Processing mechanisms of relational and non-relational memory

This doctoral thesis addresses the issues of dissociability of relational memory (RM) and non-relational memory (NRM) and of potential similarity of spatial and nonspatial RM in humans: Whether the hippocampus and perirhinal cortex are critical in RM and NRM, respectively, or whether the hippocampus is key for both RM and NRM, is under discussion. Whether the hippocampus is crucial for general RM or for spatial RM only, is also debated. Our findings from three studies on encoding and retrieval stages of spatial and non-spatial RM as well as NRM are related to the impact of normal ageing and focal lesions on the anatomy and connections of the mediotemporal lobe, prefrontal cortex, and thalamus. Study 1 investigated the potential impact differences of age-associated dysfunctions in memory-critical brain areas on memory during normal ageing. Thus, 106 healthy adults from age 20-76 were assessed in a consecutive age groups design. Spatial and non-spatial RM both declined in the 66-76 years group. This pattern accorded with the presumed course of hippocampal changes across normal ageing. An impairment of NRM commenced earlier in the 51-65 years group. Study 2 analysed the time course of novelty detection on variants of the above memory tasks using event-related potentials for encoding and retrieval in 13 healthy subjects. The event-related potentials related to RM and NRM were dissociable in an early and late time window. A late old/new effect replicated the frequently reported RMassociated old/new effect. The novelty detection P3a effect did not differ in spatial vs. non-spatial RM. Event-related potentials for subsequent hits differed between RM and NRM. Study 3 assessed the potentially differential involvement of the human thalamus in RM and NRM. Ten patients with focal ischemic thalamic lesions were compared to individualised control groups of healthy subjects matched to each individual patient on age and IQ. Six patients showed poorer RM than their respective control samples. None of the ten patients showed a significant deficit on the NRM task. Taken together, our results support the idea of RM and NRM dissociability in terms of distinct onsets of age-related declines, differential RM and NRM-related ERPs during encoding and retrieval, and disproportionate impairment of RM after focal thalamic lesions. Our observations support the notion of a common neuronal mediator for spatial and non-spatial RM with regard to similar spatial and non-spatial RM performances in terms of onsets and courses across the four consecutive age groups and topographically and temporally indistinguishable ERPs related to spatial and non-spatial novelty detection. PART 1: GENERAL INTRODUCTION 4 1 PART 1: GENERAL INTRODUCTION Retrieving the best route to one’s destination and recollecting where things have been placed, exemplify the importance of spatial relational memory (RM), in which spatial contextual details are at hand. Remembering that the alternative route is actually ideal due to current temporary roadwork and recalling the sequence of events illustrate the prominence of non-spatial RM, in which non-spatial contextual details can be retrieved. In contrast, non-relational memory (NRM) refers to the memory when no context is retrievable, albeit the induced feeling of familiarity, such as when one cannot recall the name while encountering a familiar person. The mechanisms mediating spatial and non-spatial RM and NRM in humans are still debated. To better understand these memory processes, we conducted three experimental studies incorporated in this doctoral thesis. The first study investigated the course of these memory processes across the healthy adult life span (Soei and Daum, 2008). The second study focused on the neuronal mechanisms of novelty detection in these memory processes (Soei, Bellebaum and Daum, submitted). The third study examined the role of the human thalamus in RM and NRM (Soei, Koch, Schwarz and Daum, in revision for EJN). The first part of this doctoral thesis covers the broad introduction of the issues of interest, the description of the current state of the art, and finally the outline of the aims of the studies. The second part of the thesis deals with the implications of our findings, and finally the outlook on future research. The third part covers the three studies. PART 1: GENERAL INTRODUCTION Do relational and non-relational memory processes differ? 