ala56 mice. Overall, these findings represented the first to demonstrate atypical multisensory processing behaviorally in a genetically manipulated mouse. These behavioral findings provide the foundation for beginning to investigate the underlying cellular components and circuits that may lead to altered multisensory processing related to genetic risk factors [Hornix et al., 2018].

The importance of these studies is reinforced by the fact that serotonin has been consistently implicated in ASD [Cook & Leventhal, 1996; Mulder et al., 2004; Muller et al., 2016], has been shown to be important for sensory cortical development [Cases et al., 1996; Gaspar, Cases, & Maroteaux, 2003; Salichon et al., 2001], and can modulate responses to a variety of sensory stimuli [Esaki et al., 2005; Hurley, Thompson, & Pollak, 2002; Waterhouse, Azizi, Burne, & Woodward, 1990]. Most relevant to our work, there is evidence that serotonin may play a role in cross-modal plasticity [Jitsuki et al., 2011]. Jitsuki and colleagues demonstrated that under conditions of visual deprivation serotonin receptors are critical for increased synaptic strength and more precise tuning of somatosensory responses in the mouse barrel cortex. More recently, a study from Tang and colleagues demonstrated that serotonin can modulate microcircuits in the dorsal cochlear nucleus to enhance multisensory signaling by dampening auditory-related pathways [Tang & Trussell, 2017]. The loss of multisensory gain in the SERT Ala56 knock-in mouse further supports a role for serotonin in multisensory processing, with the potential for future studies to assess the ability of serotonin transporter blockade or other approaches to rescue this behavioral deficit [Robson et al., 2018].

A Translational Bridge: From Human Clinical Findings to Rodent Experimental Systems

Psychophysical studies show strikingly similar results and benefits under multisensory conditions in humans, non-human primates, cats, rats, and mice using comparable behavioral tasks (Table 1) [Carandini & Churchland, 2013; Raposo et al., 2012; Raposo et al., 2014; Schormans, Scott, et al., 2017; Sheppard et al., 2013; Siemann et al., 2015; Stevenson, Ghose, Fister, et al., 2014]. These studies demonstrate the potential utility of rodent models for studying multisensory function and provide evidence that these behavioral results could be compared across species, but a number of questions remain. By utilizing behavioral tasks [Meijer et al., 2018; Siemann et al., 2015], future studies may assess multisensory behavioral function for mice in order to mirror or relate these findings to those being observed in human psychophysical tasks.

To this point, studies are already using very similar psychophysical experiments as observed in humans to demonstrate multisensory temporal processing and temporal binding windows (TBWs) in rats and have identified the neural structures needed for this multisensory behavior [Schormans & Allman, 2018; Schormans, Scott, et al., 2017; Schormans, Typlt, & Allman, 2017]. Therefore, can mouse studies be refined to evaluate multisensory temporal processing, including temporal binding windows, and would these tasks implicate brain regions (i.e. V2L) similar to those observed in the rat model? Interestingly, V2L has just recently been identified in the mouse as a neural structure capable of responding to audiovisual stimuli and critical for multisensory detection behavior [Meijer et al., 2020]. This would strongly suggest that V2L may have a critical role for multisensory temporal processing in the mouse, highlighting the need for future studies to design novel behavioral tasks to measure temporal binding windows in this animal model as well. As a result, a number of important translational and clinically relevant studies could be evaluated. Prior work has already begun to show atypical multisensory processing in mouse models associated with ASD (Table 2) [Gogolla et al., 2014; Siemann et al., 2017; Wada et al., 2019]. Would future investigations find similar multisensory temporal behavioral deficits in genetic mouse models of ASD as are observed in the human clinical population?

In addition, mouse models allow for the screening and testing of specific pharmacological agents that could provide important preclinical information in normalizing multisensory function [Gogolla et al., 2014]. If genetic mouse models demonstrate atypical multisensory function it may be possible to investigate associations between similar genetic mutations and multisensory deficits in clinical subsets of ASD. For example, altered multisensory processing has been shown in the SERT Ala56 mouse model [Siemann et al., 2017], do individuals with ASD who have altered serotonin signaling (genetically or via hyperserotonemia) also demonstrate atypical multisensory function or temporal processing? This could allow for future studies focused on identifying associations between genetic abnormalities and multisensory function. Hypothetically, if there were a subset of individuals with ASD that had specific genetic alterations in serotonin signaling and also demonstrated altered multisensory processing, could the use of pharmacological treatment targeting the serotonin system (i.e. an SSRI), in this specific subset, improve altered multisensory processing? By evaluating and targeting multisensory dysfunction as a potential biomarker in animal models, it may be possible to ultimately use and apply this information for the clinical ASD population. While these studies may be difficult and clearly not as straightforward as proposed, they do highlight the advantages that genetic rodent models allow for in the assessment of multisensory behavioral function, which may be translated for larger animal species and ultimately could prove to be relevant for clinical applications.

Translational Approaches for ASD: Relationships Between Multisensory and Higher-order Cognitive Function

While most studies in humans have focused on treating the core symptoms of ASD, this approach has had relatively limited success [Dove et al., 2012; Warren et al., 2011]. To this point there are only two currently approved drugs for the treatment of ASD, yet these treat co-occurring behavioral problems, rather than core symptoms [Ghanizadeh, Sahraeizadeh, & Berk, 2014]. A variety of pharmacologic compounds have been found to rescue social or repetitive behaviors observed in genetic or behavioral mouse models of ASD [Gogolla et al., 2014; Silverman et al., 2012; Silverman et al., 2015];however, those compounds that have been tested in translational clinical trials have not yet demonstrated comparable benefits in humans with autism spectrum disorder [Ooi, Weng, Kossowsky, Gerger, & Sung, 2017; Silverman & Crawley, 2014; Veenstra-VanderWeele et al., 2017].

