The hippocampus is essential for completely unconscious as well as conscious flexible memories

This scientific commentary refers to ‘Unconscious relational encoding depends on hippocampus’ by Duss et al. (doi: 10.1093/brain/awu270).

Several beliefs about the nature of organic amnesia were deeply entrenched by the early 1990s. It was believed that damage to the medial temporal lobe or its connections in the midline diencephalon and basal forebrain did not disrupt intelligence, perception, processing of information at input and immediate (short-term) memory, but selectively disrupted—following even a brief distraction—recall and recognition of recently encountered personal events or facts/concepts. In other words, amnesia was thought to affect only episodic and semantic memory, whether for pre-morbid events (retrograde amnesia) or post-morbid events (anterograde amnesia). With these kinds of memory, not only is the remembered information consciously apprehended at input and at retrieval, but the person remembering is also consciously aware that the information is a memory. That is, the person feels that the information has been encountered before, often in a particular context. In contrast, it was believed that amnesics acquired and retained other kinds of memory, such as habits and skills, normally. Although these latter types of non-declarative memory are functionally heterogeneous, they all lack the conscious feeling of memory that accompanies episodic and semantic memory.

Since the late 1990s, several of these entrenched beliefs have been challenged. On the one hand, it has been argued that amnesia is not functionally unitary: while disruption of the hippocampal, fornix and anterior thalamic system impairs recall of flexible (relational) associative memories, disruption of the perirhinal cortex or its dorsomedial thalamic projection impairs inflexible item familiarity (for a review, see Montaldi and Mayes, 2010). Relational memory representations have components linked to each other as well as to other representations, which allows flexible access through multiple cues both to the representation and from it to other representations (Cohen and Eichenbaum, 1993). Rigid representations lack these componential links that allow flexible two-way access to much stored information.

On the other hand, it has been argued that some kinds of function are not preserved as claimed. These include high-level visual perception (e.g. Lee et al., 2012) and immediate memory for item-location associations (e.g. Pertzov et al., 2013). Another strong belief that has been challenged is that priming is preserved in organic amnesics. This challenge is of particular interest because priming is a kind of unconscious memory for specific kinds of information involving items and/or associations, for which there can also be recall and recognition, i.e. conscious memory. The unconscious memory is revealed by faster, more accurate or altered processing of previously encountered items or associations even when these are not recognized or recalled. This implies that, if priming is preserved in amnesics, they must have preserved unconscious memory for previously encountered items or associations, even though their conscious memory for the same stimuli is impaired. Although conscious and unconscious memory of a given stimulus probably depend on partially distinct encoded stimulus features, some overlap between the kinds of information that support recognition and priming of the same nominal stimulus seems likely: overlap can be adjusted by selecting appropriate recognition foils to match the information that priming is based on. So, if priming is preserved, there is a problem for the dominant view that hippocampal system dysfunction disrupts storage of complex associative inputs. If priming is preserved, then amnesics are not so much impaired at storing new episodic and semantic information as unable to generate conscious memory for these things.

Challenges to the view that priming is preserved in amnesia typically focus on novel associations, such as the link between previously unrelated words or object pictures. If subjects are able to make judgements about pairs of stimuli studied together (e.g. which is bigger) faster than about otherwise similar pairs that were not studied together, that is evidence of priming. Although meta-analysis has revealed that amnesics show intact priming for previously familiar items, they are impaired relative to matched controls at priming of novel associations (Gooding et al., 2000). However, this does not prove that hippocampal damage impairs novel associative priming because healthy people will recall or recognize many more studied pairs than amnesics and may use this conscious memory to enhance their ‘primed’ performance. It is very difficult to control for this potential confound because joint tests of associative recognition and associative priming interfere with each other (Spencer et al., 2009).