5 1.1 Do relational and non-relational memory processes differ? Firstly, this section covers the general evidence on the human neuronal network mediating declarative memory. Secondly, the evidence for and against the dissociability of RM and NRM is covered. Thirdly, the issue of similarity of spatial and non-spatial RM is discussed. 1.1.1 The human neural network mediating declarative memory Numerous rodent and non-human primate studies, using electrophysiology and lesion techniques, have corroborated the accounts made in the following paragraphs. However, as the scope of this work is on human spatial and non-spatial RM and NRM, the main evidence cited in the following stems from human and non-human primate studies. This section begins with a definition of the different declarative memory types, and then deals with the neuroanatomical regions and their interconnections within the medial temporal lobe (MTL), prefrontal cortex (PFC) and thalamus in order to facilitate a critical discussion of the behavioural, neuroimaging and lesion evidence concerning these memory processes. In humans, declarative memory refers to memory that can be declared explicitly e.g. RM and NRM, while non-declarative memory refers to memory which is only implicit and does not require effortful remembering, e.g. procedural memory, perceptual representational memory, and classical conditioning, as advanced by Tulving et al. (2000): Procedural memory mediates the acquisition and later performance of cognitive and motor functions. Perceptual representational memory encodes and retains sensory information thought to underlie priming effects. Classical conditioning refers to associative learning. While implicit memory is not within the scope of this thesis, the subdivision of declarative memory into RM and NRM is in focus. In a long history of research since Endel Tulving (1972) introduced this subdivision to psychologists, different researchers have examined RM using synonyms such as episodic, associative, contextual, source, and recollective memory, and NRM using synonyms such as semantic, non-associative, non-contextual, item and familiarity memory (Eichenbaum et al., 2007). The main characteristics of these two memory processes can be described as follows in various dual-process views: RM is a relatively slow, effortful and threshold-like process related to remembering items within a specific spatiotemporal context whereas NRM is a faster and more automatic signal-detection process related to knowing an item without recalling the encoding context (Atkinson and Juola, 1974; Brown and Aggleton, 2001; Eichenbaum et al., 2007; Jacoby and Dallas, 1981; Mandler, 1980; Rugg and Yonelinas, 2003; Tulving, 1972; 2002; Yonelinas, 2001; 2002). PART 1: GENERAL INTRODUCTION Do relational and non-relational memory processes differ? 6 Illustrated in Figure 1 are the differential behavioural effects of experimental manipulations on RM and NRM: Firstly, manipulations of the study time during encoding and long delays during retrieval have large effects on both RM and NRM. Secondly, manipulations of attention and elaboration during encoding have large effects on RM and moderate ones on NRM, while manipulations of attention and speeding during retrieval have large effects only on RM. Thirdly, manipulations of bias and fluency have large effects during retrieval only on NRM, and manipulations of rote repetition during encoding, and perceptual match of the verbs and a brief delay during retrieval have moderate effects only on NRM. Likewise, under speeded processing conditions, participants are faster in accurately deciding whether an item has been previously studied or not, compared to when or where they have studied the item (Hintzman et al., 1998), suggesting that NRM is faster than RM (Yonelinas and Jacoby, 1994; 1996). Moreover, pharmacological manipulation has been shown to selectively impair the putative RMbut not NRM-associated brain potential (Curran et al., 2006). Thus, the experimental manipulations have rather distinct effects on RM and NRM, strongly supporting the notion of dissociable processes (Eichenbaum et al., 2007; Yonelinas, 2002). Since the first short report on lasting memory deficits in a patient with MTL damage (von Bechterew, 1900), it has been established across human and non-human primates and rodents that the MTL is critical to memory. The first formal anatomical study of memory loss after bilateral damage of the MTL in the famous patient H.M. yielded a deficient acquisition of declarative memories without further cognitive deficits (Scoville and Milner, 1957). A later study confirmed the bilateral symmetrical damage in the medial temporal polar cortex, most of the amygdaloid complex, most or all of the entorhinal PART 1: GENERAL INTRODUCTION Do relational and non-relational memory processes differ? 