Autism, oxytocin and interoception

Abstract

Autism is a pervasive developmental disorder characterized by profound social and verbal communication deficits, stereotypical motor behaviors, restricted interests, and cognitive abnormalities. Autism affects approximately 1% of children in developing countries. Given this prevalence, identifying risk factors and therapeutic interventions are pressing objectives—objectives that rest on neurobiologically grounded and psychologically informed theories about the underlying pathophysiology. In this article, we review the evidence that autism could result from a dysfunctional oxytocin system early in life. As a mediator of successful procreation, not only in the reproductive system, but also in the brain, oxytocin plays a crucial role in sculpting socio-sexual behavior. Formulated within a (Bayesian) predictive coding framework, we propose that oxytocin encodes the saliency or precision of interoceptive signals and enables the neuronal plasticity necessary for acquiring a generative model of the emotional and social 'self.' An aberrant oxytocin system in infancy could therefore help explain the marked deficits in language and social communication—as well as the sensory, autonomic, motor, behavioral, and cognitive abnormalities—seen in autism.

Keywords: Active inference; Autism; Bayesian predictive coding; Emotional affordance; Interoception; Neuromodulation; Oxytocin; Self-awareness; Sensory attenuation.

Fig. 1

Predictive coding. This schematic introduces predictive coding in terms of exteroceptive, proprioception and interoception: Exteroception uses primary senses such as sound, light, and discriminatory touch to build models or precepts of the external world. Proprioception processes proprioceptive and kinesthetic information from the body to allow movement and provide a sense of agency. Interoception allows models of the internal “self” to be constructed and emotional feelings to be inferred through visceral sensations such as temperature, stretch and pain from the gut, light (sensual) nondiscriminatory touch, itch, tickle, hunger, nausea, thirst, sleepiness, and sexual desire. Left panel: This illustration portrays the brain as an inference machine, using predictive coding to form perceptual representations (inferences) of the environment from light signals by extracting inherent structure from visual sensory information. In this predictive coding, visual signals are transmitted up the hierarchical processing pathway from the eyes (via the retina, optic nerve, superior colliculi, and lateral geniculate) first to primary visual sensory cortex and then to higher cortical regions that instantiate perceptual inference. The curved arrows in the figure symbolize the iterative and recursive nature of predictive coding. This illustration simplifies the hypothesis testing—that would normally occur with reference to generative models at multiple levels in the central nervous system into one loop. In the brain, neuronally encoded expectations or explanations for sensory input generate predictions that are transmitted to lower layers and ultimately sensory cortex. In hierarchical predictive coding, each level is trying to predict the level below by sending top-down predictions to form prediction errors that are then returned to improve the higher predictions. The implicit comparison of top-down predictions and bottom-up evidence allows for ascending transmission of only the relevant information (prediction error) that violates the top down predictions. These violations (prediction errors) travel up the hierarchy to inform higher cortical levels. The expected precision (inverse variance) of the prediction errors determines the degree to which these signals influence representations at higher levels. Middle panel: This depiction of embodied (active) inference, illustrates how action itself becomes the fulfillment of a forward model (prediction). The teal arrow represents the signal attenuation, in the form of corollary discharge that occurs when movements, or behaviors, are self-generated. In the case of movement, motoric predictions constitute descending signals that suppress ascending proprioceptive prediction errors that would otherwise prevent the fulfillment of the prediction. Successful attenuation, and therefore movement, largely depends upon the proper construction and utilization of predictions about the physical ‘self’. Right panel: The equivalent action in the interoceptive domain is the fulfillment of an internal prediction or drive, either by adjusting the autonomic response, or by initiating a behavior. The interoceptive equivalent to throwing a disc – as depicted in the middle panel – might be fulfilling the prediction of thirst. Prediction errors will drive either an autonomic reaction that will deploy a physiological response such as vasoconstriction or water resorption in the kidney, while at the same time boosting the affordance of water and prescribe the behavior that will lead to drinking, thus fulfilling top-down predictions. Because these actions are self-generated, they require sensory attenuation that might otherwise subvert the execution of either the behavior or the homeostatic adjustment. Our focus in this paper is on the attenuation of interoceptive prediction errors—or interoceptive attenuation. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 2

Active inference and sensory attenuation. This illustrates how sensory attenuation in active inference supports social interactions. Like action observation, social interactions require attenuation, not of proprioceptive information, but of interoceptive signals. In a predictive coding formulation of social engagement, interaction induced suppression of interoceptive signals, similar to motor induced suppression of proprioceptive signals, allows for the individual to infer the autonomic and emotional state of another, without eliciting the autonomic reflexes in self—in other words, the display of cognitive empathy without autonomic echopraxia or emotional contagion. The ability of interoceptive directed predictions to suppress otherwise inappropriate interoceptive prediction errors depends upon a complex generative model of the socio-emotional ‘self’. As explained in the text, the inferential learning necessary to create such a generative model and to deploy the appropriate predictions that emerge from the model, we argue, requires oxytocin-dependent neuronal plasticity during development.

