Selective entrainment of brain oscillations drives auditory perceptual organization

Highlights

• Facing ambiguous soundscapes, humans perceive unitary multistable organizations.

• Brain oscillations concurrently entrain to the rate of all competing sound patterns.

• Entrainment to ignored sound organizations is restricted to auditory regions.

• Entrainment in an auditory-motor network reflects the perceived sound organization.

Abstract

Perceptual sound organization supports our ability to make sense of the complex acoustic environment, to understand speech and to enjoy music. However, the neuronal mechanisms underlying the subjective experience of perceiving univocal auditory patterns that can be listened to, despite hearing all sounds in a scene, are poorly understood. We hereby investigated the manner in which competing sound organizations are simultaneously represented by specific brain activity patterns and the way attention and task demands prime the internal model generating the current percept. Using a selective attention task on ambiguous auditory stimulation coupled with EEG recordings, we found that the phase of low-frequency oscillatory activity dynamically tracks multiple sound organizations concurrently. However, whereas the representation of ignored sound patterns is circumscribed to auditory regions, large-scale oscillatory entrainment in auditory, sensory-motor and executive-control network areas reflects the active perceptual organization, thereby giving rise to the subjective experience of a unitary percept.

Introduction

Perception can be thought of as an act of inference (Gregory, 1980, Helmholtz, 1866). Modern neuroscience views the brain as a predictive machine, continuously generating internal models of the causal dynamics of the world in an attempt to interpret its observations (Bar, 2009, Friston, 2005). Although relevant to all sensory systems, this assumption especially applies to audition (Baldeweg, 2006, Garrido et al., 2009, Winkler et al., 2012). Particularly, it applies to sequential organization, which refers to the sorting of interleaved sounds (Dowling, 1973, Bregman, 1990, Sussman et al., 1999, Shamma et al., 2011, Winkler et al., 2009). Meaningful auditory objects rely on binding distributed spectrotemporal patterns into coherent streams (Bregman, 1990, Nelken and Bar-Yosef, 2009, Sussman et al., 1999). Yet, auditory information can sometimes be feasibly explained by more than one internal model. For instance, in a musical piece, a single note from an instrument could belong simultaneously to a melodic line, to a harmonic progression and to a rhythmic pattern featuring several instruments. However, despite hearing all sounds, we consciously perceive univocal organizations that we can flexibly listen to. Our subjective experience therefore conforms to the Gestalt principle of exclusive allocation (Kohler, 1947), which states that any sensory element should not be used in more than one description of the natural scene at a time. Whether this principle also applies at the neural level, specifying memory representations of the stimulus input (i.e., whether multiple internal models are held simultaneously or only the current attended one) is still a matter of intense debate (Sussman et al., 2014, Denham et al., 2014, Grossberg et al., 2004).

How the brain flexibly assigns individual events to any of the possible perceptual organizations they could fit into is optimally studied with ambiguous, multistable stimulation, because perception depends on the model currently explaining unchanging sensory input (Sterzer et al., 2009). Behavioral evidence on auditory spontaneous perceptual switches suggests that multiple alternative organizations are held simultaneously and compete to describe the acoustic scene (Denham et al., 2014, Pressnitzer and Hupe, 2006, Sterzer et al., 2009, Sussman et al., 2014). Electrophysiological studies in humans have traditionally embedded violations of established regularities within the acoustic streams in order to use change detection auditory evoked potentials, such as the mismatch negativity (MMN) (Näätänen et al., 1978), as an index of sound organization (Sussman et al., 1998, Sussman et al., 1999). However, besides yielding conflicting results, with some studies showing simultaneous encoding of alternative organizations (Pannese et al., 2015, Sussman et al., 2014) while others suggesting that only the currently perceived organization is represented (Sussman et al., 2002, Sussman, 2013, Winkler et al., 2006), evidence of this nature is intrinsically indirect and does not inform about the neural mechanisms underlying the representation of sound organization.

Several studies have shown that any existing regularity in the auditory scene is reflected in oscillatory activity tuned to its temporal pattern (Henry et al., 2014, John et al., 2001, John et al., 2002, Luo et al., 2006, Luo and Poeppel, 2007, Pannese et al., 2015). This is an interesting observation because synchronized oscillatory activity has been proposed as an effective means for neuronal communication (Fries, 2005). Moreover, since the high-excitability phase of ongoing low-frequency oscillations can be selectively entrained to events occurring in an attended stream (Schroeder and Lakatos, 2009), we speculate that neuronal entrainment could underlie our perceptual ability to flexibly reorganize sequential sounds.

We hereby designed a novel ambiguous sound sequence that allowed the study of active perceptual reorganization while controlling for sensory input. Given the quasi-rhythmic nature of most behaviorally relevant acoustic information (Patel, 2008), rhythmic attention (Jones and Boltz, 1989, Large and Jones, 1999), and its neurophysiological counterpart oscillatory entrainment (Herrmann and Henry, 2014, Schroeder and Lakatos, 2009) would likely play a key role (Pannese et al., 2015). Nozaradan et al. (2011) demonstrated that oscillatory entrainment underlies meter imagery, the voluntary organization of musical beats. However, the imagined meter was imposed on a sound sequence with acoustic energy only at the main beat rate. This leaves open the question of whether oscillatory entrainment actually helps to disambiguate a rhythmic structure that has multiple potential meters. With energy at more than one possible meter, task demands may act to enhance the attended meter while suppressing the unattended one, rather than driving the overall meter of the sequence.

To target the dynamics of large-scale neuronal slow oscillatory activity, we combined spectral analyses with source localization of EEG data, seeking to explore the distinction between the neurophysiological nature of simultaneously encoded representations of the auditory scene, and the selected internal model underlying the perceived auditory object.