Direct and indirect measures of verbal relational memory following anterior temporal lobectomy
Introduction
In 1953 the patient HM underwent extensive resection of the medial aspects of both temporal lobes for relief of medically refractory epilepsy [49]. Successful treatment of his epilepsy was achieved at a terrible and unexpected cost: HM became globally amnesic, and to this day he is unable to learn new information such as where he now lives, or the names of people he meets. Scoville and Milner described him as ‘forgetting… the events of daily life as quickly as they occur’ (p. 15).
Over time, it became apparent that HM was capable of new learning in some domains. Milner [34] first demonstrated that he was able to show considerable improvement in mirror tracing with successive practice, and preserved learning has since been shown for a variety of perceptual and motor skills [14], [36], and with some cognitive tasks [11]. Consistently, though, HM has been unable to recall having performed the task previously. Thus, it appeared that surgery had spared only those structures responsible for memories which were not mediated consciously.
Since the initial dramatic insight into the pivotal role of medial temporal lobe (MTL) structures in memory function, research on both human and animal subjects spanning four decades, and more recently, computational modelling studies [32], have provided converging support for the idea of multiple memory systems (see [52]). Demonstrations of contrasting patterns of impaired and preserved learning in a variety of amnesic groups, with varying degrees of contrast, and over a variety of kinds of material, have fuelled debate on the topic of multiple memory systems. A recent meta-analysis of such studies [19] attempted to account for the range of findings and concluded that while preserved learning can be shown, ultimate evidence for multiple systems will involve tests of novel material and novel associations; it was noted that very few such studies have been published to date.
The primary focus in such work is likely to be the anatomically-defined hippocampal system, with the current consensus that its bilaterally paired structures are fundamentally implicated in the formation and retrieval of consciously-available declarative memory (for comprehensive reviews, see [10], [17], [52]; but see [25] for an alternative view). As suggested by its name, declarative memory is conscious and reportable. In addition, it is fundamentally relational, forming conjunctions among highly processed representations, and flexible in its expression over a range of contexts.
According to this school of thought, the properties of the hippocampal system permit the binding operation necessary for relating arbitrary events, and provide a temporal buffer which holds the newly formed bond until an enduring neocortical trace is formed [32]. In the verbal domain, this binding operation is achieved easily by normals, as evidenced by the rapid acquisition of arbitrary pairings in associate learning tasks. In contrast with normal performance, and consistent with an MTL model of relational processing, HM was unable to associate arbitrary pairings [49].
Thus HM's ‘MTL’ amnesia is characterised as an impairment in declarative memory: he cannot associate the (almost inevitably) arbitrary features of someone's name and face, much less use the information flexibly in novel situations. The kinds of memory that HM can form and use without conscious awareness are termed procedural [12], nondeclarative [53], or nonrelational [10]. In a multiple memory systems framework, they are processed by a variety of extratemporal brain structures.
Most of the animal research underlying the multiple memory systems proposal has focused directly on the role of the MTL system, showing how memory for relational material is poor when structures are lesioned in isolation or in combination [52], [54]. Human studies, on the other hand, have traditionally shown that procedural memory is preserved in cases with diverse etiologies, lesion sites, and degrees of severity; most prominently, alcoholic amnesics with primary damage to midline diencephalic nuclei show the declarative/procedural dissociation [6], [8], [42].
The bias towards studying amnesias which have a predominantly extratemporal locus largely stems from their higher base rate of prevalence. A small number of cases with non-surgical and non-toxic medial temporal damage has been studied: herpes simplex encephalitis [16], posterior cerebral artery occlusion [3], and hypoxic ischaemia [57] are capable of producing moderate to severe amnesia, and in these cases declarative/procedural memory dissociations have been demonstrated.
