Neural correlates of perceptual contributions to nondeclarative memory for faces

Abstract

Face priming is a nondeclarative memory phenomenon that can be observed when recognition is facilitated for a recently encountered face. This data-driven form of priming is distinct from conceptually driven priming. Moreover, it includes two dissociable components, the facilitated access to pre-existing representations and facilitation in perceptual processing of faces. In the present study, we measured neural correlates of perceptual contributions to face priming with event-related brain potentials. Faces appeared two times (separated by 7–17 s), while participants discriminated familiar from unfamiliar faces. Half of the initial face stimuli were inverted, thereby disrupting perceptual face processing and making possible an assessment of perceptual contributions to face priming. Whereas none of the brain waves previously linked to perceptual processing of faces showed indications of priming, such effects were observed between 200 and 600 ms at left occipito-parieto-temporal recording sites. This electrical activity was present for both unfamiliar and familiar faces. The scalp topography of this effect was consistent with sources within the temporal and occipital cortices of the left hemisphere (based on a LORETA source localization). These findings suggest that priming of perceptual face processing is subserved by prolonged neural activity from 200 to 600 ms primarily in the left hemisphere. We propose that this priming reflects facilitated selection based on second-order relations among facial features.

Introduction

An improved ability to identify a face occurs when the same face had been encountered recently (Bruce and Valentine, 1985, Burton, 1998). Such experience-induced changes develop quickly, can be very long lasting, and can occur even in the absence of remembering the recent encounter. This repetition priming is a prime example of nondeclarative memory, which differs from other forms of memory in its functional properties and its neural implementation as revealed with functional imaging and lesion studies (for reviews, see Gabrieli, 1998, Richardson-Klavehn and Bjork, 1988, Roediger and McDermott, 1993, Squire and Knowlton, 2000). For example, nondeclarative memory is often preserved in neurological patients with dramatically impaired declarative memory abilities.

Improved identification of visual objects depends critically on the match of perceptual factors between initial and subsequent encounter and is therefore sometimes called ‘perceptual priming’. It is important to distinguish this data-driven form of priming from conceptually driven forms of priming, which depend on conceptual knowledge spanning multiple modalities or domains. In keeping with these ideas, data-driven priming of recognizing faces is interrupted by a shift of modality or domain (Burton et al., 1998, Ellis et al., 1996; for a review, see Burton, 1998). However, data-driven priming need not reflect only a single process. Recent findings from our laboratory showed that this priming for faces can have two dissociable components, the facilitated access to pre-existing representations and facilitation in perceptual processing of faces (Boehm et al., in press). The representations responsible for data-driven priming of faces are usually considered to comprise (a) representations of familiar faces, or face recognition units (Bruce and Young, 1986, Goshen-Gottstein and Ganel, 2000), or (b) face recognition units in combination with multimodal representations of familiar persons, or person identity nodes (Burton, 1998, Valentine et al., 1996). Facilitated access to pre-existing representations occurs only for familiar faces, whereas facilitation in perceptual processing of faces was found to be independent of face familiarity. Here, we attempted to measure neural correlates of this perceptual contribution to priming.

Much remains to be learned about the neural underpinnings of face priming. Useful perspectives on the neural implementation of memory can be provided via event-related potentials (ERPs) recorded from the human brain. ERPs are time-locked signals within the electroencephalogram, which is thought to reflect neuronal activity primarily produced by spatially aligned cortical neurons activated synchronously (Münte et al., 2000). In contrast to many other methods for measuring human brain activity, ERPs allow a precise description of the time course of neural events. In general, ERP correlates of memory retrieval can be observed as differences between ERPs to repeated and new stimuli (Friedman and Johnson, 2000, Rugg, 1995) and associations between retrieval subprocesses and specific ERPs are currently under investigation. For repetitions of faces after short intervals in the time range of approximately 10 s, as in the current study, a so-called early repetition effect or N250r (ERE/N250r) and a centro-parietal so-called late repetition effect or N400 (LRE/N400) have been reported (Boehm and Sommer, 2005, Boehm et al., 2005, Pfütze et al., 2002, Schweinberger et al., 1995; for a review, see Schweinberger and Burton, 2003). The ERE/N250r is a frontal/fronto-polar positivity peaking around 300 ms when measured against a linked-mastoid reference (Boehm et al., 2005). Because the ERE/N250r has usually not been found for unfamiliar faces when immediate face repetitions were excluded, it has been concluded that it relates to changes in the access to facial representations (face recognition units) due to short-term repetition (Pfütze et al., 2002, Schweinberger and Burton, 2003, Schweinberger et al., 1995). The LRE/N400 appears as a centro-parietal positivity (or reduced centro-parietal negativity) between about 400 and 600 ms and is usually interpreted as reflecting changes in the access to semantic knowledge about the depicted person (for a review, see Schweinberger and Burton, 2003).

