Single-neuron encoding of surprise in auditory processing.

ABSTRACT

The main role of structures in ascending sensory systems is to extract raw features of sensory input and compartmentalize the information-bearing elements for use by the brain. Information-bearing elements can be apparent, as in the case of stimulus frequency or intensity (Ehret and Merzenich 1988; Tramo et al. 2002; Yu et al. 2010). The features of sound that drive neuronal fi ring at higher auditory centers, however, remain elusive. In their exciting article, Gill and colleagues (2008) show how “surprise” is a dimension of auditory experience that alters fi ring patterns of central auditory neurons. By elaborating the method for calculating and extracting spectro-temporal receptive fi elds (STRFs), the authors demonstrate that auditory neurons, mainly those from hierarchically higher-order areas, modulate their discharge rates in response to sound elements that deviate from expected values. This work is the fi rst to capture and separate encoding due to surprise from the ongoing encoding of spectral and temporal elements of acoustic cues (Theunissen et al. 2004). The coding of auditory information was studied in a highly social songbird species, the zebra fi nch (Taeniopygia guttata), which frequently engages in vocal exchange as part of its normal behaviour (for reviews, see Zeigler and Marler 2004). On the receiving (sensory) end of this exchange, the acoustic elements of the incoming birdsong, including notes and syllables, are encoded by auditory neurons (for reviews, see Mello et al. 2004; Gentner 2004). As with words in human speech, for a song to be recognizable over repeated use, the order of all of its individual sound elements must also be largely preserved across time. Consequently, songbirds naturally generate expectations not only for specifi c songs but also for the general structural rules, internal correlations or probability statistics that apply to song elements. To determine if surprise was predictive of altered neuronal activity, electrophysiological recordings were made in key structures of the ascending auditory pathway, including the songbird analogue of the mammalian inferior colliculus (nucleus MLd), the primary auditory forebrain area (Field L2) or an association auditory forebrain area (CLM) (Vates et al. 1996; Mello et al. 1998). One of the main goals of this work was to isolate the impact of surprise on auditory encoding for different cells (Gill et al. 2008). To this end, different forms of STRF were compared, including a STRF that was specifi cally developed to capture the impact of fi ring due to unmet expectations in stimulus structure (a surprise-STRF). In order to drive neuronal fi ring by surprise, Gill and colleagues generated song stimuli in which certain song elements were louder or softer than expected. Deviations were only introduced as changes in power for a particular element given a brief sample of “stimulus history”. This manipulation allowed for the measured and elegant application of “surprise” embedded on the song elements without having to interpret surprise in the context of the entire song. The authors show that surprise-STRF had far greater predictive strength relative to other STRF metrics and, therefore, was useful to parse out and quantify changes in fi ring given the probability of that change occurring based on prior experience. Surprise-STRFs were shown to have provided improvement in predictive power for select neurons at all three levels of the auditory pathway that were tested. Great gains in prediction were, however, frequently made by surprise-STRFs in the higher-order auditory area CLM, for two dominant cell types named by the authors as off-set and complex auditory neurons. Interestingly, in neurons that are surprise-responsive, Gill and colleagues found that the degree of altered fi ring was relatable, in linear terms, to the magnitude of change introduced. In addition, surprise coding was directionally sensitive; surprises to augmented stimulus power could be encoded at an entirely different sub-set of neurons than cells tuned to the surprise of a lower than expected stimulus power.