Attenuation of auditory evoked potentials for hand and eye-initiated sounds

Highlights

• Saccade initiation of tones evoked less sensory attenuation than button-presses.

• Motor-to-sensory mapping can be partly generalized to multiple contingencies.

• Predictive and associative processes may both contribute to sensory attenuation.

Abstract

Reduction of auditory event-related potentials (ERPs) to self-initiated sounds has been considered evidence for a predictive model in which copies of motor commands suppress sensory representations of incoming stimuli. However, in studies which involve arbitrary auditory stimuli evoked by sensory-unspecific motor actions, learned associations may underlie ERP differences. Here, in a new paradigm, eye motor output generated auditory sensory input, a naïve action–sensation contingency. We measured the electroencephalogram (EEG) of 40 participants exposed to pure tones, which they produced with either a button-press or volitional saccade. We found that button-press-initiated stimuli evoked reduced amplitude compared to externally initiated stimuli for both the N1 and P2 ERP components, whereas saccade-initiated stimuli evoked intermediate attenuation at N1 and no reduction at P2. These results indicate that the motor-to-sensory mapping involved in speech production may be partly generalized to other contingencies, and that learned associations also contribute to the N1 attenuation effect.

Introduction

Our brains must regulate a continuous stream of sensory input in order to flexibly generate behaviour and allow interaction with the world. A well-established example of such regulation is sensory attenuation, where the sensory input evoked by self-initiated actions is marked by reduced phenomenological (e.g., Blakemore, Frith, & Wolpert, 1999; Cardoso-Leite, Mamassian, Schütz-Bosbach, & Waszak, 2010; Sato, 2008) and neurophysiological representations (e.g., Baess, Jacobsen, & Schröger, 2008; Houde, Nagarajan, Sekihara, & Merzenich, 2002; Schafer & Marcus, 1973) compared to identical, externally initiated sensory input—a phenomenon typified by the difficulty of tickling ourselves (Weiskrantz, Elliott, & Darlington, 1971). Functionally, sensory attenuation serves to conserve attentional resources and to enable sensory processing in situations where volitional actions would otherwise desensitize sensory receptors, such as during speech production (Bendixen, SanMiguel, & Schröger, 2012). It has also been proposed as fundamental for self-identity, such that dysfunctional attenuation could lead to psychotic symptomology (Feinberg, 1978, Ford et al., 2001).

In the auditory domain, the N1 or N1 m component (an evoked potential or magnetic field which appears approximately 100 ms after the onset of an auditory stimulus) is used as a cortical index of sensory attenuation, because its amplitude, compared to externally initiated stimuli, is consistently reduced for both self-initiated vocalizations (Curio, Neuloh, Numminen, Jousmaki, & Hari, 2000; Heinks-Maldonado, Mathalon, Gray, & Ford, 2005; Houde et al., 2002) and button-press-initiated stimuli (Aliu, Houde, & Nagarajan, 2009; Baess et al., 2008; Martikainen, Kaneko, & Hari, 2005; McCarthy & Donchin, 1976; Schafer & Marcus, 1973; Sowman, Kuusik, & Johnson, 2012). The predominant explanation for these findings invokes a theory of motor control in which a forward model predicts future behavioural states and their sensory consequences (Wolpert, Ghahramani, & Jordan, 1995). According to this theory, the sensory consequences of volitional action can be predicted based on an efference copy (i.e., a copy of the motor command), and sensory attenuation reflects the subtraction of this prediction from actual sensory input (Bays & Wolpert, 2007). Conversely, externally initiated sensory input, for which there is no motor information to form an accurate prediction, will remain unmodulated by the efference copy mechanism (see Timm, SanMiguel, Keil, Schroger, & Schonwiesner, 2014).

Forward prediction is strongly implicated in speech production (Hickok, 2012), which involves a well-defined range of motor output (e.g., shape formed by lips) producing specific, habitual, sensory consequences (i.e., uttered syllables). As candidate language fibre pathways have been identified (Dick & Tremblay, 2012), upon which the efference copies of speech motor output conceivably travel, there is also a plausible neural basis for forward prediction in speech production. However, the N1 attenuation literature largely comprises experiments based on arbitrary action–sensation contingencies, usually hand movements to press a button and elicit a tone. This is problematic, because it is not clear that motor-to-sensory mapping can be generalized from speech production to auditory input evoked by motor actions unrelated to speech (Horváth, 2015). Unlike speech-induced auditory attenuation, for which there are distinct neural networks proposed to be involved (Behroozmand et al., 2016; Chang, Niziolek, Knight, Nagarajan, & Houde, 2013; Greenlee et al., 2013) which likely encode specific acoustic properties of the upcoming sound, internal predictions arising from non-speech motor actions might be comparatively more crude. This presents the possibility that sensory attenuation of speech and non-speech stimuli are driven by different mechanisms.