One relatively common form of MTL dysfunction has not contributed much to discussions about multiple memory systems. The complex partial seizures of temporal lobe epilepsy (TLE) commonly have an epileptogenic focus in one or other hippocampus; neuroimaging often shows unilateral hippocampal atrophy, and in medically refractory cases where unilateral anterior temporal lobectomy (ATL) is performed, neuropathological examination of resected tissue usually demonstrates hippocampal sclerosis [2].
Given that unequivocally lateralised pathology is a precondition for surgery, most studies of the neuropsychological sequelae of TLE have used postoperative cases, exploiting the opportunity of providing histological verification of pathology. Memory impairment is common and is characteristically material-specific: with the left hemisphere dominant for language in most of the population, left-sided temporal lobectomy (LATL) is typically associated with some degree of verbal memory deficit [7], [35], [37], [40], while right-sided temporal lobectomy (RATL) is associated with memory impairment in the visuospatial domain [26], [51]. The latter findings are demonstrated less reliably (see [29]) and research interest has focused on verbal memory impairment.
Most research on memory in TLE has been conducted in the context of clinical epileptology, with diagnostic implications in terms of localising a seizure focus [27]. In addition, such studies have contributed to an understanding of lateralisation of cognitive function. However, a TLE population seems well suited for studying cognitive models of memory more generally. The site and nature of neuropathology is relatively uniform, and the prevalence of the disorder is relatively high. Surprisingly, little work has been done with this population to examine whether preserved memory can be demonstrated.
Blaxton's [4] study of conceptually- and data-driven memory processing in TLE contributes to cognitive modelling of memory, but is couched in processing rather than neuroanatomical terms; as such, her data tell a story which runs parallel to the theme of neurally separable memory systems. She found that conceptual processing of information was impaired in left-sided TLE, irrespective of whether testing was explicitly focused on to-be-remembered material (semantic cued recall), or was incidental (category production). On the other hand, perceptually based memory was preserved, again irrespective of the mode of testing.
In a study designed to address multiple memory systems theory, Zaidel, Oxbury, and Oxbury [62] showed that despite poor free recall of a list of words LATL subjects performed as well as RATL subjects and normal controls on a word-stem completion task. That is, they tended to produce list words when asked to think of the first word that came to mind when a three-letter word-initial stem cue was provided (e.g. ELE-). Drawing parallels with similar outcomes in extratemporal amnesic populations [22], [23], the authors adopted a broadly equivalent terminology describing contrasts between implicit and explicit memory (cf. procedural vs. declarative memory; see [47]). Their data are compatible with the usual conceptualisation of procedural memory: under test conditions which are less demanding than free recall in terms of processing resources, the stem reactivates the recently activated representation of the word, constituting a partial repetition of the original learning context. Another study by Zaidel and colleagues [61], however, revealed minimal incremental accuracy improvement when stem completion was performed, which was equivalent for LATL and RATL cases.
The memory processes and representations tapped by stem completion are likely to be finely-tuned, inflexible, and nonrelational. As such, the findings of Zaidel et al. [62] with ATL subjects extend the domain of preserved learning in amnesics to an MTL-disordered population, but their use of lists of individual words does not provide the ‘acid test’ of MTL function. More potent data would involve a dissociation between impaired and preserved memory using relational materials in a paired associate learning (PAL) paradigm. Data from our group [43] demonstrate the impaired part of the dissociation, using an ‘explicit’ method of testing: subjects with left-sided TLE tested presurgically were impaired in learning the ‘hard pairs’ (e.g. cabbage–pen, crush–dark) of the Associate Learning subtest of the Wechsler Memory Scale (WMS; [58]). Right-sided presurgical subjects showed normal levels of performance. Another study using WMS hard pairs found memory to be equivalent in left- and right-sided adolescent presurgical cases, but also found that left-sided performance dropped significantly following surgery [61].
