Introduction
Interoception refers to the sensation and representation of internal physiological signals, from organs such as the heart, stomach, lungs, and skin (Craig
2003). Its importance for survival is clear, but it is also increasingly seen as central to the development and organisation of higher-level cognition. For example, interoceptive accuracy (IA; the extent to which one can accurately identify interoceptive signals) has been found to be associated with ability in the domains of emotion-processing (Barrett et al.
2004; Pollatos et al.
2005), decision-making (Dunn et al.
2010), self-regulation (Herbert et al.
2007), empathy (Fukushima et al.
2011) and theory of mind (Shah et al.
2017). These wide ranging associations are in line with theories of embodied cognition that imply cognition is situated in bodily systems, and also point towards a role for interoception in both self and other processing (Gallese
2007; Barrett and Simmons
2015; Barrett
2017). This has led to a growing interest in the importance of interoception for disorders such as autism spectrum disorder (ASD).
ASD is a neurodevelopmental disorder which is characterised by severe behavioural impairments in social-communication and behavioural flexibility (American Psychiatric Association
2013). At the cognitive level, ASD is characterised by a difficulty with representing others’ mental states (theory of mind; Yirmiya et al.
1998) and with one’s own mental states (metacognition; e.g., Brosnan et al.
2016; Grainger et al.
2014; McMahon et al.
2016). Given recent findings linking interoception with higher level cognitive processes, such as theory of mind (Shah et al.
2017) and self-regulation (Herbert et al.
2007), it has been argued that difficulties with self-other processing in ASD may be the result of an underlying impairment in interoceptive awareness. Quattrocki and Friston (
2014) suggest that abnormal regulation of the oxytocin (a hormone linked to social bonding) system in ASD may represent the biological basis of interoceptive impairments, which in turn result in problems linking more abstract concepts associated with these bodily signals. The idea is that a difficulty interpreting one’s own internal bodily signals early on may interfere with learning about the association between these low level bodily signals and other higher level feelings and thoughts, restricting the comprehension of oneself, which in turn would have similar effects on understanding other selves. However, empirical studies of interoception in ASD have produced decidedly mixed results.
The majority of studies assessing interoception in ASD have used the heartbeat tracking task (Schandry
1981). In this task participants are asked to count the number of times they feel their heart beat in a given time interval without touching their body (i.e., feeling their pulse). The closer the participant’s reported number of heart beats is to their actual number of heart beats, the greater their interoceptive accuracy (IA). Six studies have used the heartbeat tracking task to measure interoceptive accuracy among individuals with ASD and findings are summarised in Table
1. In short, while diminished accuracy was observed in two studies, four studies found only negligible and non-significant between-group differences suggesting that interoceptive accuracy is not impaired in ASD. One plausible explanation for the mixed findings is proposed by the alexithymia hypothesis (Bird and Cook
2013), which posits that difficulties processing emotions are only evident in ASD individuals who also have high levels of alexithymia (defined as the inability to identify or describe ones emotions; Taylor
1984). According to this view interoceptive abnormalities are a central component of alexithymia, not autism, and the suggested link between ASD and interoception can be explained by the high comorbidity of ASD and alexithymia (approximately 50% of ASD individuals also have alexithymia, five times that of the general public; Berthoz and Hill
2005), as opposed to ASD itself. Therefore, mixed results in the literature might emerge because levels of alexithymia vary from study to study, with only studies including ASD individuals with high levels of alexithymia likely to show a difference in interoception (Shah et al.
2016).