7 cortex (ERC) and approximately half of the partly atrophic hippocampal formation, sparing the perirhinal (PRH) and parahippocampal cortex (PHC), and atrophic cerebellum and mamillary nuclei (Corkin et al., 1997). Subsequent studies supported the idea of memory deficits associated with MTL lesions in humans (Lah and Miller, 2008; Spiers et al., 2001). In non-human primates, damage of the MTL was also linked to memory dysfunctions in pioneering observations (Buckmaster et al., 2004; Mishkin, 1982; Zola-Morgan et al., 1982). However, the early reports suffered from inconsistent classification schemes regarding the neural areas and anatomical projections (subregionally, MTL-neocortically and MTL-subcortically) (Suzuki and Amaral, 2003b). Following a contemporary nomenclature of the boundaries within the primate MTL (Lavenex and Amaral, 2000; Suzuki and Amaral, 2004), the MTL encompasses broadly the PRC, PHC, ERC, and hippocampus (HC). The HC can be subdivided into the dentate gyrus, Ammon’s horn, and subiculum. Information exchange between the MTLs in both hemispheres is mainly enabled by means of the ventral hippocampal commissure between the anterior hippocampi, and the dorsal hippocampal commissure between the parahippocampal gyri (Demeter et al., 1985). Notably, while some researchers exclude the amygdala from the MTL (Murray and Wise, 2004), others have found the same activity pattern during retrieval in the HC and the amygdala (Rutishauser et al., 2006; 2008). Further, the MTL as an entity serving primarily memory functions has been questioned recently, suggesting that each MTL area mediates distinct functions such as memory and perception. For example, the PRC has been suggested to be involved in the perceptual analysis of single items and in the binding of individual stimulus features to a coherent representation of the object, which requires perceptual and mnemonic competence (Bussey et al., 2002; 2006; Bussey and Saksida, 2005; Lee et al., 2005; Murray et al., 2007; Murray and Bussey, 1999), while other investigations have seemingly refuted this account (Levy et al., 2005; Shrager et al., 2006; Squire et al., 2006). These two issues are beyond the scope of this work and thus not covered hereafter. Anatomically, the PRC and PHC cover the lateral bank and the fundus of the collateral sulcus in the anterior and posterior parahippocampal gyrus (PHG), respectively (Pruessner et al., 2002). Both cortices consist of numerous subdivisions (PRC (35, 36cl, 36cm, 36d, 36rl, and 36rm); PHC (TFl, TFm, and TH)) (Lavenex et al., 2004) with distinct chemoarchitectonic and cytoarchitectonic features (Suzuki and Amaral, 2003a). The incoming information has been proposed to be substantially integrated via intrinsic associational connections (Lavenex and Amaral, 2000), which are mostly intrasubdivisional in the PHC and predominantly intersubdivisional in the PRC (Lavenex PART 1: GENERAL INTRODUCTION Do relational and non-relational memory processes differ? 8 et al., 2004). The latter has been suggested to be at a higher level than the PHC in the hierarchy of associational cortices, due to the direction of feedforward and -backward projections (Lavenex et al., 2004). Using retrograde tracing, Suzuki and Amaral (1994a) have elucidated the cortical afferents to the PRC and PHC: The majority of neocortical afferents into the PRC originate from the unimodal visual areas TE and rostral TEO in the inferior temporal cortex, but also from the area TF in the PHC, and little from the orbitofrontal and insular cortices, processing information about the quality of the object i.e. the “what” information. The majority of neocortical afferents into the area TF in the PHC come from the V4, the more caudal TE and TEO, the retrosplenial cortex, the superior temporal sulcus, insular cortex, frontal cortex, and posterior parietal lobes processing polymodal spatial information i.e. the “where” information. Area TH receives afferents from the retrosplenial cortex and the superior temporal sulcus, but not from TE and TEO. These “what” and “where”-information are forwarded to the ERC, situated in the medial PHG (Suzuki and Amaral, 2003a) and serving as the main interface between the PHG and the HC (Lavenex and Amaral, 2000). The subsequent information processing is largely segregated, as the PRC projects mainly to the lateral entorhinal area, while the PHC projects predominantly to the medial entorhinal area, with the projections being mostly reciprocal (Suzuki and Amaral, 1994b). Finally, these two types of information converge into the cornu ammonis 1, 2 and 3, subiculum and dentate