How is sensory processing altered across exteroceptive and interoceptive domains in autism?

1

How is sensory processing altered across

exteroceptive and interoceptive domains in autism?

Iris Proff, University of Amsterdam

Abstract

Autism spectrum conditions (ASC) are characterized by social interaction deficits, repetitive behaviors and restricted interests. In recent years, there has been increasing interest in formerly neglected sensory abnormalities occurring in ASC. Quantative and qualitative tests of sensory perception reveal a large set of findings in both exteroceptive perception of the environment and interoceptive perception of sensations in the body. An array of theories has been proposed, each of which accounts for a subset of these findings. It remains unclear in how far these theories interact or overlap and in what way findings from different domains are connected. This review reviews aims to build a bridge between largely disconnected research fields of interoception and exteroception in ASC. Two major explanatory frameworks are proposed that incorporate isolated theories of specific perceptual processes in ASC which account for a large part of the reviewed findings. The stimulus binding framework holds that a decreased excitation-inhibition balance in ASC causes reduced neural synchronization which leads to deficits in binding of stimuli from both exteroceptive and interoceptive domains into unified percepts. The Bayesian framework holds that prediction errors are weighted disproportionally high in ASC, which leads to an inaccurate model of the environment and one’s body, from which no accurate predictions can be drawn. We argue that the neural mechanisms proposed by both frameworks act in conjunction and that abnormalities in behavior observed in autistic individuals are rooted in abnormalities in perceptual processes.

Abbreviations

ACC – anterior cingulate cortex

ADOS – Autism Diagnostic Observation Schedule AI – anterior insula

AQ – Autism Quotient

ASC – autism spectrum conditions BAQ – Body Awareness Questionnaire BPQ – Body Perception Questionnaire DLPFC – dorsolateral prefrontal cortex

DSM – Diagnostic Statistical Manual of Mental Disorders

ERP – event related potential

MAIA – Multidimensional Assessment of Interoceptive Awareness

MMN – mismatch negativity

MRS – magnetic resonance spectroscopy PI – posterior insula

PFC – prefrontal cortex

2

1 Introduction

Autism spectrum conditions (ASC)1 _{were first} defined by the latest version of the Diagnostic and statistical manual of mental disorders (DSM-5; American Psychiatric Association, 2013) and encompass the former diagnoses of autism, pervasive developmental disorder and Asperger’s syndrome. ASC are defined by core symptoms in social interaction and communication as well as repetitive behavior patterns, which typically arise in the first three years of development. In recent years, there has been increasing interest in another set of symptoms, namely abnormalities in sensory processing (Robertson & Baron-Cohen, 2017). This interest is based on frequent reports of hypo- and hyper-reactivity of autistic individuals to sensory stimuli, such as apparent indifference to pain or temperature, fascination by lights or aversive reactions to noisy environments. There are two major types of sensory signals: exteroceptive signals, originating from the external environment and interoceptive signals, originating from sources inside the body. Exteroceptive signals can be further divided into visual, auditory, tactile, gustatory and olfactory sensory modalities. While a growing body of research provides evidence that both interoceptive and exteroceptive processing is altered in ASC, both research fields are largely segregated. The goal of the present review is to examine associations between interoceptive and exteroceptive abnormalities, to outline in how far

1

For the sake of simplicity, we will use the term autistic to refer to people on the autistic spectrum.

existing neural and computational theories of ASC can account for findings from both domains and to construct a unifying framework incorporating the diverse findings and theories.

Even though autism has been used as clinical diagnosis since 1938 (Asperger, 1938), it was not until the 1990s that researchers observed abnormalities in exteroceptive perception in autistic subjects and started to explore the idea that such differences might underly the autistic phenotype. If autistic individuals perceive the world differently, this might give rise to differences in behavior. An early account of altered perception in ASC is the weak central coherence theory stating that ASC is marked by a reduced capacity to bind local stimuli to a global percept (Happé, 1996). Other researchers postulated an attentional over-focus (Casey, Gordon, Mannheim, & Rumsey, 1993), general perceptual enhancement (Mottron, Burack, Dawson, Soulières, & Hubert, 2001), neural synchronization deficits (Brock, Brown, & Boucher, 2002), an altered excitation/inhibition balance (Rubenstein & Merzenich, 2003), local over-connectivity (Markram, 2007) or global neural under-connectivity (Just, Keller, Malave, Kana, & Varma, 2012) as the core mechanisms underlying abnormalities autistic perception. Most recently, a Bayesian account of autistic perception has been developed (Pellicano & Burr, 2012), stating that a disbalance between prior predictions and incoming sensory evidence leads to abnormal perception and consequently abnormal behavior.

3 It remains controversial in how far these theories are conflicting, overlapping or complementary and if they describe mechanisms that are present across the whole autistic spectrum or characterize only specific subgroups.

The study of interoceptive differences in ASC is even more recent. Initiated by a discourse on how interpretation of interoceptive signals might fail in ASC (Quattrocki & Friston, 2014), several groups have investigated interoceptive abnormalities in ASC over recent years. Interoception is the processing of internal bodily sensations by the nervous system (Khalsa et al., 2018) and can be understood as the afferent component of homeostatic control loops within the body. Bodily signals such as temperature, pain, visceral sensations, hunger or heartbeat reach the brain via lamina-1 neurons in the spinal dorsal horn, projecting to the brainstem and further to the hypothalamus, thalamus and finally reaching the insula (Craig, 2003). Posterior insula (PI) has been identified as the primary interoceptive cortex representing the objective physiological state of the body, while anterior insula (AI) has been found to contain a subjective meta-representation of interoceptive signals (Craig, Chen, Bandy, & Reiman, 2000). In the AI, exteroceptive and interoceptive signals converge with valence and cognitive information, possibly forming the basis for subjective feeling states underlying the sense of self (Craig, 2009). A seminal paper by Garfinkel and colleagues (2015) defined three dimensions of interoception, which are now widely used in the field. The authors distinguish between

interoceptive accuracy (the accuracy in assessing

interoceptive signals, measured by objective behavioral tests), interoceptive sensibility (the subjective experience of interoceptive signals, measured by questionnaires) and interoceptive

awareness (also termed interoceptive insight, the

metacognitive ability to judge one’s interoceptive accuracy, measured by confidence-accuracy correspondence, such as area under an ROC curve or meta d’). However, no consensus has been reached regarding what constitutes an interoceptive signal and on how to measure interoception (Khalsa et al., 2018). These issues will be discussed in detail below.

In Chapter 1, we will review objective and subjective measures of exteroceptive processing. We will assess how visual, auditory, tactile and multisensory processing is altered in ASC and evaluate if certain patterns recur in different modalities. In Chapter 2, we will turn towards objective and subjective measures of interoceptive processing and examine if there is robust evidence pertaining to interoceptive alterations in ASC. Chapter 3 constitutes a review of studies investigating the integration of exteroceptive and interoceptive information in ASC. Based on the reviewed findings, we will discuss existing theories of autistic perception and combine them into two major explanatory frameworks in Chapter 4. We will end by pointing out the limitations of the current approach and the conclusions that can be drawn from the reviewed literature.