Table 1
Details of previous studies of interoceptive accuracy in ASD
| Children 21 ASD, (M age: 12.3, SD 2.8), 24 NT (M: 11.5, SD 2.5), matched for age and IQ | ASD = NT No group difference | Visual counting task: a subset of children, 8 TD (33%) and 12 ASD (57%), counted dim visual stimulus over same interoception time intervals. Performance did not differ between groups | Not measured |
| Adults 19 ASD (M: 32.9, SD 11.5) 19 NT, (M: 32.9, SD: 14.4), matched for age, gender and IQ | ASD = NT No group difference | Time estimation task: All participants were instructed to judge the duration of three randomized intervals. No correlation between time estimation and IA | Negative correlation between Alexithymia and IA in NT (exp1) and ASD and NT (exp2) |
| Adults 20 ASD (M 28.1, SD 8.8), 20 NT (M 27.8, SD 3.4), matched for age and gender, but IQ not measured |
ASD < NT (d = 1.10)
ASD significantly lower IA
| None | Not measured |
| Children 52 ASD (M 12.2, SD 2.9) 42 NT, (M 11.03, SD 2.8) and Adults 19 ASD (M: SD), 19 NT (M: SD:), matched for age and gender, significantly differ on IQ, both across whole sample and within age groups | ASD = NT When groups split by IQ (over or under 115) for the groups under 115 the NT’s showed positive correlation between age and IA (r = 0.539) while ASD showed negative correlation (r = − 0.339) | Visual counting task: all adults, and 28 ASD (54%) and 21 NT (50%) children counted frequency of visual stimulus. NT significantly better (p = 0.004) and significant correlation between IA and task for NT’s only (r = 0.35) | Not measured |
| Children 30 ASD, (M: 12.5, SD 2.9), 30 NT (M: 11.9, SD 3), matched for age, VIQ, PIQ, and FSIQ |
ASD < NT (d = 1.09)
ASD significantly lower IA
| Auditory counting task: 12 ASD (40%) and 17 NT (57%) participants had to silently count auditory-presented tones that numbered 5, 20 or 33. No group differences on task. No other comparisons reported | Not measured |
| Adults 46 ASD (M:40.2 SD 11.7), 48 NT (M: 41.2, SD 12.6), matched for age, sex, VIQ, PIQ, and FSIQ | ASD = NT No group difference | None | ASD Hi (> 60), n = 27, and low (< 60), n = 19, in Alexithymia did not differ in IA; No correlation between IA and Alexithymia |
A recent study tested this hypothesis directly comparing a group of autistic adults who manifested clinically significant levels of alexithymia (n = 27) with an age, sex, IQ, and ASD severity-matched group of autistic adults who did not meet criteria for alexithymia (n = 19) (Nicholson et al.
2018). Crucially, there was no hint of any difference between these two sub-groups of ASD participants in terms of interoceptive accuracy, which provided clear evidence against the alexithymia hypothesis. One of the potential limitations of this recent study, however, was the lack of a control task to guard against alternative explanations for the ASD group’s undiminished performance on the heartbeat tracking task. One potential difficulty with the heartbeat tracking task is that it might be possible to achieve apparently good interoceptive accuracy on it by counting silently. Given that the average heart rate is around 60 bpm, simply counting seconds internally could be sufficient to achieve high interoceptive accuracy even if one was unable to detect one’s heartbeat (Brener and Ring
2016). Therefore, in the current study we adopted a version of a time estimation control task (employed by Shah et al.
2016), which required participants to count silently during several unspecified periods and then estimate how many seconds had passed during the time. If participants with ASD (or neurotypical comparison participants) were compensating for underlying difficulties with interoception by silently counting, then performance on the time estimation control task should be associated significantly with performance on the heartbeat tracking task and group differences on the latter should appear once time estimation is controlled for.
A second more general limitation with the field as a whole concerns the reliance on the heartbeat tracking task as a single measure of interoceptive accuracy. The task is used frequently because it is relatively easy to administer, has good test–retest reliability (Mussgay et al.
1999), and is sensitive to individual differences (Christensen et al.
2018; Dunn et al.
2010; Garfinkel et al.
2015). However, some have questioned whether interoception is a unitary construct and suggested that it actually involves multiple independent systems, which may not be causally linked (Garfinkel et al.
2016a). If this latter view is correct, then it may be that individuals with ASD show diminished interoceptive accuracy on one domain, but not another domain. Experiment 1 addresses these issues.