gyrus (Witter and Amaral, 1991), again segregated as projections from the lateral entorhinal area ascend into the rostral HC, while the dorsal HC receives input from the medial entorhinal area (Witter et al., 1989b). This segregation of connectivity domains within the ERC and its interconnections with the HC subregions indicates a parallel processing mode in the ERC. In contrast, the information processing appears to be serially in both the PRC and PHC, leaving the information processed in the HC to be highly integrated, supermodal or amodal (Lavenex and Amaral, 2000). However, there are some reciprocal connections between the PRC, PHC, and lateral and medial entorhinal areas (Suzuki and Amaral, 1994b). The HC subsequently feeds the even more highly integrated, multifaceted, and abstract information back to the ERC, then PRC and PHC, and finally, to the neocortical areas from which the input into the MTL originated (Lavenex and Amaral, 2000). The PFC, interacting with the MTL, has also been associated with memory processes (Fletcher et al., 1997; Fletcher and Henson, 2001; Simons and Spiers, 2003). As the PFC projects reciprocally to three different regions within the mediodorsal thalamic nuclei (MD), the PFC itself has accordingly three functional areas (Fuster, 1997). The first area, the dorsolateral PFC (DLPFC), receives projections from the parvocellular lateral region of the MD in the thalamus and cortical projections from the PART 1: GENERAL INTRODUCTION Do relational and non-relational memory processes differ? 9 parietal cortex. The DLPFC not only contributes to successful encoding during RM (Dolan and Fletcher, 1997; Murray and Ranganath, 2007; Staresina and Davachi, 2006) and working memory (Ghashghaei and Barbas, 2001; Wendelken et al., 2008), but is well known for mediating executive functions (Fuster, 1997). These presumably support the strategic encoding and retrieval of contextual details (Squire, 1994; Squire and Zola, 1998). The DLPFC is also part of the “dorsal visual processing stream”, which transmits the perceptual object-locations representations from the visual to the frontal cortices (Goodale et al., 2004). The second area of the PFC refers to the frontal eye fields, which receive projections from the thalamic pars paralamellaris, but have not been linked to memory. The third PFC area is the orbitofrontal cortec (OFC), which receives projections from the magnocellular medial part of the MD. It is part of the “ventral visual processing stream”, and transmits the object-detail representations from the visual to the frontal cortices (Goodale et al., 2004). The OFC receives further projections from cortical and subcortical areas associated with long-term memory and emotional processing, and autonomic functions (Ghashghaei and Barbas, 2001). The anterior PFC does not receive projections from the MD, but its activity has been found to correlate with RM (Allan et al., 2000; 2005b; Simons et al., 2005a). The thalamus has been dubbed the major relay to the cerebral cortex, as almost all knowledge about the external world is based on visual, auditory, somatosensory, cerebellar and other input that have had to pass through the thalamus (Sherman and Guillery, 2006). The thalamus is located on each side of the midline, and entails several nuclei, each transmitting distinct afferent signals via well-defined pathways to a major neocortical area. In the following, mostly the memory-related MD and its pathways will be highlighted (Aggleton et al., 1986; Saunders et al., 2005; Witter et al., 1989a). The MD is the largest structure in the medial thalamus and is located medial to the internal medullary lamina. The ERC and PRC efferents join the ventral amgydalothalamic pathway, stria terminalis and a relay station in the bed nucleus of the stria terminalis to terminate in the medial portion of the MD, which also receives projections from the basal forebrain (Russchen et al., 1987). ERC projections also join the hippocampothalamic pathways (via the fornix and via the caudal thalamic pole) to terminate in the anterior, midline and lateral dorsal nuclei. The midline nuclei project back to the originating area (Insausti et al., 1987), while the MD projects to the PFC and the anterior and lateral dorsal nuclei project to the cingulate cortex and HC (Goldman-Rakic and Porrino, 1985; Ray and Price, 1993; Vogt et al., 1987). PART 1: GENERAL INTRODUCTION Do relational and non-relational memory processes differ? 