4

2 Exteroceptive processing

2.1 Search criteria

This part of the review contains studies published between 1998 and 2019 comparing exteroceptive accuracy or subjective report (henceforth referred to as exteroceptive sensibility) in a sample of subjects diagnosed with ASC with an age-matched control group. When perception is investigated using social stimuli (e.g. faces), sensory effects might be confounded with social abnormalities that influence perception. Therefore, to be able to draw clear conclusions about sensory processes in ASC, this review does not consider studies involving social stimuli. One exception to this rule are audio-visual speech stimuli, as these constitute a primary area of research into multisensory binding in ASC. However, we limited our search to studies focusing on purely phonetic and non-semantic aspects of language processing.

The literature on exteroceptive processing in ASC is extensive and contains numerous inconsistencies. Therefore, we only report frequently replicated results, as well as novel methods considered to bear high potential for future application. We do not consider studies that do not apply standardized diagnostic methods (DSM or ADOS).

2.2 Visual perception

The perceptual enhancement theory (Mottron et al., 2001; Mottron, Dawson, Soulières, Hubert, & Burack, 2006) postulates that sensory processing

abnormalities in ASC arise from enhanced perception of low-level stimulus features. However, in the visual domain, a compelling number of studies measuring visual acuity (Bölte et al., 2012; Kéïta, Mottron, & Bertone, 2010; Tavassoli et al., 2016), orientation (Freyberg, Robertson, & Baron-Cohen, 2016; Grubb et al., 2013; Koh, Milne, & Dobkins, 2010), motion direction (Manning, Tibber, Charman, Dakin, & Pellicano, 2015) and flicker discrimination (Bertone, Mottron, Jelenic, & Faubert, 2005; Pellicano, Gibson, Maybery, Durkin, & Badcock, 2005) of simple stimuli indicate that basic visual discrimination abilities in ASC are indistinguishable from those in control subjects.

Visual search

The most widely replicated finding related to visual processing in ASC is superior performance in visual search. Visual search tasks require subjects to determine if a target stimulus is present in a cluttered field of distractors. Typically, these tasks contain a disjunctive search condition, where the target has one unique feature (e.g. a unique colour), and a more difficult conjunctive search condition, where the target shares each of its’ features with some of the distractors (Treisman, 1982). An ASC advantage in terms of shorter response times and lower error rates has been shown for conjunctive search conditions (Kaldy, Kraper, Carter, & Blaser, 2011; O’Riordan, 2004; O’Riordan, Plaisted, Driver, & Baron-Cohen, 2001; Plaisted, O’Riordan, & Baron-Cohen, 1998; Shirama, Kato, & Kashino, 2017) as well as for

5 difficult disjunctive search conditions (Joseph, Keehn, Connolly, Wolfe, & Horowitz, 2009; O’Riordan, 2004; O’Riordan et al., 2001; Shirama et al., 2017). Moreover, the effect is stronger when the target is absent (Joseph et al., 2009; O’Riordan, 2004; O’Riordan et al., 2001; Plaisted et al., 1998; Shirama et al., 2017) and at large set sizes (O’Riordan, 2004; O’Riordan et al., 2001; Plaisted et al., 1998; Shirama et al., 2017). Two studies have shown that the ASC advantage in visual search remains when applying backwards masking (Shirama et al., 2017) or dynamical search (Joseph et al., 2009), both of which disrupt top-down serial search strategies.

Even though superior visual search in ASC has been widely replicated, some studies failed to find an effect (Baldassi et al., 2009; Constable, 2010; Grubb et al., 2013; Keehn, Shih, Brenner, Townsend, & Müller, 2013). Out of these, one study measured the performance in relative discrimination thresholds, rather than in response times or error rates (Baldassi et al., 2009). This might be an invalid approach, as it has been shown that discrimination thresholds do not correlate with reaction times in visual search (Brock, Xu, & Brooks, 2011). Discrepancies in findings might be due to a dependency on motor ability (Lindor, Rinehart, & Fielding, 2018), which is commonly impaired in ASC (Green et al., 2009). The ASC advantage may be masked by reduced motor abilities, as visual search requires fine control of saccades.

In sum, there is substantial evidence for a superior visual search ability in ASC, particularly in difficult search conditions.

Parallel processing

It has been argued that the ASC advantage in visual search might be accounted for by increased parallel processing and consequently decreased automatic filtering of stimuli (Remington, Swettenham, & Lavie, 2012). This is supported by findings showing that autistic subjects require higher perceptual load to ignore distracting information (Remington, Swettenham, Campbell, & Coleman, 2009; Remington et al., 2012), indicating higher perceptual capacity. Moreover, the number of scrutinized items in visual search increased with set size in autistic toddlers, but not in controls (Kaldy et al., 2011). Compellingly, increased parallel processing might also explain why the ASC advantage in visual search prevails when the target is absent and at large set sizes, as in these cases many stimuli must be processed simultaneously.

Taken together, these findings indicate that autistic individuals process more visual information in parallel, which is beneficial to visual search, but detrimental when a task requires filtering out distracting information. However, this conclusion seems to contradict findings from crowding paradigms, where subjects need to ignore distracting stimuli flanking the target. Some studies indicate superior (Baldassi et al., 2009; Kéïta et al., 2010) while others indicate typical (Constable, 2010; Freyberg et al., 2016; Grubb et

6 al., 2013) performance of autistic subjects in crowding tasks. This might indicate that the detrimental effects of increased distractor processing may only manifest in situations with high perceptual load.

Local-global processing

Disembedding tasks, which require subjects to respond to local aspects of a stimulus, while ignoring its global features (Mottron, Burack, Iarocci, Belleville, & Enns, 2003), can be used to test processing of local and global stimulus features. A recent review of 20 disembedding studies shows that autistic subjects frequently display shorter response times in disembedding tasks, while accuracy is typical (Horlin, Black, Falkmer, & Falkmer, 2016). Supporting a local processing style, autistic individuals have been shown to focus on local, rather than global aspects of complex stimuli using Machine Learning on eye tracking data (Wang et al., 2015) and match-to-sample tasks (Nayar, Voyles, Kiorpes, & Di Martino, 2017).

Addressing the notion that global processing deficits prevail in autistic individuals, a number of studies investigated global motion coherence perception, requiring subjects to segregate signal stimuli from global noise. While several studies found decreased motion coherence perception in ASC (Milne et al., 2002; Pellicano et al., 2005), others did not find group differences (Chen et al., 2012; Jones et al., 2009; Koldewyn, Whitney, & Rivera, 2010; Manning et al., 2015; Peiker, Schneider, et al., 2015). Two studies

provide evidence that differences in global motion perception might depend on stimulus duration, where an autistic disadvantage is only present at very short stimulus durations (200 ms; (Robertson, Martin, Baker, & Baron-Cohen, 2012; Robertson et al., 2014).

These results suggest that local processing of visual stimuli is superior in autistic individuals, while fast integration of local stimuli in the global context is impaired.

Binocular rivalry

In binocular rivalry paradigms, distinct visual stimuli are presented to each eye of the participant, such that conscious perception switches between both percepts. Two studies have found that switch rates in binocular rivalry are diminished in autistic subjects (Robertson, Kravitz, Freyberg, Baron-Cohen, & Baker, 2013; Spiegel, Mentch, Haskins, & Robertson, 2019) and that they show longer durations of mixed percepts (Robertson et al., 2013). The authors argue that this speaks to reduced lateral inhibition between neuronal populations encoding the competing stimuli. A conflicting study found no difference in mixed percept duration (Said, Egan, Minshew, Behrmann, & Heeger, 2013). This discrepancy might arise out of the use of less complex and non-colourful stimuli, which have been shown to evoke less lateral inhibition (Alais & Melcher, 2007).