In Experiment 1, we gave 21 adults with ASD and 21 NT age- and IQ-matched comparison adults the classic heartbeat tracking task to measure cardiac interoception, as well as a version of a task used by Murphy et al. (
2018) to test respiratory interoceptive accuracy. We included control tasks designed to assess the extent to which alternative, compensatory strategies might be used by individuals with ASD to succeed on the two interoception tasks in the absence of underlying interoceptive competence. We also administered measures of mindreading, depression and anxiety to allow full characterisation of the sample and for the exploration of underlying cognitive/personality correlates of interoceptive accuracy. We predicted that adults with ASD would be unimpaired on both the cardiac and respiratory interoception tasks and that this would reflect truly undiminished interoceptive accuracy (rather than compensatory strategy use).
General Discussion
Taken together, the results from experiments 1 and 2 add to the growing literature on interoception in ASD and provide a potential developmental explanation for how interoception changes across the lifespan in this condition. The findings partly support theories that reference the importance of altered interoceptive processing in the development of the ASD phenotype (Quattrocki and Friston 20
2014). However, the results also suggest that interoceptive difficulties in childhood may resolve by adulthood in people with ASD (unlike difficulties with mindreading and emotion processing, which remain impaired across the lifespan in ASD).
In Experiment 1, we replicated previous research findings of no cardiac interoceptive impairment in ASD adults (Shah et al.
2016; Mash et al.
2017; Nicholson et al.
2018), irrespective of the extent to which participants manifested alexithymic traits (cf. Nicholson et al.
2018). While one previous study (Garfinkel et al.
2016b) found an impairment in ASD adults, this study did not match groups for IQ, which leaves open the possibility that the finding was the result of between-group differences in general cognitive ability, rather than between-group differences in diagnostic status. While the current study found a clear medium-to-large cardiac interoceptive impairment in ASD children (
d = 0.72), which is in line with another recent study by Palser et al. (
2018), two other studies of cardiac interoception in children with ASD have observed no such impairment (Schauder et al.
2015; Mash et al.
2017). One issue to consider here is that, as highlighted by an anonymous reviewer of the manuscript, the average interoceptive accuracy of neurotypical children in Experiment 2 was higher than the average interoceptive accuracy of neurotypical adults in Experiment 1. While post hoc analysis showed that this difference was non-significant,
t = 1.71,
p = 0.10,
d = 0.52, BF
10 = 0.95, the fact that neurotypical children were even numerically superior to neurotypical adults is somewhat surprising. Perhaps these control children were unrepresentatively able, which produced artificial group differences in interoceptive accuracy that would not have been observed if “super controls” had not been employed. This is a possibility, but we guarded against this by ensuring the groups were matched for background variables, such as IQ and time estimation ability, which might have influenced interoceptive accuracy. The group difference in interoceptive accuracy remained significant after age, IQ, time estimation ability, and average heart rate were controlled in an ANCOVA. If the control children showed higher levels of interoceptive accuracy because they were super controls, then we might reasonably have expected to see group differences reduced or eliminated after controlling for other relevant variables, but this did not occur. When considering all these findings together, one interpretation is that there is heterogeneity in the cognitive profile of children with ASD and that cardiac interoception is impaired in only some children. However, whether such an impairment represents a core cognitive cause of ASD features is a matter for debate.
A number of theories have argued that emotional processing relies on inferring the causes of interoceptive signals (Wiens
2005; Seth
2013; Barrett
2017), and neuroimaging studies of the link between interoception and emotion provide some evidence in favour of this view (Critchley et al.
2004; Zaki et al.
2012). This fits when considering ASD, given that previous research has highlighted that difficulties in identifying and understanding emotions both in oneself and in others is a common cognitive-level difficulty in this disorder (Gaigg
2012; Uljarevic and Hamilton
2013; but see Williams and Happé
2010). If interoception is linked with emotion-processing development in neurotypical individuals, then early impairments interoception among children with ASD may underpin later (and persistent) difficulties with emotion-processing. In other words, it may be that a “decoupling” of interoception and emotion processing among some children with ASD results in emotions never being fully anchored within the body, making emotions difficult to understand in self and others across the lifespan even once interoception difficulties have resolved.