10 1.1.2 Relational and non-relational memory in humans It is debated, whether there is a functional subdivision of labour in the human MTL, with the HC mediating RM and the PRC subserving NRM (Eichenbaum et al., 1994) or whether the MTL functions as a unified system critical for both RM and NRM (Squire, 1994). This issue has been investigated using neuroimaging methods such as non-invasive electrophysiology to detect the time-varying electric fields at the scalp surface generated by synchronously active neuron populations, with the measure most used in memory research being the event-related potential (ERP) (section 1.2.2.1). This section focuses on evidence from another neuroimaging technique and patients studies. Whilst non-invasive functional magnetic resonance tomography (fMRI) does not test whether a particular region is necessary for the task at hand (Logothetis, 2008), it offers fine-grained spatially-resolved localisation of the neural network, and conveys the neural correlates of cognitive task processing. Using this technique, differential activations of the HC vs. PRC / anterior PHC have been found to correlate with either RM or NRM: The HC has been suggested to play a role in RM (Davachi et al., 2003; Davachi and Wagner, 2002; Giovanello et al., 2004; Pihlajamaki et al., 2004), whereas the PRC has been assumed to support NRM (Brown and Aggleton, 2001; Henson et al., 2003; Montaldi et al., 2006; Pihlajamaki et al., 2004). A recent study showed that the PRC activity was predictive of subsequent NRM even under associative conditions where word pairs were encoded as a single conjunctive item (Haskins et al., 2008), which accords to behavioural findings (Diana et al., 2008). This is seemingly in good agreement with the notion, that the HC and adjacent cortex mediate RM and NRM, respectively (Eichenbaum et al., 1994). However, following Kopelman et al. (2007), it may be possible, that the correlational nature of fMRI observations in healthy participants and focal lesion findings in patients point in different directions: Single and group studies of hypoxic patients with lesions affecting either the HC or of a patient with PRC resection due to intractable epilepsy, have as yet not yielded a clear picture with respect to the notion of RM and NRM dissociability. On the one hand, disproportionate impairment depending on the lesion site has been observed (Bowles et al., 2007; Giovanello et al., 2003; Holdstock et al., 2005; Mayes et al., 2004; Turriziani et al., 2004; Yonelinas et al., 2002). On the other hand, proportionate deficits on RM and NRM after critical lesions or stimulations have also been reported across a variety of material and group-matched NRM performance (Cipolotti et al., 2006; Coleshill et al., 2004; Gold et al., 2006; Kopelman et al., 2007; Kopelman and Stanhope, 1998; Manns et al., 2003; Reed and Squire, 1997; Stark et al., 2002; Stark and Squire, 2003; Wais et al., 2006; Wixted and Squire, 2004). Interestingly, the MRI volume of either the HC or PHC showed the highest correlation with both RM PART 1: GENERAL INTRODUCTION Do relational and non-relational memory processes differ? 11 and NRM measures (Kopelman et al., 2001; 2007; Kramer et al., 2005). Thus, after early single-process theories (Gillund and Shiffrin, 1984; Hintzman, 1988; Murdock, 1982), a contemporary single-process view suggests that both memory components differ only in quantity of memory strength and are both critically mediated by the HC and adjacent cortex (Slotnick and Dodson, 2005; Squire, 1994; Squire et al., 2007; Squire and Zola, 1998; Wixted, 2007). This perspective was recently supported by an fMRI study showing that both the HC and the PRC predicted memory strength of subsequently remembered information (Shrager et al., 2008). According to Kopelman et al. (2007), the dispute cannot be easily resolved at present, as according to Eichenbaum et al. (2007), studies with (severe) hypoxic patients may be potentially compromised by damage outside the HC. Further, present structural MRIs may not be sensitive enough to reveal all damage apparent in post-mortem histology and may lead to conflicting results (Holdstock et al., 2008; Rempel-Clower et al., 1996). 