Oscillatory abnormalities

Several studies have reported abnormal oscillatory responses to visual stimuli in the gamma frequency

7 band (30-120 Hz) in ASC. Evoked oscillations in the gamma range in response to complex or illusory stimuli have been found to be decreased (Buard, Rogers, Hepburn, Kronberg, & Rojas, 2013; Stroganova et al., 2012), opposed to typical gamma band responses to simple visual stimuli (Milne, Scope, Pascalis, Buckley, & Makeig, 2009; Stroganova et al., 2012). Conversely, gamma power has been shown to be elevated during sustained visual attention in two independent samples of autistic children (Orekhova et al., 2007).

When presented with periodic visual stimuli, neural ensembles synchronize their activity with the frequency of the stimulus, as well as with multiples of the stimulus frequency, a process known as neural entrainment (Lazarev, Simpson, Schubsky, & DeAzevedo, 2001). Autistic subjects show reduced neural entrainment both at the stimulation frequency (Snijders, Milivojevic, & Kemner, 2013), as well as at multiples of the stimulation frequency (Lazarev, Pontes, & deAzevedo, 2009).

Moreover, evidence points towards a deficiency in synchronizing oscillations between brain regions. Specially, interhemispheric synchronization in the beta range (12-30 Hz) in sensory cortices when presented with visual stimuli is reduced in ASC (Isler, Martien, Grieve, Stark, & Herbert, 2010; Lazarev, Pontes, Mitrofanov, & deAzevedo, 2013; Peiker, David, et al., 2015).

This evidence suggests that while the production of neural oscillations is unaffected or even enhanced, the accurate timing of neural

oscillations in response to visual stimuli is impaired in ASC.

Visual predictions

A single study investigated the effect of visual predictions in ASC. In a cued visual association task, autistic subjects gazed longer at the correct location, when the incorrect location was predicted by the cue (Greene et al., 2019). This finding indicates a decreased reliance on predictions in ASC.

2.3 Auditory perception

There is evidence for a deficit in temporal processing of auditory stimuli in autistic children. Children with ASC require 48% longer stimulus onset asynchronies (SOAs) to determine the order of two rapidly presented tones (Kwakye, Foss-Feig, Cascio, Stone, & Wallace, 2011) and display fewer auditory ERPs to the second tone (Oram Cardy, Flagg, Roberts, Brian, & Roberts, 2005) than controls. Moreover, time estimation of auditory stimuli is impaired in autistic children (Szelag, Kowalska, Galkowski, & Pöppel, 2004).

These findings could be accounted for by an abnormal temporal encoding of auditory stimuli in the brain. This is supported by range of EEG and MEG studies reporting delayed M100 responses to tones (Edgar, Fisk, et al., 2015; Edgar, Khan, et al., 2015; Gandal et al., 2010; Roberts et al., 2010).

8 Auditory mismatch negativity

A large number of studies have investigated auditory deviance processing in ASC by examining mismatch negativities (MMN) in auditory oddball paradigms. Reported differences in amplitudes and latencies of MMNs are highly inconsistent. In a recent meta-analysis, out of 22 reviewed studies only 6 counterbalanced features of standard and deviant stimuli (Schwartz, Shinn-Cunningham, & Tager-Flusberg, 2018). Among these, despite conflicting results, the authors report a tendency to reduced MMN amplitude in ASC subjects, but no clear pattern of latency differences. However, differences in stimulus features might account for the heterogeneity of the results. A clearer picture arises when results are reviewed separately for the deviant features, namely pitch, phoneme category or duration.

All six studies reported by Schwartz investigated differences in pitch. Three studies by Lepistö and colleagues consistently show an increased MMN amplitude for pitch changes for speech sounds as well as preserved or increased amplitudes for pitch changes for tones (Lepistö et al., 2005; Lepistö, Nieminen-von Wendt, von Wendt, Näätänen, & Kujala, 2007; Lepistö et al., 2006) in ASC. Crucially, these studies apply very small differences in pitch (12 Hz). Distinguishing the pitch of notes is harder than of words (Heaton, Hudry, Ludlow, & Hill, 2008), possibly accounting for different findings for these two stimulus categories. Moreover, two studies applying big pitch differences of 200 and 400 Hz respectively, found no difference of MMN amplitudes (Roberts

et al., 2011; Weismüller et al., 2015). The last reported study showed smaller MMN amplitudes for pure tones (Dunn, Gomes, & Gravel, 2007) with a pitch deviance of 200 Hz. However, subjects watched a movie that was not silenced and were thus confronted with competing auditory stimuli. The difference in amplitude disappeared when no movie was shown, suggesting that the reported decreased MMN amplitudes might result from reduced filtering of auditory information in ASC (as discussed below).

None of the four studies investigating differences in phoneme-category reported effects on MMN amplitudes. Finally, three studies investigated duration deviance and found that MMN amplitudes for duration changes were decreased in autistic children (Lepistö et al., 2005, 2006), but increased in autistic adults (Lepistö et al., 2007). The age-related differences in duration processing might indicate a delayed development of accurate temporal auditory processing in ASC.

Taken together, ASC subjects appear to show higher MMN amplitudes for pitch deviance, when it is difficult to perceive and competing with other auditory information. On the other hand, duration deviance processing seems to be impaired in autistic children, in line with temporal processing deficits mentioned above. These findings indicate that deviance processing in ASC is highly dependent on stimulus features and not generally impaired or increased.

9 Sound segregation

A range of studies applying diverse methods provide evidence for an impairment in filtering of auditory stimuli based on a stimulus features in ASC. Autistic subjects have been shown to fail to automatically segregate streams of tones with distinct frequency ranges (Bouvet, Mottron, Valdois, & Donnadieu, 2013; Lepistö et al., 2009) and spatial locations (Teder-Sälejärvi, Pierce, Courchesne, & Hillyard, 2005). When subjects are required to identify auditory stimuli among distractors with common features, the failure to automatically filter auditory stimuli in ASC leads to processing advantages (Bouvet et al., 2013; Lin, Yamada, Komine, Kato, & Kashino, 2015). One study examined how perceptual load influences auditory filtering in ASC, using a dual task paradigm (Remington & Fairnie, 2017). Autistic subjects’ performance in an auditory detection task declines less with increasing perceptual load, indicating enhanced perceptual capacity goes along with weaker auditory filtering.

Another type of sound segregation is the segregation of local and global features. Paralleling evidence of enhanced visual disembedding, two studies found an increased ability to extract local features from melodies in ASC (Bouvet, Simard-Meilleur, Paignon, Mottron, & Donnadieu, 2014; Mottron, Peretz, & Menard, 2000). In both studies, the perception of global features of the melody was indistinguishable between groups.

In sum, when presented with complex auditory stimuli made up of several parts, autistic subjects tend to apply less automatic filtering or

grouping of stimuli and perceive each part independently. Depending on task demands, this might lead to processing advantages or disadvantages.

Pitch discrimination

A range of studies have reported an advantage of autistic children in the perception of subtle pitch changes of pure tones (Bonnel et al., 2003; Heaton et al., 2008; O’Riordan & Passetti, 2006), as well as better pitch memory (Heaton, 2003). The prevalence of absolute pitch in ASC is estimated to be between 5% (Rimland, 1978) and 11% (DePape, Hall, Tillmann, & Trainor, 2012). Conversely, absolute pitch is rare in the general population, with estimations of less than 0.01% (Takeuchi & Hulse, 1993). A study involving a large autistic sample (n = 72) found that only a subgroup of 20% showed exceptional pitch discrimination, while no differences were found on the group level (Jones et al., 2009). Thus, musical savants might represent a specific phenotype within the ASC population.