A related possibility concerning decoupling is that local interoceptive signals are not integrated together in a global sense, therefore restricting their influence on motivational and behavioural states which drive subsequent goal-directed action (Hatfield et al.
2019). This fits the pattern of weak central coherence often observed in people with ASD (Happé and Frith
2006), and suggests that interoceptive signals may be attended to by people with ASD in a more narrowly detail-focused manner than neurotypical people tend to attend to them. From this perspective the decoupling of interoceptive processing may not be specific to its links to emotional information per se, but global processing on a more general level, including the inability for interoceptive signals to inform emotional processing, but also that these processes struggle to properly bind with other information sources such as memory, perception and decision making. Future research aimed at measuring interoceptive, emotional and other forms of cognitive processing within the same population could test the plausibility of this account, while also shedding light on the idea that the decoupling may be specific to emotional and interoceptive processing.
Another alternative possibility is that the process of building theories of emotions (or mental states, generally) is impaired in ASD
independent of interoception difficulties (e.g., Carruthers
2009; Williams
2010). This could also explain the findings that emotion-processing difficulties persist in ASD even after interoceptive difficulties have resolved, and also that emotion-processing abilities are not necessarily reliably associated with interoceptive accuracy (see Nicholson et al.
2018; also, the lack of a significant association in the current study). Future research should aim to unpick the developmental trajectories of both emotional and interoceptive processing in children with ASD, and investigate how these trajectories compare to those in neurotypical children.
The current results also speak to other theories such as the alexithymia hypothesis (Bird and Cook
2013), given that neither cardiac nor respiratory interoceptive accuracy was significantly associated with number of alexithymic traits in participants who completed Experiment 1. Moreover, there were no significant differences in interoceptive accuracy between individuals with ASD who scored over the clinical threshold for alexithymia (n = 13) and those who scored under threshold (n = 8). These findings challenge any strong claim that alexithymia and interoception are inextricably linked. A caveat here is that sample size was relatively modest and so drawing conclusions from the results of association (or subgroup) analyses should be done with caution. Nonetheless, the sample size is very similar to the sample size in previous studies that have reported an association between the two abilities (see Table
2). The alexithymia hypothesis is a plausible theory that explains some of the existing findings in the interoception-, emotion processing- and ASD-related literature. The current findings do not speak to all aspects of the theory and there are several aspects that may well be correct. The current findings suggest only that comorbid alexithymia does not appear to influence interoceptive abilities in adults with ASD as the theory predicts it should.
Aside from the contribution of the current results to our understanding of ASD, the results also have implications for theories of the structure of interoception. In particular, the finding that accuracy on the measure of cardiac interoception was not associated significantly with accuracy on the measure of respiratory interoception supports the suggestion that there may not be a unitary interoception faculty that processes all forms of interoceptive input (Garfinkel et al.
2016a). The finding that measures of interoceptive accuracy in different domains are not associated significantly does not show that the outputs from independent low-level interoceptive systems are not combined at a higher level of the processing hierarchy (e.g., at the level of integration or interpretation). However, it does imply that at the level of monitoring these systems are distinct. This can inform futures studies that aim to measure both interoceptive monitoring and integration across multiple domains.
In conclusion, the current study suggests that interoceptive accuracy impairments may be present in children with ASD, but that these impairments resolve over time and are absent by adulthood. This provides a developmental framework for understanding interoception in ASD and suggest future research should focus on how interoceptive and emotional processing relate to each other in children. Equally, it provides motivation for future studies aimed at improving interoception in childhood with the idea that this may have the potential to enhance the link between emotional processing and interoception early on, which could improve emotional processing in ASD through development and into adulthood. These findings add to the growing literature on interoception in ASD and provide insight into strategies for future research.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.