Processing and short-term retention of relational information in amnesia

Abstract

In a recent eye-movement study [Psychol. Sci. 11 (2000) 454], amnesic patients failed selectively to exhibit long-term effects of memory for the relations among the constituent elements of scenes. This failure could be due to a deficit specifically in long-term relational memory, as we have suggested; or in retention of relational information over any delay, whether involving perceptual processing and short-term maintenance or long-term memory, consistent with suggestions from recent studies of the hippocampus; or in on-line processing of relational information, as would occur in perceptual or feature binding. Here we show robust eye-movement effects of relations among elements of scenes in amnesia in a short-delay matching task, with the same materials and in the same amnesic patients in which long-delay conditions elicited failure. These findings document intact processing and short-term retention of relational information in amnesia, indicating that amnesia associated with hippocampal damage results in a relational memory deficit, specifically of long-term memory.

Introduction

Several notable findings in the neuropsychological literature (Chun & Phelps, 1999; Gabrieli, Cohen, & Corkin, 1988; Ryan, Althoff, Whitlow, & Cohen, 2000) have provided strong support for the claim that amnesia associated with hippocampal damage results in a selective deficit in memory for relations among items (Cohen, 1984, Cohen & Eichenbaum, 1993; Cohen, Poldrack, & Eichenbaum, 1997; Cohen & Ryan, 2003; Cohen et al., 1999; Eichenbaum, 1997; Eichenbaum & Cohen, 2002; Eichenbaum, 1999). Particularly striking was Ryan et al.’s (2000) finding that amnesic patients failed selectively to exhibit the relational manipulation effect, an eye-movement effect that is sensitive to memory for relations. While these findings implicated the hippocampus in relational long-term memory; the present work examines whether the hippocampus is responsible for the general processing of relations.

In Ryan et al. (2000), subjects viewed a series of images of real-world scenes, then saw some of the scenes repeated in their original form and saw other scenes in manipulated form. Each manipulated scene involved a change in the relations among the constituent elements of a scene studied earlier. In the critical viewing block, a given scene was viewed as a repeated scene by some subjects (i.e., was always viewed in that form) and as a manipulated scene by other subjects (i.e., was being viewed in that form for the first time, after having undergone a manipulation of the original relations among items). In this way, repeated and manipulated scenes were physically identical, differing only in their viewing history. Normal subjects exhibited differential viewing of the manipulated scenes compared to viewing of the (physically identical) repeated scenes. But it was the pattern of the differential viewing that is so critical. The eyes were attracted disproportionately to the very area of relational manipulation in the manipulated scenes—a pattern we have called the relational manipulation effect. Subjects directed a greater proportion of viewing time and a greater proportion of total fixations to the manipulated critical region compared to viewing of the same critical region for other subjects for whom no change had occurred in the region. Similarly, subjects made more transitions into/out of the critical region with their eyes when the critical region had been manipulated versus unmanipulated.

This effect occurred whether or not viewers were consciously aware of or had explicit memory for the changes. In a separate experiment, college-aged subjects were asked to report whether the scenes had changed from a previous viewing, and to indicate the nature of the change. It would be possible for subjects to be aware of the change without being able to correctly specify the exact nature of the change (e.g., a cat moved from the left side of the scene to the right side). However, in all cases, when subjects reported they were aware that a scene had changed, they were able to specify the exact nature of the change. Therefore, subjects were considered to be unaware of the change only if they reported that the scene had not changed in any way when, in fact, a change had occurred.

When the trials were separated on the basis of whether or not subjects had conscious awareness for the change, it could be seen that conscious awareness for the change was not a requirement for elicitation of the relational manipulation effect, that is, the eyes were attracted to the critical region on all three measures of viewing (proportion of fixations, proportion of time, or number of transitions into/out of the critical region) even when subjects were unaware that any change had occurred in the scene. With no dependence on conscious awareness of the changes, and given that subjects are never asked to make explicit memory judgments in this task, this eye-movement effect is seen as providing an indirect (implicit) measure of memory for relations.

Amnesic patients failed entirely to show this relational manipulation effect under free-viewing conditions; none of the patients exhibited it. Yet, they showed normal memory for items, which was simultaneously assessed with indirect (implicit) eye-movement measures, manifested in differential viewing of repeated scenes compared to novel scenes. Thus the deficit in amnesic patients was selective to memory for relations.

The selective failure of amnesic patients to display long-term effects of memory for relations in this paradigm, as with other examples of impairment in memory for relations among items (see Cohen & Ryan, 2003, Eichenbaum & Cohen, 2002), could be due to any one of several possible deficits to different stages in the processing and representation of relations: (1) in the processing of relational information, including the on-line processes that are part of perceptual or feature binding; (2) in the retention of relational information across any delay, whether involving short-term maintenance or long-term memory; or (3) specifically in long-term relational memory. That is, while the findings point unambiguously to the role of the hippocampal system in representing the relations among items, they cannot, alone, specify the particular stage in the chain of processing and retention of relational information that is impaired in amnesia.