1.1.3 Spatial vs. non-spatial relational memory in humans It is disputed, whether the human HC is confined to spatial memory and mapping of large-scale space consistent with the cognitive map theory (O'Keefe, 1999; O'Keefe and Nadel, 1978) or whether it plays a general role in RM, with spatial processing being only one aspect of RM (Cohen et al., 1997; Eichenbaum et al., 1999; 2001; 2004). Positron emission tomography studies with healthy human subjects focusing mainly on spatial processing and navigation have supported the idea of a more important role of the HC in spatial compared to non-spatial RM processing: Right-sided activation of the HC/PHG has been found to correlate with recall of the navigation routes and recognition of object locations with landmark cues (Johnsrude et al., 1999; Maguire et al., 1997). FMRI studies also supported the spatial processing view: Activations of the HC have been found to correlate with spatial but not social RM, and with virtual navigation based on spatial RM (Kumaran and Maguire, 2005; Rauchs et al., 2008; Ross and Slotnick, 2008). Spatial novelty detection correlated with the anterior HC activation, while posterior HC was associated with location processing in a virtual environment task (Doeller et al., 2008). Human lesion studies focusing on spatial processing and navigation have provided supporting evidence: Holdstock et al. (2000) showed in a patient with a bilateral HC lesion that the HC is more important for consolidation of allocentric as opposed to egocentric spatial memory. Astur et al. (2002) described ten patients with unilateral removal of the HC due to medically intractable epilepsy; who were significantly slower to PART 1: GENERAL INTRODUCTION Do relational and non-relational memory processes differ? 12 find the hidden platform in a virtual Morris water maze. Moreover, after training, not a single patient employed a spatial strategy. Brandt et al. (2005) reported deficient spatial RM and navigation in a virtual Morris water maze for ten patients with chronic acquired vestibular loss resulting in HC atrophy. Maguire et al. (2006) showed in a patient that chronic bilateral HC damage impaired navigation based on remote spatial representations. In contrast, fMRI studies with healthy human subjects, in which RM of perceptually and conceptually distinct items was assessed, have supported the hypothesis of a general role of the HC in RM: Duzel et al. and Fenker et al. (2001; 2003; 2005) showed in healthy human volunteers, and in one patient with early HC damage, that the HC plays a role in spatial, non-spatial and emotional RM. Eldridge et al. (2000; 2005) reported that the dentate gyrus, CA2 and CA3 were more active during verbal and nonverbal RM formation, while the subiculum was more active during retrieval. Kohler et al. (2005) reported that the right-sided middle HC was more active during both nonverbal spatial and non-spatial RM, while Davachi and Wagner showed that the HC proper activation correlated with relational processing of word triplets (2002). Studies of human patients with HC lesions assessing different types of RM have also been conducted: Cave and Squire (1991) reported spatial and non-spatial RM deficits of seven patients with lesions of the hippocampal formation, suggesting that the HC is important for rapid acquisition of relational information. Kumaran (2007) found in four patients with selective HC damage similar impairments on a spatial and non-spatial configural learning task. Shrager et al. (2007) examined six patients with damage of the HC and found similarly impaired performance on tests of object location memory with and without a view-point shift between study and test. Spiers et al. (2001) reviewed the literature on the consequences of damage of the HC on anterograde amnesia, corroborating the account that the patients’ RM deficits are of a general nature, as the deficits span materials like words and paired associates, sentences and stories, pictures of objects and scenes and combinations of words and pictures. This account has also been supported by rodent and non-human primate studies that were not mentioned for the sake of brevity (for extensive reviews Eichenbaum et al., 1999; Eichenbaum and Fortin, 2005; Wood et al., 1999). In summary, the neuroimaging and patient studies addressing not only spatial but also non-spatial and emotional RM using perceptually and conceptually distinct items in the verbal and nonverbal domain support the notion of a more general function of the HC (Cohen et al., 1997; Eichenbaum et al., 1999; 2001; 2004). PART 1: GENERAL INTRODUCTION State of the art 13 1.2 State of the art As outlined in the previous sections, the evidence for the account that different neuronal networks are engaged in RM and NRM comes from manifold lines of investigation: Differential behavioural sequelae of damage of critical areas, dissociable effects of experimental manipulations, and distinct correlating activation patterns in neuroimaging studies. This section extends the state of the art concerning our three studies beyond the published or submitted versions of the articles. 