An alternate (or perhaps complementary) explanation of sensory attenuation relates to learned associations between actions and sensations. Operationally, learned associations have been proposed to “pre-activate” potential sensory input, increasing baseline neural activity at a greater rate than the signal increase resulting from an incoming stimulus (Roussel, Hughes, & Waszak, 2013). This account differs from forward prediction in that sensory attenuation is attributed to poorer stimulus discrimination (and consequent reduction of the corresponding sensory representations) rather than an efference copy directly reducing the internal response to a predicted stimulus. An associative explanation can be reasonably applied to the classic paradigm of pressing a button to hear an auditory stimulus, given that such a contingency conceivably draws upon a wealth of pre-existing action–sensation associations that have been experienced and learnt over the course of a lifetime. Indeed, there is evidence to suggest that prior experience can affect sound perception (Repp & Knoblich, 2007), and some suggestion that contingency strength alters neurophysiological response in the auditory (Baess et al., 2008; SanMiguel, Widmann, Bendixen, Trujillo-Barreto, & Schröger, 2013) and visual (Roussel, Hughes, & Waszak, 2014) domains (see Horváth, 2015 for a discussion). Hence, efference copy modulation may not be wholly responsible for all cases of sensory attenuation, and an account based on learned associations may explain observed effects in contingencies that lack the highly specific motor-to-sensory mapping that exists during speech production.

One possible test of the contribution of learned associations would be to employ a novel contingency (i.e., one never experienced before) between motor output and sensory input. For example, volitional eye movements cannot directly cause sounds in our natural environment, and therefore it seems implausible that an efference copy of the eye motor command would be sent to the auditory cortex. Hence, for a contingency between an eye movement and an auditory stimulus, it is almost certain that no learned association exists, which provides a strong test of whether learned associations are an essential component of sensory attenuation. Moreover, if neurophysiological auditory attenuation were still observed for this novel contingency, it implies that motor-to-sensory mapping is indeed generalizable to indirect action–sensation contingencies. This result would support the forward prediction model of sensory attenuation. However, if sensory attenuation was not observed, a limit for generalizability will have been identified, which might suggest that learned associations are driving button-press-elicited auditory N1 attenuation. Alternatively, sensory attenuation may be driven by a combination of these processes; for example, it seems possible that forward prediction could be established for novel contingencies, such that levels of N1 attenuation are mediated by associative strength.

Accordingly, the primary aim of the present study was to determine whether auditory event-related potential (ERP) attenuation, particularly with respect to the N1 component, would occur following an action–sensation contingency for which no prior learning exists, and thus provide a valuable contribution to the discussion about the processes underlying sensory attenuation. To do so, our experimental design contained a new condition which associated eye movements (the motor output) with tones (the auditory sensory input), a pairing which cannot occur outside of artificial contexts, and for which it is difficult to imagine an analogous, naturally occurring pairing of events. Specifically, participants were presented a pure tone (as is common in auditory ERP studies of this nature, e.g., Baess et al., 2008) following a singular, volitional saccadic movement. To determine whether N1 attenuation occurred, their resultant electrophysiological response, following subtraction of a motor condition (i.e., the same eye movement without stimulus presentation), was compared to an externally initiated condition in which tones were presented without any participant input. Given the possibility raised above that a combination of efference copy and associative processes drive sensory attenuation, we expected to observe N1 attenuation for saccade-initiated stimuli compared to the externally initiated condition. Even so, it seemed unlikely that a novel contingency could produce the same level of attenuation as previously seen in button-press-initiated experiments, as we have substantial prior experience with auditory sensory input following hand motor output.

In view of this hypothesis, our secondary aim was to quantitatively compare self-initiation effects associated with different regions of motor output (i.e., hand and eye). To achieve this aim, our within-subjects design also included an established button-press-initiated condition, for which convincing N1 and P2 attenuation has been demonstrated (Mifsud et al., 2016, Oestreich et al., 2016, Whitford et al., 2011). We chose to additionally assess the P2 component (a large voltage positivity which peaks approximately 200 ms after stimulus onset), which reflects the processing of specific auditory features (Shahin, Roberts, Pantev, Trainor, & Ross, 2005), for two reasons. First, SanMiguel, Todd, and Schröger (2013) have suggested that it may provide a more direct measure of sensory-specific prediction effects than the N1, because, unlike N1 effects, P2 attenuation was uniform over different stimulus onset asynchronies. Hence, it seems prudent to report P2 effects so as to enable comparison between different paradigms. Second, the P2 component has previously been shown to discriminate between self-initiated conditions which differ by motor output region (i.e., hand and foot; van Elk, Salomon, Kannape, & Blanke, 2014), if not necessarily contingency strength (that is, we have substantive experience with both hand and feet producing auditory stimulation), and is therefore relevant given the disparate eye and hand regions targeted in the present study.