Oscillatory abnormalities

A range of MEG studies have reported abnormal oscillatory responses to auditory stimuli in autistic subjects. Specifically, evoked gamma power (40-60 Hz) was decreased in response to pure tones, while induced gamma power was increased (Edgar, Khan, et al., 2015; Rojas, Maharajh, Teale, & Rogers, 2008), also reflected by a higher phase locking factor of neural responses in ASC (Gandal et al., 2010; Rojas et al., 2008). Moreover, when presented with an acoustic click train of 40 Hz,

10 autistic subjects showed decreased neural entrainment in the gamma range (Wilson, Rojas, Reite, Teale, & Rogers, 2007). One study found that gamma range coherence between parietal-temporal areas was higher in typically developing infants than in infants who later developed ASC, when presented with speech sounds (Righi, Tierney, Tager-Flusberg, & Nelson, 2014).

These results parallel findings from the visual domain and indicate impaired neural synchronization in response to auditory stimuli, both within and between brain regions.

Auditory predictions and habituation

Three recent studies have investigated the effects of auditory predictions and habituation in autistic subjects, consistently showing diminished attenuation of auditory processing. Two of these studies modulated the predictability of deviant stimuli in an auditory oddball paradigm. The first study found that the MMN amplitude to deviant stimuli decreased less over time (as deviants become less surprising) in autistic subjects (Hudac et al., 2018). The second study modulated the global context by modulating the occurrence rate of deviant stimuli. When the deviant stimulus occurred more frequently (and was thus less surprising), control subjects showed an attenuated MMN response, whereas this effect was weaker in autistic subjects (Goris et al., 2018). Crucially, the results from both studies can be interpreted either as decreased top-down predictions or as decreased habituation to recurring deviant stimuli in the autistic sample. The third study introduced

predictions by contrasting self-generated (and thus predictable) to externally generated auditory stimuli. Whereas control subjects showed a decreased M100 response to self-generated stimuli, this attenuation was not present (van Laarhoven, Stekelenburg, Eussen, & Vroomen, 2019).

These findings suggest that the attenuating effect of predictions and habituation to recurring auditory stimuli is diminished in ASC.

2.4 Tactile perception

Most research on tactile perception in ASC has focussed on subjective self- or parent-reports. A few recent studies applied vibrotactile stimulation to objectively estimate tactile perception thresholds. Results from these studies are mixed, likely due to due small sample sizes (Mikkelsen, Wodka, Mostofsky, & Puts, 2018). One consistently replicated finding is an absent adaptation effect during tactile stimulation in ASC. In control subjects, detection thresholds for static stimulation are lower than for dynamic stimulation that increases slowly in strength until it reaches a detectable level, while this effect is absent in autistic subjects (Puts et al., 2017; Puts, Wodka, Tommerdahl, Mostofsky, & Edden, 2014; Tavassoli et al., 2016). Moreover, the effect of adaptation on tactile amplitude discrimination and spatial localization of supra-threshold vibrotactile stimuli has also been found to be absent in ASC (Puts et al., 2017, 2014; Tommerdahl, Tannan, Cascio, Baranek, & Whitsel, 2007).

11 Adaptation to both perceptible and imperceptible tactile stimuli is thought to depend on lateral GABAergic inhibition (Blankenburg et al., 2003; Tommerdahl, Favorov, & Whitsel, 2010). Magnetic resonance spectroscopy (MRS) studies have found decreased GABA levels in somatosensory cortex of autistic adults (Sapey-Triomphe, Lamberton, Sonié, Mattout, & Schmitz, 2019) and children (Puts et al., 2017). Furthermore, in autistic children, GABA levels in somatosensory cortex correlate with the difference between static and dynamic detection thresholds (Puts et al., 2017).

Together, these findings indicate that tactile processing abnormalities in ASC might be linked to decreased sensory habituation caused by a neuronal inhibition deficiency.

Oscillatory abnormalities

As in the visual and auditory domains, a dysfunction of gamma wave synchronization in ASC has been observed in response to tactile stimuli. Using a 25 Hz vibrotactile stimulus, one study found that the entrained response at 50 Hz (thus within the gamma range) was diminished in ASC compared to controls (Khan et al., 2015).

Moreover, perception thresholds of autistic subjects on tactile temporal order judgment tasks are unaffected by the presentation of a competing vibrotactile stimulus on a nearby stimulation site (Tommerdahl, Tannan, Holden, & Baranek, 2008). Conversely, perception thresholds of typically developing individuals increase 3- to 4-fold with the presentation of the competing

stimulus. The authors argue that this indicates a lack of synchronization between nearby neural ensembles in ASC which impedes a binding of both stimuli, such that they can be perceived individually.

2.5 Multisensory perception

When perceiving two stimuli from different sensory modalities, the brain needs to solve the so-called binding problem. Thus, it needs to determine if the stimuli were caused by the same or distinct events or objects. The more proximal two stimuli are perceived in time, the more likely they are to be bound into a unified percept in the brain (Stevenson et al., 2016). The maximum temporal distance between two stimuli perceived as synchronous is termed the temporal binding window. A small tolerance of temporal delay is necessary, as modalities differ in conductance speed (e.g. light travels faster than sound) as well as neural transmission speed (King, 2005).

Temporal binding windows can be assessed by behavioral tasks, most often synchrony or temporal order judgments of two subsequent stimuli from different sensory modalities. A recent meta-analysis of 10 studies reports a general tendency for enlarged multisensory temporal binding windows in ASC (Zhou et al., 2018). However, findings are heterogeneous. Some studies have reported larger temporal binding windows in ASC when using non-linguistic audio-visual stimuli (Foss-Feig et al., 2010; Kwakye et al., 2011; Noel, Lytle, Cascio, &

12 Wallace, 2018) or speech stimuli (Noel, De Niear, Lazzara, & Wallace, 2018; Noel, De Niear, Stevenson, Alais, & Wallace, 2017). Conversely, others have not found such differences using non-linguistic (Noel et al., 2017; Stevenson et al., 2018; Stevenson, Siemann, Schneider, et al., 2014; Turi, Karaminis, Pellicano, & Burr, 2016) or speech stimuli (Grossman, Schneps, & Tager-Flusberg, 2009; Smith, Zhang, & Bennetto, 2017). When applying visuo-tactile stimuli, temporal binding windows have been reported to be enlarged in ASC (Greenfield, Ropar, Smith, Carey, & Newport, 2015; Noel, Lytle, et al., 2018).

Two studies have reported that the typical fast adaptation to asynchronous non-linguistic audio-visual stimuli observed in nonclinical populations (Van der Burg, Alais, & Cass, 2013) is absent in ASC (Noel et al., 2017; Turi et al., 2016). Whereas control subjects adapt within one trial to asynchrony between two stimuli, autistic subjects do not adjust their perception of synchrony based on past stimuli. The underlying mechanism of this effect is disputed, and it might be interpreted either as a failure of top-down predictions or of decreased bottom-up adaptation to asynchrony.

A more indirect test of synchrony perception is constituted by undirected watching paradigms using eye tracking. Subjects are simultaneously presented two video clips, where visual and auditory information is synchronous in one, but asynchronous in the other. Autistic children consistently show reduced preference for synchronous videoclips (Bebko, Weiss, Demark, & Gomez, 2006; Falck-Ytter, Nyström, Gredebäck,

Gliga, & Bölte, 2018; Falck-Ytter, Rehnberg, & Bölte, 2013; Grossman, Steinhart, Mitchell, & Mcilvane, 2015).