1.2.1 Ageing and spatial and non-spatial relational and non-relational memory The first study focused on age-related changes of spatial and non-spatial RM and NRM. In the next subsections, the scope is extended to general age-related changes in memory, associated structural changes in memory-related brain structures, and finally the caveats of existing experimental approaches. 1.2.1.1 Age-related functional changes in memory As introduced, declarative memory comprises RM and NRM, while nondeclarative memory comprises procedural memory and perceptual representational memory (section 1.1.1). These distinct memory processes are differently affected by rising age. Procedural memory and perceptual representational memory remain fairly stable across healthy ageing, while declarative memory typically declines with rising age to different extents (Balota et al., 2000). Several candidate mechanisms underlying the age-related memory decline have been proposed. The concept of general slowing and resulting deteriorated memory performance (Salthouse, 1996) is difficult to reconcile with similar patterns of fewer benefits and more costs for older adults on both free and forced recall tests and on timed and self-paced tests (Henkel, 2007). The concept of reduced processing resources or attentional capacity in ageing (Kahneman, 1973) assumes that difficult cognitive tasks require more resources than simpler tasks, but the concept has been criticised as too vague without addressing the neural correlates, and it cannot account for the entire age-related memory decline (Salthouse et al., 1988). This raises the question which additional factor may play a pivotal role, possibly a specific deficit to associate unrelated pieces of information (Naveh-Benjamin, 2000). A long tradition of psychological research employing recall and single-item recognition tests has shown, that RM for personal experiences undergoes a more pronounced age-related decline than the relatively stable NRM for the general knowledge about the world (Balota et al., 2000; Yonelinas, 2002) as observed in extreme-group comparisons (old vs. young participants) (Naveh-Benjamin, 2000; Naveh-Benjamin et al., 2003; 2004), or that age effects are reliably more variable in RM relative to NRM PART 1: GENERAL INTRODUCTION State of the art 14 (Spencer and Raz, 1995). Results of studies showing that RM and NRM were proportionately reduced by healthy ageing in extreme-group comparisons (Mark and Rugg, 1998; Schacter et al., 1997) have been linked to ceiling effects (Yonelinas, 2002). However, this view has been challenged in a recent within-subjects-designed study using three different process estimation methods for RM and NRM in an extreme-group comparison (Prull et al., 2006). Two process estimation methods revealed that both RM and NRM estimates were negatively affected by ageing, while the third method found age-related decreased RM but invariant NRM estimates. All three methods confirmed the proposal of a decline of RM with ageing (Prull et al., 2006). Support for the age-related decline of RM and NRM came from a study employing the process-dissociation procedure in a large extreme-group comparison (Toth and Parks, 2006). In another extreme-group comparison, age-related significant deficits in spatial and non-spatial RM have been reported, but interpretation may have suffered from a ceiling effect in the later task (Driscoll et al., 2003). Our own ageing study across four consecutive age groups of the healthy adult life span (section 3.1) showed a differential course of decline extending the notion of age-related decrement in spatial and non-spatial RM and NRM. The last three studies are powerful as different estimation methods were employed yielding converging results and they do not suffer from ceiling effects (Prull et al., 2006; Soei and Daum, 2008; Toth and Parks, 2006). 1.2.1.2 Age-related structural changes in the relevant neural correlates This section covers the structural and morphological impact of healthy ageing in the memory-related brain structures outlined above (section 1.1.1). Pathological ageing, such as in Parkinson’s or Alzheimer’s disease, and age-related changes in the transmitter systems and electrophysiological neuronal properties are not covered for the sake of brevity. Within the MTL, research has been mostly conducted on the HC, but less on ERC, PRH or PHC. Beginning with the HC, it has been reported that age-related HC volume losses were significant (Raz et al., 2004a; 2004b) and accelerated relative other brain regions (Jernigan et al., 2001; Schuff et al., 1999). It has been further observed that HC volume, on average, decreased annually at a faster pace, with a moderate decline in the adults < 50 years, but a twice as fast decline in