Although the focus of this intervention research, both in humans and in model systems, has focused on these core symptoms, we argue that it may be beneficial to test treatments or interventions that target altered (multi)sensory function. Since multisensory integration is critical for our perception of the world and is integral for the interpretation and understanding of social communicative signals, it is possible that treatment that targets of these sensory disturbances may have therapeutic effects that extend to core symptom domains. Indeed, if multisensory integration is a key building block in the construction of higher-order cognitive representations, it may serve as a tractable target for intervention that would have benefits that have cascading impact across both the processing hierarchy as well as in development [Baum et al., 2015a; Robertson & Baron-Cohen, 2017; Stevenson et al., 2018; Stevenson, Siemann, Schneider, et al., 2014; Thye et al., 2018]. With this possibility, it is important to carefully evaluate the potential relationships between multisensory processing and clinical domains such as social communication in ASD [Cascio, Woynaroski, Baranek, & Wallace, 2016; Feldman et al., 2018; Martínez et al., 2019; Patten, Labban, Casenhiser, & Cotton, 2016; Sartorato, Przybylowski, & Sarko, 2017; Wallace et al., 2020; Yaguchi & Hidaka, 2018].

Currently, the hypothesis that atypical multisensory processing underlies core ASD symptoms is plausible, yet it has only some evidence in humans, and much of that evidence is indirect. For example, Stevenson et al. demonstrated that atypical multisensory temporal processing in individuals with ASD resulted in poorer performance in perceiving a multisensory illusion that utilizes speech related stimuli [Stevenson, Siemann, Schneider, et al., 2014]. In addition, the brain region (STS) shown to be critical in the perception of this illusion [McGurk effect) (Beauchamp et al., 2010; Nath et al., 2011; Nath & Beauchamp, 2012], is also a key hub for the processing of social communicative stimuli [Pelphrey, Morris, & Mccarthy, 2004; Stevenson et al., 2011; Zilbovicius et al., 2006], and has been found to be both structurally and functionally atypical in ASD [Boddaert et al., 2004; Redcay, 2008]. In addition, studies have found that sensory disturbances correlate with greater deficits in social communication, along with the presence of more repetitive behaviors in individuals with ASD [Boyd et al., 2010; Boyd, McBee, Holtzclaw, Baranek, & Bodfish, 2009; Feldman et al., 2019; Foss-Feig, Heacock, & Cascio, 2012; Hilton et al., 2010; Kern et al., 2007; Schauder, Muller, Veenstra-VanderWeele, & Cascio, 2015; Stevenson et al., 2017; Stevenson et al., 2018; Watson et al., 2011; Zhou et al., 2018].

A number of future investigations could evaluate these relationships. Studies have shown that multisensory training paradigms with feedback can result in the narrowing of the TBW and increased functional connectivity between the STS and regions of the auditory and visual cortices in typically developed individuals [De Niear, Gupta, Baum, & Wallace, 2018; De Niear, Koo, & Wallace, 2016; Powers 3rd et al., 2012; Powers, Hillock, & Wallace, 2009]. Work has just begun to investigate if these same paradigms result in improved neural and behavioral multisensory function in individuals with ASD [Feldman et al., 2020], and plan to assess if these effects result in lasting changes to other domains such as social communication. There is evidence that seemingly irrelevant sensory stimuli can impact primary sensory systems [Ghazanfar & Schroeder, 2006]. Future studies could also utilize sensory-based therapies and/or perceptual training to investigate the underlying mechanisms for which improving unisensory function can result in increased multisensory function or temporal processing [Stevenson, Wilson, et al., 2013], and if this ultimately leads to lasting and positive changes in higher order domains for both typically developed and clinical populations [Wallace et al., 2020].

Animal models could also be used to evaluate the relationship between multisensory processing, social function, and repetitive behavior. Gogolla and colleagues demonstrated that a pharmacological intervention in mouse models associated with ASD not only normalized atypical multisensory neural responses, but it also reversed the presence of repetitive behaviors found in these animal models [Gogolla et al., 2014]. They did not, however, establish causality directly. Future studies should use optogenetics, electrophysiology, and pharmacology to more directly investigate the known circuits and neural populations that support multisensory function. For example, if relevant subcortical (i.e. superior colliculus) and/or cortical brain regions (i.e. unisensory cortices, V2L, or parietal cortex), critical for multisensory processing are selectively silenced either pharmacologically, through cooling blockade or lesioning methods, and/or specific neuronal populations (i.e. parvalbumin-positive interneurons) are reversibly silenced through optogenetic techniques [Olcese et al., 2013], does this result in impaired social function or the presence of repetitive/restricted behavior? The development of multisensory operant tasks in mice [Meijer et al., 2018; Siemann et al., 2015] also raises the opportunity to evaluate whether training in animal models can improve multisensory function, which may parallel approaches being developed in humans that seek to capitalize on multisensory training [De Niear et al., 2016; De Niear et al., 2018; Powers 3rd et al., 2012; Powers et al., 2009].

Conclusion

Atypical multisensory integration may be a tractable phenotype that can be observed across species and is becoming increasingly recognized as an important feature of ASD. In contrast to social communicative behaviors, which are quite species-specific, or repetitive behaviors, which take on very different forms in rodents than in humans, multisensory integration appears to have preserved brain mechanisms and is readily tested through behavioral assays that have been developed across species. With the coupling of these behavior paradigms and genetic manipulations, it may be possible to investigate molecular, cellular, and systems-level questions, and then attempt to translate these findings to the human clinical population. This approach is further supported by emerging evidence that suggests that altered (multi)sensory function has cascading effects that impact core symptoms, making this area of study not just tractable, but also potentially highly impactful.