Taken together, there is substantial evidence for impaired perception of synchrony in ASC that manifests in increased temporal binding windows of multisensory stimuli.

McGurk illusion

Multisensory illusions can be used to quantify how strongly one modality influences perception in another modality. Most frequently studied is the McGurk illusion (Mcgurk & Macdonald, 1976), where subjects are presented with videoclips of spoken syllables and are asked to report which syllable they heard. In the illusory condition, the auditory stimulus is ‘ba’, whereas the visual percept is ‘ga’. Often, subjects report the bound percept ‘da’, synthesizing information from auditory and visual modalities. This ‘McGurk-effect’ is thought to be a measure of successful audio-visual binding, which is crucial for speech understanding (Stevenson et al., 2016). A range of studies have found that autistic subjects show a decreased McGurk-effect (Bebko, Schroeder, & Weiss, 2014; DePape et al., 2012; Irwin, Tornatore, Brancazio, & Whalen, 2011; Mongillo et al., 2008; Stevenson, Siemann, Schneider, et al., 2014; Stevenson, Siemann, Woynaroski, et al., 2014; Williams, Massaro, Peel, Bosseler, & Suddendorf, 2004), indicating weaker influence of visual signals on auditory perception in ASC. Conversely, three studies did not find group differences in the frequency of bound percepts (Iarocci, Rombough,

13 Yager, Weeks, & Chua, 2010; Stevenson et al., 2018; Woynaroski et al., 2013).

Multisensory facilitation and competition

Multisensory facilitation occurs when perception of a stimulus is facilitated by an additional stimulus in another modality. Using simple audio-visual response time tasks, two studies have found a lack of multisensory facilitation in autistic subjects (Brandwein et al., 2013; Ostrolenk, Bao, Mottron, Collignon, & Bertone, 2019). While control subjects responded quicker to audio-visual stimuli than to auditory or visual stimuli alone, this effect was absent in the ASC group. Similarly, one study found that visual search performance was facilitated by auditory cues in controls, but not in autistic subjects (Collignon et al., 2013).

When applying electrophysiology, multisensory facilitation is reflected by the difference between the summed response to unisensory stimuli and the response to multisensory stimuli. This difference, also referred to as multisensory difference wave, has been shown to be decreased or even absent in autistic subjects in visuo-tactile (Russo et al., 2010) as well as audio-visual stimulation (Brandwein et al., 2013).

In certain contexts, multisensory stimuli might compete for attention and thus impede, rather than facilitate, perception (Sinnett, Soto-Faraco, & Spence, 2008). One EEG study investigated the effect of multisensory competition in ASC applying an audio-visual attention task. In each trial, subjects were required

to respond to either the visual or the auditory modality, while ignoring the other (Jeremy Murphy, Foxe, Peters, & Molholm, 2014). Autistic subjects showed increased interference by the non-attended condition as well as a smaller modulation of preparatory alpha oscillations. These findings indicate that the lack of multisensory facilitation in ASC is accompanied by increased multisensory competition and an inability to filter stimuli based on modality.

2.6 Exteroceptive sensibility

Exteroceptive sensibility refers to the subjective experience of exteroceptive stimuli, assessed by self- or parent-reports. Exteroceptive sensibility is usually assessed using items measuring hypo-sensibility (e.g. ‘Appears not to notice loud and

sudden noises’), hyper-sensibility (e.g. ‘Bothered by sounds’), sensation seeking (e.g. ‘Touches people and objects’) and sensation avoiding (e.g. ‘Holds hands over ears to protect ears from sound’). A set

of studies involving large sample sizes found consistently heightened exteroceptive sensibility in ASC across all sensory modalities (Ben-Sasson et al., 2007; Kern et al., 2006; Minshew & Hobson, 2008; Silva & Schalock, 2012; Tavassoli, Miller, Schoen, Nielsen, & Baron-Cohen, 2014; Tomchek & Dunn, 2007). Intriguingly, the above listed studies report that ASC is associated with increased scores in both hypo- and hyper-sensibility as well as in sensation seeking and sensation avoiding measures, even though these measures appear to be opposed on first sight.

14 A possible resolution to this paradox is provided by a study that aimed to characterize subgroups within a large sample of autistic children (Liss, Saulnier, Fein, & Kinsbourne, 2006). The authors performed a cluster analysis on subjects’ sensory profile and ASC symptomology and found that two clusters were characterized by exteroceptive hyper-sensibility, strong attentional focus and exceptional memory. One of these clusters was further characterized by good cognitive, but weak social functioning, as well as by a strong prevalence of repetitive behavior and sensory seeking. Moreover, the authors report a correlation between hyper- and hypo-sensibility measures, indicating that these effects might arise from a common underlying mechanism. These findings suggest that a subset of the autistic population is characterized by an extreme sensibility to exteroceptive stimuli. Overly focused attention, repetitive behaviors and sensory seeking might be compensatory mechanisms which then again lead to hypo-sensibility to socially relevant stimuli outside of the locus of attention.

2.7 Exteroception: summary

The research reviewed in this chapter points towards a tendency for increased locally focused perception without global impairment in ASC in both visual and auditory domains. Moreover, there is ample evidence for a temporal processing deficit in ASC, including: impairment of visual motion coherence at short durations, delay of auditory ERPs, impairment of auditory temporal order judgments, enlarged temporal binding windows of

multisensory stimuli, and decreased processing of auditory duration deviance.

Another recurrent theme is impaired automatic filtering of sensory information and consequently increased processing of distractors, observed in the visual auditory and audiovisual domain. Decreased filtering might be a result of an increased perceptual capacity of autistic subjects, providing them with the ability to process large amounts of information in parallel. This again might account for the widely replicated superiority of autistic subjects in difficult visual search.

Another common theme observed across modalities is a diminished effect of top-down predictions on the perception of visual, auditory and audiovisual stimuli (Goris et al., 2018; Greene et al., 2019; Noel et al., 2017; Turi et al., 2016; van Laarhoven et al., 2019). However, deviance processing measured by oddball paradigms, which arguably relies on top-down predictions, is not unequivocally impaired in ASC.

Compellingly, research investigating exteroceptive processes in ASC directly linked to GABAergic inhibition yields robust results. First, adaptation to tactile stimuli is dependent on GABAergic lateral inhibition and decreased tactile adaptation effects have been found in ASC. Similarly, autistic individuals adapt less to multisensory asynchrony and to recurring deviant auditory stimuli. Even though the underlying mechanisms of these effects are debated, they might also result from a deficit in lateral inhibition. Second, binocular rivalry is a paradigm often used to test GABAergic inhibition, as it relies on lateral

15 inhibition of neuronal populations encoding the competing stimuli. Higher switch rate and longer periods of mixed percepts are thought to be markers of low GABAergic inhibition (Van Loon et al., 2013), and both effects have been found in ASC. Third, GABAergic inhibition is necessary to produce synchronous neural oscillations, which have been reported to be weakened in ASC, in response to visual, auditory stimuli and tactile stimuli.

3 Interoceptive processing

3.1 Search criteria

This part of the review contains all studies comparing interoceptive accuracy or sensibility in a sample of subjects diagnosed with ASC with an age-matched control group. Interoceptive awareness in ASC has only been assessed in a single study and will thus not be considered in this review. Moreover, due to the scarceness of the literature, studies that relate interoception with autistic traits in healthy populations are included.