Though these ERPs reflect differences in processing between repeated and new faces, it is unlikely that these processing differences relate to data-driven priming for the following reasons. In case of the ERE/N250r, this effect has not been found at a repetition lag of 15 min (Schweinberger et al., 2002a), whereas face priming generally can be measured at this lag. In order to describe neural correlates of data-driven priming in general, specific conditions can be employed to discriminate priming from declarative memory and other repetition-related changes; manipulations such as a shift of modality or domain and varying the level of processing are especially useful because they selectively affect one type of memory (for a review, see Roediger and McDermott, 1993). Whereas data-driven priming for faces is sensitive to a domain-shift from names to faces, the LRE/N400, in contrast, has been found to be similar for different stimulus domains; moreover, the LRE/N400 can also be elicited when the prime and target faces belong to different but related famous persons (Boehm et al., 2005, Pfütze et al., 2002, Schweinberger, 1996, Schweinberger et al., 1995). Some recent studies have applied conditions that dissociate priming from other types of memory and demonstrated ERP correlates of priming for words, reporting small positivities of 1–2 μV around 400 ms that were most pronounced at parieto-central or occipital sites (Boehm et al., 2005, Joyce et al., 1999, Paller and Gross, 1998, Paller et al., 1998, Rugg et al., 2000, Rugg et al., 1998, Rugg and Nieto-Vegas, 1999). For priming of unfamiliar faces, small negativities around 350 ms at central or parietal electrode sites have been reported, which contrast to fronto-temporal positivities at ∼500 ms described for priming of familiar faces (Boehm et al., 2005, Henson et al., 2003, Nessler et al., 2005, Paller et al., 2003).

The quest for neural correlates of priming of face perception can be informed further by considering specific processes of face perception, often referred to as structural encoding (Bruce and Young, 1986, Burton, 1998). Structural encoding entails a particular facility in processing configural information within faces, and three distinct types of configural processing have been distinguished, namely processing of first-order relations, holistic processing, and processing of second-order relations (Maurer et al., 2002). First-order relations, such as the spatial arrangement of a pair of eyes above a nose with a mouth below, define faces as a unique class of visual stimuli. Holistic processing groups individual facial features to a gestalt-like whole, making processing of individual features more difficult than when presented alone. Second-order relations (i.e., distances and angles between eyes, nose, and mouth) are considered critical for distinguishing between individual faces.

The extent to which these three types of configural processing contribute to priming of structural encoding is currently unknown. This question can be partially addressed by analyzing several ERP components that have been purported to bear close relationships to structural encoding—P100, vertex positivity (VPP) or P150, and N170 (Bentin et al., 1996, Itier and Taylor, 2002, Jeffreys, 1996, Linkenaer-Hansen et al., 1998); magnetic counterparts of these ERP components may also be considered, with some caveats (Liu et al., 2002). The functional significance of these components, however, is still under some debate. For example, some investigators hypothesized that P100 relates to processing of first-order relations and N170 to processing second-order relations (Itier and Taylor, 2002, Liu et al., 2002), but others failed to establish a link between N170 and processing specific to the individuality of faces (Bentin and Deouell, 2000, Eimer, 2000a). By determining which ERPs reflect priming of structural encoding, valuable information may be gained concerning both the contribution of types of configural processing to face priming and the functional significance of these ERPs.

In order to measure ERP correlates of priming at the level of structural encoding, we recorded electrical brain activity from scalp electrodes, while participants were asked to discriminate familiar from unfamiliar faces (Fig. 1). Each individual face was presented twice with one to four intervening faces. Repeated faces were always presented upright, whereas half of the familiar and half of the unfamiliar faces were initially presented inverted (Yin, 1969). We chose to use a shift from inverted to upright presentation because all three types of configural processing are disrupted by such face inversion, resulting, for example, in reduced recognition accuracy (Leder and Bruce, 2000, Leder et al., 2001, Maurer et al., 2002). A common assumption is that inverted faces are processed to a relatively higher degree on the basis of their isolated features (Maurer et al., 2002), as indexed by increased activity in brain areas usually more involved in processing other visual objects (Aguirre et al., 1999, Haxby et al., 1999). Given the great divergence between how upright versus inverted faces are processed, priming of structural encoding can occur with upright-to-upright repetition but not with inverted-to-upright repetition (Boehm et al., in press). A pure indication of priming of structural encoding can thus be obtained by computing the difference between ERPs to repeated faces that had been presented upright at initial encounter and ERPs to repeated faces that had been presented inverted at initial encounter.