Studies which have used a relational paradigm to address the other (preserved, ‘implicit’) part of the dissociation have produced ambiguous findings. Graf and Schacter's [21] study of a mixed group of (largely extratemporal) amnesics appeared to show intact stem completion in a PAL paradigm: having learned a pair such as WINDOW-REASON, amnesics could not recall REASON directly when presented with WINDOW as a cue, but correctly completed the paired stem WINDOW-REA- when asked to think of the first word which came to mind; such completions were made more often than when a control pair-stem was presented (e.g. RIPE-REA-). However, these findings soon received significant qualification, with further analysis showing that only a mildly amnesic subset of their sample demonstrated preserved implicit memory [48]; for these subjects, the interpretation was that explicit memory was sufficiently intact to support stem completion and so the data become irrelevant (see [20], [33]). Subsequent PAL stem completion work with severe amnesics has also failed to find evidence of spared learning [50].
The present study addresses the status of preserved memory for relational material, using a TLE population. We adopt the view, broadly consistent with views expressed by Gooding, Mayes, and van Eijk [19], that the postulation of preserved relational memory runs counter to the tenets of a multiple memory systems framework. According to theorists such as Squire, Cohen, and Eichenbaum, relational memory is definitionally declarative, and is processed by the hippocampal system; relational memory is not supported by extratemporal memory systems in any enduring sense. Thus, evidence of preserved relational memory in amnesia might suggest that MTL structures are not uniquely responsible for forming conjunctions, thereby implicating another relational processing system.
As the notion of preserved relational memory is not captured in the terminology of a declarative/procedural dichotomy, and even less so in the explicit/implicit dichotomy, the present study adopted a representationally neutral terminology which focused on how the evidence of a memory trace was derived. Consistent with the terms used by Richardson-Klavehn and Bjork [41], memory of an event was measured either directly (e.g. the task involved recall of the event) or indirectly (e.g. the task produced evidence of the formation of a memory trace, but did not require consciously mediated attention to the event itself). The present study tested for the existence of a relational memory trace which was not able to be expressed under direct recall conditions, by using a novel indirect method which actually precluded declarative knowledge of the relationship from playing a role in performance. Thus, in a mildly amnesic TLE sample, any interpretation of relative savings was not complicated by the possibility that declarative memories were brought to bear in the task (cf. the account for other mildly amnesic subjects [48]).
Processing load was reduced by adopting a recognition memory paradigm [24], circumventing any purely retrieval-based contribution to poor cued recall performance. However, simple recognition of learned pairs from a set including distractors would require an unwieldy item set, and in principle, could be achieved declaratively. Adapting the masked priming technique of Forster and Davis [18] provided an elegant solution.
Masked priming was originally devised to minimise strategic or episodic influences among stimulus words in studies of lexical processing. The technique involved serial presentation of a forward masking stimulus, a brief prime, and then a target to which the subject responded; critically, control studies showed that subjects were not able to report on attributes of the prime ([18]; indeed, most subjects did not realise that the prime was presented). Targets were classified as words more rapidly (and sometimes more accurately) in a lexical decision task when they were preceded by a masked prime which was identical (though presented in a different case), relative to when they were preceded by a control word. In this context, priming was thought to pre-activate the lexical representation of the target word. Later work adapted the paradigm to explore distinctions between representations in episodic and semantic memory: links formed between pairs in a learning phase were found to be activated by a masked prime, facilitating episodic recognition of a consciously-perceived single word target [5].
The present study teams together paired associate learning and masked priming in two distinct phases. In Phase 1 of the experiment the subject attempts to learn arbitrary pairings of words presented visually on a computer screen, and a direct measure of memory is provided by cued recall. In Phase 2, any link that has been formed is probed while subjects make episodic recognition judgements on the second element of the pair (deciding whether or not it had appeared in the list of learned pairs can be performed without great difficulty by mildly amnesic subjects, even if they cannot recall the pairings between words). The link is probed by presenting the matching element of the pair immediately before the word to be recognised; the word appears only briefly, is visually masked, and the subject is not consciously aware of having seen it. Although unavailable to conscious report, the masked prime is processed rapidly, and if a link has been formed in Phase 1, serves to facilitate the recognition response to the immediately following, clearly visible target word. Facilitation is measured by a reduction in reaction time (RT) (and potentially an increase in accuracy) to make a recognition judgement on what is perceived by the subject to be a single word; most priming effects are revealed in terms of shorter response latency, with corresponding effects occasionally shown in terms of accuracy, if sufficient errors are indeed made. Priming provides an indirect measure of memory for the novel pairing, as the task being performed makes no reference to (and indeed, obscures) the process being measured.