Some authors argue that abnormalities in interoception observed in ASC are not linked to autistic traits per se, but rather to alexithymia (Shah, Hall, Catmur, & Bird, 2016), defined as inability to describe or assess one’s emotions. Even though alexithymia is not listed as a symptom of ASC in the DSM or ADOS, it has an estimated 50% co-occurrence rate with ASC, opposed to a prevalence of only 10% in the general population (Berthoz & Hill, 2005; Hill, Berthoz, & Frith, 2004). Arguably, alexithymia could be considered a symptom of ASC rather than a separate condition. Therefore, we additionally included studies

relating interoceptive accuracy or sensibility to alexithymia.

3.2 Interoceptive accuracy

Heartbeat perception

Heartbeats constitute discrete interoceptive signals which are observable by the individual and easily measurable. These properties made them the primary focus of research assessing interoceptive accuracy. In heartbeat tracking tasks (Schandry, 1981), subjects are required to count their heartbeats over different time intervals (typically between 25 and 60 seconds) without taking their pulse. Whereas some studies found evidence that interoceptive accuracy assessed by heartbeat tracking tasks is impaired in autistic adults (Garfinkel et al., 2016; Mul, Stagg, Herbelin, & Aspell, 2018) and children (Nicholson, Williams, Carpenter, & Kallitsounaki, 2019; Palser, Fotopoulou, Pellicano, & Kilner, 2018), others have found no such differences (Failla et al., 2019; Nicholson et al., 2019, 2018; Schauder, Mash, Bryant, & Cascio, 2015; Shah et al., 2016). Similarly, some authors have found that interoceptive accuracy is impaired in adults with alexithymia (Herbert, Herbert, & Pollatos, 2011; Shah et al., 2016), while others did not find such differences (Mul et al., 2018; Nicholson et al., 2019, 2018; Zamariola, Maurage, Luminet, & Corneille, 2018).

In heartbeat discrimination tasks (Katkin, Reed, & Deroo, 1983; Whitehead, Drescher, Heiman, & Blackwell, 1977; Wiens & Palmer, 2001), subjects are asked to judge if a series of auditory or visual stimuli is synchronous or asynchronous to

16 their own heartbeat. Three studies have found no differences in interoceptive accuracy measured by heartbeat discrimination tasks between healthy controls and autistic adults (Garfinkel et al., 2016; Mul et al., 2018) or children (Palser et al., 2018). However, these results have to be interpreted with caution as the number of trials in each of these studies used was far below the recommended number of 40 trials to acquire reliable results (Kleckner, Wormwood, Simmons, Barrett, & Quigley, 2015).

Even though heartbeat perception tasks constitute the standard in the field, there is substantial controversy about their construct validity. It has been argued that heartbeat tracking tasks rather test the accuracy of subjects’ beliefs about their own heartbeat than their actual ability to perceive heartbeats (Brener & Ring, 2016). Confirming this concern, one study manipulated heart rate in a sample with pace makers and found, strikingly, that heartrate estimates did not consistently increase when increasing actual heart rate from 60 to 120 beats per minute (Windmann, Schonecke, Fröhlig, & Maldener, 2019). Moreover, heartbeat tracking studies consistently find that subjects underestimate heartbeats and that interoceptive accuracy is negatively correlated with heart rate (Tsakiris, Ainley, Pollatos, & Herbert, 2019). While some authors argue that this undermines the validity of the paradigm (Zamariola, Maurage, et al., 2018), others argue that underestimation is a logical consequence of any signal detection tasks and that interoceptive perception necessarily depends on individual

physiological properties, in this case heart rate (Tsakiris et al., 2019). Concerning the heartbeat discrimination task, it has been argued that accuracy depends on the temporal delays of heartbeats and heartbeat sensations, which differ between individuals (Brener & Ring, 2016). Interestingly, even though they are thought to measure the same ability, accuracy scores for heartbeat tracking and discrimination tasks do not always correlate in the reported studies (Mul et al., 2018; Palser et al., 2018), but see (Garfinkel et al., 2016). These arguments call out for refinement and standardization of existing methods as well as for development of novel paradigms to assess interoceptive accuracy.

Novel interoceptive tasks

One study investigated the relation of autistic traits and alexithymia with subjects’ ability to estimate the speed of their breath (Murphy, Catmur, & Bird, 2018), requiring subjects to reproduce an air blow of a certain speed. The authors compared performance of a healthy sample in two conditions, one in which exteroceptive auditory information was available to subjects and one where it was blocked. Interoceptive accuracy was defined as the difference in accuracy of subjects’ estimate of the speed of the air blow between the exteroceptive and interoceptive condition. Alexithymia, but not autistic traits, predicted a weaker reliance on internal information. However, one might argue that the tactile sensation of air flow is still present in the interoceptive condition

17 and the paradigm thus does not strictly measure an interoceptive signal.

The same authors investigated the association of muscle effort estimation with autistic traits and alexithymia in a sample of healthy subjects (Murphy, Catmur, et al., 2018). The results indicate that alexithymia, but not autistic traits, correlate negatively with muscle effort estimation.

Finally, one study tested accuracy of physiological arousal estimation in a sample of autistic adults and a control sample (Gaigg, Cornell, & Bird, 2018), by comparing subjective arousal ratings and skin conductance responses to emotionally arousing pictures. The authors found that alexithymia, but not autistic traits, predict a lower correspondence of subjective and objective arousal measures. One might argue that this result is not surprising, as arousal is a dimension of emotion and it is not clear which interoceptive signal is perceived when estimating arousal.

3.3 Interoceptive sensibility

Along with interoceptive accuracy, a range of studies investigated the relation of subjective interoceptive sensibility with ASC and alexithymia, yielding conflicting results. One study found evidence for increased interoceptive sensibility in autistic adults (Garfinkel et al., 2016). The authors further report that autistic subjects score higher on interoceptive sensibility relative to interoceptive accuracy, whereas the opposite effect was found in the control group. Conversely, other studies found decreased interoceptive sensibility in autistic

adults (Fiene & Brownlow, 2015; Mul et al., 2018) and children (Palser et al., 2018), as well as a negative association of interoceptive sensibility with alexithymia (Mul et al., 2018; Zamariola, Vlemincx, Corneille, & Luminet, 2018).

This apparent discrepancy might be resolved when considering the different questionnaires used to assess interoceptive sensibility. Garfinkel and colleagues (2016) as well as Palser and colleagues (2018) applied the Body Perception Questionnaire (BPQ; Porges, 1993), which assesses the frequency of being aware of pathological or non-pathological bodily sensations, such as swallowing or stomach pain. The other studies used the Multidimensional Assessment of Interoceptive Awareness (MAIA; Mehling et al., 2012) or the Body Awareness Questionnaire (BAQ; Shields, Mallory, & Simon, 1989), both of which assess the belief in accurate interpretation and control of body signals, using statements such as ‘I

notice distinct body reactions when I am fatigued’

or ‘I can use my breath to reduce tension’. Crucially, frequent perception and accurate interpretation of body signals are distinct processes. Therefore, arguably, only the BPQ measures interoceptive sensibility as introduced by the Garfinkel model of interoception (2015).