Before proceeding any further, it is important to clarify some anatomical terms. By “hippocampal system” we refer here, as in our other writings (e.g., Eichenbaum & Cohen, 2002), to the hippocampus (including the hippocampus proper, the dentate gyrus, and the subicular complex) and the parahippocampal region (including the entorhinal, perirhinal, and parahippocampal cortices). Much research on memory and amnesia considers the entire hippocampal system, but more recent work in the field focuses specifically on the hippocampus, and we have argued (Cohen & Ryan, 2003, Eichenbaum & Cohen, 2002) that it is the hippocampus itself that critically subserves the relational memory functions under investigation here.

Ryan et al. (2000) interpreted the failure of amnesic patients to show the relational manipulation effect as a deficit in long-term memory for relations. One reason for emphasizing the long-term memory component is the abundant evidence that the deficit caused by hippocampal damage in (human and animal) amnesia is delay-dependent: The marked memory impairment that is characteristic of amnesia is exhibited not in immediate test conditions but rather after a significant delay after learning. A second reason to emphasize long-term retention is that the task required subjects to answer orienting questions about each scene, which ensured that all viewers, control subjects and amnesic patients alike, successfully processed the very relations that, in manipulated scenes, would undergo change.

It remains possible, however, that the critical deficit shown by amnesic patients in memory for relations could be a more general one, occurring earlier in the processing and representation of relational information than previously appreciated. Two developments in the literature recommend consideration of this possibility. First, work on attention and its critical role in the representation of objects has emphasized perceptual or feature binding, i.e., the binding together, in on-line processing, of the different attributes of an object that are simultaneously represented in different cortical maps, e.g., color, contour, and spatial location. This process is thought to give rise to the integrated percept and representation of each object (see Treisman & Gelade, 1980). Johnson and colleagues (e.g., Chalfonte & Johnson, 1996) have suggested that it is this early aspect of processing and representation that is supported by the hippocampal system and is compromised in amnesia.

Second, several recent findings in the functional-imaging literature have pointed to hippocampal activation in working memory tasks, i.e., even under short-delay conditions (Mitchell, Johnson, Raye, & D’Esposito, 2000; Ranganath & D’Esposito, 2001; Stern, Sherman, Kirchoff, & Hasselmo, 2001). For example, Ranganath and D’Esposito (2001) found activation bilaterally in anterior hippocampus during a delayed match-to-sample task with faces, in which the delay was on the order of several seconds; this region was not active during the encoding or retrieval phases of a long-term memory task also using faces. The observed hippocampal activation was greater when novel faces, as opposed to familiar faces, were used as the stimuli for the working memory task, suggesting that greater involvement of the hippocampus is needed in order to establish and maintain representations of novel stimuli across even short delays. A similar conclusion can be drawn from Stern et al.’s (2001) findings of greater hippocampal activation bilaterally during a two-back working memory task with novel scenes rather than with familiar scenes.

While findings from neuroimaging indicate the participation of hippocampus in working memory tasks, standing in stark contrast are the neuropsychological studies with amnesic patients which suggest that the hippocampus is not critical for intact working memory performance. On a variety of short-delay tasks using digits, words, pictures, mazes, etc., and in trial unique delayed matching or nonmatching to sample tasks similar to the neuroimaging studies described above, amnesic patients perform normally if the maintenance of information critical to the task at hand is needed only over a short delay (see Aggleton et al., 2000; Cave & Squire, 1992; Holdstock, Shaw & Aggleton, 1995; Milner, Corkin, & Teuber, 1968; Warrington & Baddeley, 1974; but see Owen, Sahakian, Semple, Polkey, & Robbins, 1995).

In order to address the possibilities raised by these multiple lines of work, the present study examined directly the processing and short-term retention of relational information in amnesic patients, using the same eye-movement monitoring methods and the same materials reported in Ryan et al. (2000). This permitted us to test whether amnesia compromises these earlier stages of processing and representation of relations, or whether the deficit in amnesia is instead selective to long-term memory for relations.

The work reported here used a short-delay matching task, in which amnesic patients viewed pairs of scenes. The first member of each pair was followed, within a few seconds of blanking the screen, by either the same scene or a manipulated version of the scene. Disproportionate attraction of the eyes to the region of change in the manipulated scenes by amnesic patients would indicate memory for the (changed) relations within the scene, under these delay conditions. Such an outcome would show that amnesia is not fundamentally a deficit in the processing and short-term retention of relational information. Together with the failure to show the relational manipulation effect in the long-term memory (free-viewing) version of this task (Ryan et al., 2000), this outcome would isolate the deficit in amnesia (and the role of the hippocampal system) to long-term memory for relations. By contrast, a failure of amnesic patients to show eye-movement effects diagnostic of memory for relations even under the short-delay conditions of this study would indicate a more general deficit in relational processing and representation.