Masked priming of recognition memory provides a technique which should be sensitive to the formation of links between previously unrelated material, while precluding conscious declarative processing. Using this indirect method with an MTL-disordered subject population, the present study provides a strong test of the main premise underlying the proposal for multiple memory systems: if the MTL system is unique in dealing with relational memory, then MTL damage should be associated with poor learning across the board, and masked priming should not be able to detect evidence of learning over and above what can be shown directly using a conscious task. Specifically, LATL subjects would be expected to perform worse than RATL subjects using a direct measure of memory (cued recall in Phase 1 of the experiment), and an indirect measure (masked priming in Phase 2).
Section snippets
Subjects
Twenty-five subjects contributed data for the study. Recruited from the Comprehensive Epilepsy Programme at the Austin and Repatriation Medical Centre, they had undergone unilateral anterior temporal lobectomy for treatment of medically refractory complex partial seizures. Unilateral localisation of epileptogenic focus had been achieved by combining diagnostic information from video-EEG telemetry, structural MRI, functional imaging such as ictal SPECT and interictal PET, and neuropsychological
Phase 1: serial learning with cued recall
Fig. 1 shows summary data for LATL subjects and RATL comparison subjects for serial learning of concrete and abstract word pairs. ANOVA revealed robust main effects of Subject Group, F(1, 23)=21.30, P<0.001, Word Type, F(1, 23)=40.82, P<0.001, and Trial, F(3, 69)=99.14, P<0.001. Interpretation of these effects was qualified by a significant three-way interaction, F(3, 69)=6.35, P<0.005, which appeared to reflect steady learning of both kinds of material for RATL subjects, and of concrete pairs for LATL
General discussion
Current theorising about multiple memory systems proposes that different brain systems have specific roles in capturing different aspects of learning. At the heart of the theory is the putatively unique role of MTL structures in encoding relational/declarative memory, with extratemporal brain systems being responsible for nonrelational/procedural learning. None of the various conceptual frameworks used to distinguish between the systems (e.g. declarative/procedural, explicit/implicit,
Acknowledgements
We would like to express our gratitude to the patients who participated enthusiastically in the study; we also thank two anonymous reviewers for their comments on an earlier version of the manuscript.
References (62)
Hippocampal adequacy versus functional reserve: predicting memory functions following temporal lobectomy
Archives of Clinical Neuropsychology
(1995)The language-as-fixed-effect fallacy: a critique of language statistics in psychological research
Journal of Verbal Learning and Verbal Behavior
(1973)Acquisition of motor skill after bilateral medial temporal lobe excision
Neuropsychologia
(1968)- et al.
Unilateral medial temporal lobe memory impairment: type deficit, function deficit, or both?
Neuropsychologia
(1998) - et al.
A meta-analysis of indirect memory tests for novel material in organic amnesics
Neuropsychologia
(2000) Some comments on the special cognitive functions claimed for the hippocampus
Cortex
(1994)Right hippocampal excision impairs learning and recall of a list of abstract designs
Neuropsychologia
(1986)- et al.
Further analysis of the hippocampal amnesic syndrome: 14 year followup study of H.M.
Neuropsychologia
(1968) - et al.
Right hippocampal impairment in the recall of spatial location: encoding deficit or rapid forgetting?
Neuropsychologia
(1989) - et al.
Memory: brain systems and behavior. Special issue: learning and memory
Trends in Neurosciences
(1988)