3.4 Interoception: summary

Findings from heartbeat tracking tasks indicate that ASC and alexithymia may be associated with decreased interoceptive accuracy, even though many studies failed to find that effect, speaking to the noisy nature of heartbeat signals, imperfect

18 methods to measure of interoceptive accuracy and/or high individual variability. Results from novel interoceptive tasks suggest that an impairment of interoceptive accuracy is linked to alexithymia, rather than to autistic traits. Even though interoceptive tasks aim to measure pure interoceptive perception, one might argue that any interoceptive perception is infused with cognitive information (such as the beliefs about one’s physiological state) as well as exteroceptive information (such as the tactile sensation of heartbeats on the skin). Interoceptive perception might thus be interpreted as an inherently integrative process

Studies investigating interoceptive sensibility suggest that the salience of body signals is decreased during development in autistic children, possibly leading to decreased ability to interpret and control body states later in life. This again might result in inefficient regulation and thus increased salience of bodily signals, in line with findings of impaired autonomic regulation in autistic populations (Benevides & Lane, 2013).

In sum, even though research on interoception in autistic populations is still in its early phase, results indicate a disruption of interoceptive processing and homeostatic control in a subgroup of autistic individuals, possibly those with co-occurring alexithymia. This disruption is marked by a decreased ability to perceive, interpret and control interoceptive signals, possibly leading to increased occurrence of salient interoceptive signals.

4 Integration of interoceptive and

exteroceptive processing

4.1 Behavioral tasks

To our knowledge, only one study explicitly addressed the integration of interoceptive and exteroceptive information in ASC (Noel, Lytle, et al., 2018). The authors measured temporal binding windows using audio-visual, visuo-tactile and cardio-visual stimuli and found that temporal binding windows are enlarged for all stimulus types in the autistic sample. Strikingly, the authors report a 4-fold enlargement for cardio-visual stimuli, an effect which is several orders of magnitude larger than for visuo-tactile and audio-visual stimuli. They furthermore report that accuracy in the cardio-visual task correlated strongly with interoceptive accuracy assessed by heartbeat tracking. These results point towards a major disruption of the integration of interoceptive and exteroceptive information in ASC, exceeding the disruption of multisensory exteroceptive integration.

Importantly – even though meant as measures of interoceptive accuracy – classical cardio-visual heartbeat discrimination tasks also require subjects to integrate interoceptive and exteroceptive information. However, the cardio-visual processing task applied by Noel and colleagues (2018) contains important differences to classical heartbeat discrimination. First, subjects monitored their pulse using their thumb on their wrists. Second, each trial consisted of a single exteroceptive stimulus, rather than a train of stimuli, leading to a greatly increased number of

19 trials. Third, exteroceptive stimuli were visual flashes, rather than tones. Fourth, the authors used various SOAs, enabling them to estimate temporal binding windows. Even though there is a need for further replication in larger samples, the results indicate that the method applied might constitute a valid approach to assess the integration of interceptive and exteroceptive information.

4.2 Anterior insula connectivity

Anterior insula (AI) has been identified as central region for the subjective experience of interoceptive signals (Craig, 2003). AI is activated when subjects pay attention to their own heartbeat (Critchley, Wiens, Rotshtein, Öhman, & Dolan, 2004; Schulz, 2016; Zaki, Davis, & Ochsner, 2012) and it tracks subjects’ subjective experience of temperature (Craig et al., 2000). Moreover, anterior insula receives input from exteroceptive sensory regions (Sterzer & Kleinschmidt, 2010), dopaminergic midbrain regions and amygdala encoding hedonic value (Menon & Levitin, 2005), as well as prefrontal cortex (PFC) encoding motivational and cognitive information (Eckert et al., 2009). This led to the hypothesis that anterior insula functions as an information hub, where exteroceptive and interoceptive information is combined with hedonic and cognitive factors to create subjective feeling states (Craig, 2009; Uddin, 2015). AI has also been shown to be part of a salience network, including dorsal ACC, DLPFC, superior temporal pole, amygdala, and the dopaminergic midbrain (Seeley et al., 2007). This

network is thought to prescribe salience to internal or external stimuli and thus mediate between default mode network and executive network (Goulden et al., 2014; Sridharan, Levitin, & Menon, 2008). The presumed role of the AI as an information integration hub makes it a candidate region to investigate the integration of exteroceptive and interoceptive signals in ASC and thus of interest for this review.

Resting state fMRI studies have frequently reported abnormalities in functional connectivity of the AI in ASC. Increased functional connectivity between sensory areas and AI (Green, Hernandez, Bookheimer, & Dapretto, 2016) as well as between AI and other regions of the salience network (Uddin et al., 2013) has been reported. Conversely, other studies have found reduced connectivity between amygdala and AI (Ebisch et al., 2011; von dem Hagen, Stoyanova, Baron-Cohen, & Calder, 2013), as well as PFC and AI (von dem Hagen et al., 2013).

These findings indicate an altered connectivity pattern of the AI in ASC, possibly leading to abnormal attribution of salience and attention to stimuli. Specifically, the reviewed findings suggest a shift of salience towards exteroceptive (mediated by sensory cortices) and away from emotional stimuli (mediated by amygdala) or motivational cognitive information (mediated by PFC).

20

5 Computational and neural

accounts of sensory processing in

ASC

Having outlined the most important findings from exteroceptive and interoceptive processing in ASC, the question arises how these findings are linked. There exists a range of neurobiological, psychological and computational theories which focus on different aspects of autistic perception (see Box 1). We argue that rather than being mutually exclusive, many of these theories are complementary (describing different phenomena within autistic perception) or analogous (describing the same phenomenon on different levels of abstraction or in different sensory domains). Furthermore, we propose that these theories can be combined to two major explanatory frameworks that account for differences in autistic perception and how these lead up to social and behavioral symptoms of ASC. The first we call stimulus binding framework, which posits that a deficiency to bind stimuli into a unified percept is the core characteristic of autistic perception. The second we call the Bayesian

framework, which holds that the main deficiency in

ASC is a failure of perceptual inference. In the following sections, we will unpack both frameworks by elaborating on how they integrate existing theories and account for findings reviewed above.

5.1 The stimulus binding framework

The weak central coherence theory (Happé, 1996; Happé & Frith, 2006) postulates that ASC is characterized by difficulties to process information in its context, leading up to a more locally focused processing style. When applied to exteroceptive perception, weak central coherence can be understood as a global deficit in the efficient and accurate binding of stimuli or stimulus-parts to a unified percept. This can account for performance advantages in tasks where stimuli must be perceived individually, such as visual search (e.g. Joseph et al., 2009) or visual or auditory disembedding and segregation tasks (e.g. Horlin et al., 2016), as well as for a local processing style (e.g. Wang et al., 2015). Moreover, it can also account for the reduced ability to group stimuli based on features which is reflected in reduced auditory filtering based on frequency or spatial location (Bouvet et al., 2013; Lepistö et al., 2009; Teder-Sälejärvi et al., 2005). In the multisensory domain, a deficit in stimulus binding accounts for weaker binding of visual and auditory stimuli in the McGurk illusion (e.g. Bebko et al., 2014), the lack of multisensory facilitation, as stimuli are processed individually (Brandwein et al., 2013; Ostrolenk et

21 al., 2019), and a less accurate perception of synchrony, reflected by increased multisensory binding windows (Zhou et al., 2018). One might argue that a deficiency in stimulus binding is reflected in diminished interference between competing stimuli, as seen in the tactile domain (Tommerdahl et al., 2008). This might lead to processing advantages, such as higher perceptual capacities and advantages in parallel processing (Remington & Fairnie, 2017; Remington et al., 2009, 2012).