Fig. 3

Oxytocin and the development of emotional affordance. This schematic describes normal (and autistic) neurodevelopmental trajectories, in terms of (simplified) neural architectures underlying predictive coding of autonomic (emotional) signals. The three panels illustrate the development of associative connections we imagine underlie the acquisition of emotional responses during three stages of development. The anatomical designations should not be taken too seriously—they are just used to illustrate how predictive coding can be mapped onto neuronal systems. In all of these schematics, red triangles correspond to neuronal populations (superficial pyramidal cells) encoding prediction error, while blue triangles represent populations (deep pyramidal cells) encoding expectations. These populations provide descending predictions to prediction error populations in lower hierarchical levels (blue lines). The prediction error populations then reciprocate ascending prediction errors to adjust the expectations (red lines). Arrows denote excitatory connections, while circles denote inhibitory effects (mediated by inhibitory interneurons). Left panel: in the first panel, connections are in place to mediate innate (epigenetically specified) reflexes – such as the suckling reflex – that elicit autonomic (e.g., vasovagal) reflexes in response to appropriate somatosensory input. These reflexes depend upon high-level representations predicting both the somatosensory input and interoceptive consequences. The representations are activated by somatosensory prediction errors and send interoceptive predictions to the hypothalamic area—to elicit interoceptive prediction errors that are resolved in the periphery by autonomic reflexes. Oxytocin is shown to project to the high-level representations (the amygdala) and the hypothalamic area, to modulate the gain or precision of prediction error units. In this schematic, its effects are twofold: oxytocin attenuates the gain of hypothalamic prediction error units and augments the gain of higher level units. Middle panel: this shows the architecture after associative learning, during which high-level representations in the anterior cingulate or insular cortex have learned the coactivation of amygdala representations and exteroceptive cues (e.g., the mother's face during suckling). These high-level representations now predict the exteroceptive visual input and (through the amygdala) somatosensory and autonomic consequences. Right panel: in this schematic, visual input (e.g., the mother's face) is recognized using the high-level representation in the anterior insular or cingulate cortex. However, in this case, interoceptive prediction error is attenuated so that it does not elicit an autonomic response. In other words, although the high-level emotional representation is used to recognize exteroceptive cues, lower-level transcortical reflexes are inhibited. In autism, we presume that oxytocin is deficient, such that sensory attenuation is impaired – leading to disinhibition of autonomic responses and the failure to recognize a mother's face in any other context – other than during suckling. This failure of sensory attenuation may underlie autonomic hypersensitivity, failure of emotional recognition, attention to emotional cues, theory of mind and central coherence. The dotted green line in this figure acknowledges that there may not be any direct projections from the origin of oxytocin cells (in the supraoptic and paraventricular nuclei of the hypothalamus) to secondary or primary somatosensory cortex. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 4

Oxytocin dependent neuromodulation compared to dopamine. This schematic uses the same format as Fig. 3 to illustrate the similarities between the role of oxytocin in modulating the precision of interoceptive signals and the role of dopamine in modulating proprioceptive signals. Again, the anatomy and details should not be taken too seriously; however, the overall architecture depicted here emphasizes their similarity – in terms of opposing neuromodulatory roles that could increase the precision of higher-level prediction errors, while decreasing the precision of lower-level prediction errors. For oxytocin, we have illustrated this by an augmentation of gain in the amygdala and an attenuation of sensory precision in the hypothalamic region. For dopamine, the same complementary effects are illustrated in the dorsal and ventral striatum, mediated by D1 (go pathway) and D2 (no-go pathway) receptors, respectively. In both cases, context sensitive neuromodulatory signaling is listed by descending projections (predictions) from cortical regions involved in emotional regulation (oxytocin) or action selection (dopamine). Deficits of oxytocin – we propose – lead to a failure of interoceptive processing, while deficits of dopamine (for example in Parkinson's disease) compromise proprioceptive processing and the initiation of action.