For some years, Henke and her colleagues have been developing a priming paradigm that completely avoids this confound and also, unlike standard priming, clearly shows evidence of unconscious relational priming that may depend on the hippocampal system. They have presented novel associations between words or pictures very briefly with a preceding and following mask so that participants are not consciously aware of which stimuli they have been shown. In a subsequent test, where they must consciously identify pairings of components that are related in a semantic way to the unconsciously studied pair, their judgements are altered for pairs related to those studied relative to similar but unrelated pairs. For example, Reber et al. (2012) subliminally showed transitively-related word pairs (A–B and B–C) or unrelated pairs (E–F and G–H). A minute later participants were supraliminally shown pairs that were either transitively related like A–C or unrelated like E–H, and had to judge whether the words were semantically related. They more often judged transitively related pairs to be related semantically. Hippocampal activity was higher during unconscious encoding of transitively related pairs and during the unconscious retrieval of such pairs. This inferential unconscious memory is clearly relational, although functional MRI correlations cannot prove that hippocampal activity is essential for it. To do this requires a lesion approach.

Duss et al. (2014) in this issue of Brain, have now addressed this matter. They found that 11 amnesics with hippocampal system damage were not only impaired at unconscious relational associative verbal memory, but also at an indirect memory test where they remained conscious of the primed stimuli at both study and test. Participants were subliminally shown unrelated word pairs like ‘violin–lemon’ and ‘table–car’ up to nine times for 17 ms each time with masks immediately before and after. At test 5 min later, they were supraliminally shown pairs either related to those previously seen, such as ‘cello–mandarin’, or unrelated, such as ‘harp–truck’. Following nine presentations, the controls judged the related pairs to be semantically related more often than they did the unrelated pairs, whereas the patients, as a group, did not. However, the patients performed normally at another task at which they were also never conscious of the subliminally studied words. In this task, subliminal presentation of a single word like ‘fisher’ was followed immediately by supraliminal presentation of a related word (e.g. ‘angler’) or an unrelated word (e.g. ‘credit’), and participants were required to judge the pleasantness of the latter. It took significantly longer to judge pleasantness for related test words, but the effect was the same in amnesics and their controls. The authors plausibly argue that the impaired unconscious associative memory was relational and dependent on the hippocampal system. In contrast, the preserved unconscious semantic single-word memory comprised rigidly bound components; it was therefore not relational and depended not on the hippocampal system, but on prefrontal language areas.

These anatomical conclusions were mainly based on extensive MRI investigation of all the controls and six of the patients who proved scannable, using structural MRI with manual hippocampal volumetry and voxel-based morphometry, functional MRI and resting state functional MRI. The patients had a mixture of hippocampal, fornix and/or anterior thalamic nucleus damage or had aetiologies (Herpes simplex or hypoxia) consistent with such damage. Even if structurally intact, the hippocampus would not have been functionally normal (Snaphaan et al., 2009). Unconscious relational memory in the controls probably required good hippocampal functional connections to several structures, including the anterior temporal neocortex, which plays a role in semantic memory processing. Strikingly, three high-performing patients, unlike the other eight, showed preserved unconscious relational memory; their residual hippocampus also showed a similar, but weaker, pattern of connectivity to that of controls, whereas the other scanned, but impaired, patients did not. The authors argue that, if the hippocampus and its key connections are working above a critical level, unconscious relational memory can be supported normally, although conscious relational memory is greatly impaired. This strongly suggests that relational memory, whether conscious or unconscious, is critically dependent on processing and perhaps storage mediated by the hippocampal system, but that good unconscious memory, being much weaker, requires far fewer of that system’s resources.

Replications of this important study are desirable in patients with more confined hippocampal system damage, such as otherwise matched patients with different degrees of fornix damage. However, lesion differences outside the hippocampal system in, for example, parts of visual cortex, did not differ consistently between high-performing and impaired patients, which suggests that such damage did not disrupt unconscious relational memory. A problem remains with the relationship between the completely unconscious kind of relational priming and novel associative priming where there is consciousness of studied stimuli. Preserved amnesic performance has recently been reported for the latter kind of priming (Verfaellie et al., 2012). Duss et al. suggest that the amnesics in that study may have had sufficient residual hippocampal functionality to perform normally. This proposal needs testing but would be more plausible if associative priming was less relational in the Verfaellie et al. (2012) study. Although identifying associative or even item memories as relational is not easy, being relational is probably a matter of degree. This warrants further investigation, as do the preserved kinds of non-relational priming, which, unlike item familiarity memory and despite conflicting evidence, Duss et al. argue may be unaffected by perirhinal cortex lesions. Unless this is wrong, in contrast to relational memory, different neural systems must mediate conscious and unconscious non-relational memory.

References