In the interoceptive domain, there is evidence for disrupted integration of stimuli as

well. Studies point to dysfunctional integration of interoceptive and exteroceptive information in ASC (Noel, Lytle, et al., 2018). Moreover, it can be argued that in order to accurately interpret and predict interoceptive signals, these must be combined with exteroceptive stimuli from the environment. If binding of external and internal signals goes astray during development, this might lead to the observed pattern of increased salience but reduced ability to interpret interoceptive signals in autistic adults (Fiene & Brownlow, 2015; Garfinkel et al., 2016; Mul et al., 2018). Moreover, in order to correctly identify one’s emotional

Box 1. Theories of autistic perception

Excitation-inhibition balance. Decreased GABA levels in autistic individuals lead to a global deficit in neural inhibition (Robertson et al., 2016).

Neural synchronization hypothesis. ASC are characterised by a general failure to produce synchronized neural oscillations, leading to an inability to bind stimuli (Simon & Wallace, 2016).

Under-connectivity hypothesis. Long-range under-connectivity between frontal and posterior regions of the brain underlies the autistic phenotype (Just et al., 2012).

Intense world hypothesis. Increased local neural connectivity and plasticity leads to hyper-perception and -attention in ASC (Markram 2007).

Weak central coherence. ASC are characterized by an inability to process information in its context, leading up to a locally focused processing style (Happé & Frith, 2006).

Perceptual enhancement. Low level perceptual abilities are enhanced in ASC (Mottron et al., 2001).

Attentional over-focus. An intense attentional focus and difficulties to disengage lead to the emergence of autistic symptoms (Keehn, Müller, et al., 2013) .

Bayesian account. Attenuation of prediction errors in the exteroceptive (Lawson et al., 2014) and interoceptive domain (Quattrocki & Friston, 2014) is decreased in ASC due to neuromodulator dysfunction. This leads to an inadequate generative model.

22 states, interoceptive and exteroceptive information needs to be combined. Therefore, a reduced ability to bind these types of stimuli together could account for the high occurrence of alexithymia within the autistic population (Berthoz & Hill, 2005).

Despite the compelling number of findings in line with weak central coherence, the theory can be criticized for being descriptive, rather than explanatory. However, there are several neural mechanisms observed in ASC that might underlie a deficiency in stimulus binding.

Neural synchronization hypothesis

It has been argued that the neural mechanism underlying the global deficit in stimulus binding is a deficiency to synchronize neural oscillations (Brock et al., 2002; Simon & Wallace, 2016). Gamma range (30-120 Hz) synchrony between neural ensembles encoding different stimulus properties is thought to be a mechanism crucial for sensory integration (Rodriguez et al., 1999) and thereby the brain’s solution to the binding problem. Local synchronization of nearby ensembles subserves unisensory feature binding, whereas long distance synchronization of neural ensembles located in different brain regions enables multisensory binding. Due to biophysical constraints, long distance neural synchronization is restricted to frequencies below the gamma range. However, synchronization of gamma waves between brain regions can be achieved by hierarchical cross-frequency coupling, where lower frequency oscillations drive higher frequency

oscillations (Simon & Wallace, 2016). Hence, in order to achieve successful sensory integration between brain regions, local synchronization of gamma waves, long distance synchronization of lower frequencies, as well as local cross-frequency coupling need to be intact.

Deficiencies in two out of three of these mechanisms have been found in ASC. Research has consistently identified a deficiency to produce locally synchronized gamma range activity in response to visual, auditory and tactile stimuli (e.g. Edgar, Khan, et al., 2015), as well as entrained responses to periodic stimuli (Khan et al., 2015; Wilson et al., 2007). Conversely, induced oscillatory power in response to stimuli seems to be preserved or increased (Rojas et al., 2008). This suggests that the timing of gamma waves relative to a stimulus is impaired in ASC, while gamma wave production remains intact. Moreover, applying visual or auditory stimuli to a deficiency in long-range synchronization, reflected by decreased interhemispheric (e.g. Peiker, David, et al., 2015) and parietal-temporal coherence (Righi et al., 2014) of beta and gamma waves. To date, there is no evidence for altered cross-frequency coupling in response to sensory stimuli in ASC.

Interestingly, during resting state and sustained visual attention, neural synchronization across the delta, theta, alpha and gamma frequencies (Cornew, Roberts, Blaskey, & Edgar, 2012; Edgar, Khan, et al., 2015; Orekhova et al., 2007) as well as cross-frequency coupling between alpha and gamma waves is increased in ASC (Berman et al., 2015). Thus, the core deficiency in

23 ASC seems to pertain to the temporal precision of synchronizing neural activity to the onset of a stimulus, possibly due to a noisier baseline state that is not tuned for incoming stimuli (Edgar, Khan, et al., 2015), and not the production of neural oscillations per se. Arguing along these lines, the limited ability to produce neural oscillations to external stimuli with temporal precision accounts for temporal processing deficits observed in ASC, such as delayed auditory ERPs (e.g. Edgar, Khan, et al., 2015) and impaired temporal order judgments (Kwakye et al., 2011).

Crucially, a deficit in precise temporal processing and accurate binding of auditory and visual information might give rise to language impairments often observed in ASC (Stevenson et al., 2018).

Excitation-inhibition disbalance

On the molecular level, a deficit in GABAergic inhibition can account for abnormalities in neural synchronicity observed in ASC. It has been shown that GABAergic inhibition is necessary to produce synchronized gamma waves in response to sensory stimulation (Sohal, Zhang, Yizhar, & Deisseroth, 2009), but that downregulation of interneuron inhibition suppresses resting-state gamma oscillations (Saunders et al., 2013), which mirrors the pattern observed in ASC. There is ample neurobiological evidence for increased GABA levels in ASC. Post-mortem and in vivo MRS studies have found increased GABA receptor expression in frontal, sensory and motor cortices in ASC (Gaetz et al., 2014; Harada et al., 2011; Oblak, Gibbs, &

Blatt, 2010; Puts et al., 2017; Sapey-Triomphe et al., 2019), pointing towards decreased inhibition in the autistic brain. Furthermore, increased blood and cortical glutamate levels (Brown, Singel, Hepburn, & Rojas, 2013; Joshi et al., 2013; Shinohe et al., 2006) and the high prevalence of comorbid epilepsy in ASC (Lee, Smith, & Paciorkowski, 2015) points towards increased excitation. Moreover, genes relating to the function of inhibitory interneurons (Marin, 2012) and glutamate receptors (Lee, Choi, & Kim, 2015) are associated with ASC.

Next to its effect on neural synchrony, an increased excitation-inhibition ratio can account for multiple other findings from the exteroceptive domain that are unrelated to stimulus binding, as outlined above. These include diminished adaptation effects to tactile stimuli (e.g. Puts et al., 2017), decreased switch rates and longer mixed percepts in binocular rivalry (Robertson et al., 2013). Possibly, reduced adaptation to multisensory asynchrony (Noel et al., 2017; Turi et al., 2016) and to recurring auditory deviants (Goris et al., 2018; Hudac et al., 2018) might also be accounted for by reduced lateral inhibition. Behavioral and biological evidence directly related to a GABAergic deficit in ASC is the most widely replicated and least disputed, indicating that an altered excitation-inhibition ratio might be a key mechanism underlying sensory abnormalities associated with the autistic phenotype (Robertson, Ratai, & Kanwisher, 2016; Rubenstein & Merzenich, 2003).