the older adults (Raz et al., 2004a; Raz et al., 2004b; Raz et al., 2005). Similarly, a linear relationship for the rate of HC volume loss in healthy adults in their sixties and seventies has been reported (Cohen et al., 2006). However, stable HC volume across ageing has also been found (Sullivan et al., 2005; Van Petten et al., 2004). Studies using stereological methods suitable for estimating the total number of neurons in the brain have reported a substantial loss in the PART 1: GENERAL INTRODUCTION State of the art 15 HC subiculum (~52%), in the hilus of the dentate gyrus (~31%), but almost none in the CA1 region across the age range (West, 1993; West et al., 1994). A marker for the neuronal integrity is N-acetyl-aspartate as evident in MR spectroscopy (Barker, 2001), which levels have been shown to be decreased in the HC across the healthy adult life span (Schuff et al., 1999), and in healthy old compared to young subjects (Driscoll et al., 2003). While increased dendritic extent in the dentate gyrus in healthy old relative to middle-aged subjects has been reported (Flood et al., 1987), apical and basal dendritic branching of pyramidal neurons in the CA1 region were stable across the healthy adult life span (Hanks and Flood, 1991). Taken together, the volume of the human HC has been found to decline curvilinearly with rising age, with an accelerated decrease after age 50 to 60 (Cohen et al., 2006; Driscoll et al., 2003; Raz et al., 2004a; 2004b; 2005; Schuff et al., 1999), but conflicting results call for a cautious interpretation (Sullivan et al., 2005; Van Petten et al., 2004) as the evidence for a positive relationship between HC volume and RM performance is variable and surprisingly weak (Schiltz et al., 2006; Van Petten, 2004). Differential age-related changes across the adult life span have been reported for the number of neurons (West, 1993; West et al., 1994) and for the extent of dendrites of the HC subregions (Flood et al., 1987; Hanks and Flood, 1991). Decreased levels of N-acetyl-aspartate in the HC have also been observed in the elderly (Driscoll et al., 2003; Schuff et al., 1999). Concerning the other MTL regions, it has been reported that the ERC volume does not change with age, and declines only in pathological but not in healthy ageing (Small et al., 2002). Further, ERC volume decreased annually on average only minimally (0.32%), with the adults < 50 years showing virtually no annual pace of decline (0.11%), unlike the more affected older adults (0.53%) (Raz et al., 2004b; 2005). However, these findings are at odds with an observation of age-related ERC volume reduction across the adult life span (Simic et al., 2005). For the PRC, a stable volume across healthy ageing has been reported (Insausti et al., 1998a; 1998b), although little is known about the functional integrity of the PRC (Burke and Barnes, 2006). With respect to the PHC, no age-related volume differences have been observed (Raz et al., 1997; Van Petten et al., 2004). In the PFC, the DLPFC and OFC volume declined across a five-year change with a high annual percent change (0.91%, and 0.85%) (Raz et al., 2005). The inferior frontal and OFC volume declined across the adult life span as assessed via MRI (Resnick et al., 2003), and PFC volume shrinkage has been reported in a sample of healthy humans across the adult life span (Raz et al., 1998). All volumes of frontal lobe regions have been shown to be affected strongly but differentially by age using differential estimation methods (Jernigan et al., 2001; Raz et al., 1997; Tisserand et al., 2002). An average PART 1: GENERAL INTRODUCTION State of the art 16 linear decline of 5% per decade within the PFC after the year 20 has been estimated (Raz et al., 2004a). For old relative to young non-human primates, a substantial 32 % reduction in neuron number in all layers of the DLPFC (BA 8A) and a conserved neuron number in BA 46 of the PFC has been reported (Smith et al., 2004). Decreased dendritic extent in the medial PFC has been reported in healthy humans across the adult middle to old life span (de Brabander et al., 1998). Successively lower synaptic density has been observed in layers I to VI particularly in the PFC and declining with age in humans (Liu et al., 1996) and non-human primates (Bourgeois et al., 1994). Driscoll et al. (2003) further observed decreased N-acetyl-aspartate levels in frontal white matter in the elderly, while volume of the frontal regions was not assessed. The greatest age-related declines in white matter integrity have been observed in the PFC in healthy humans using diffusion tensor imaging in extreme-group comparisons (Head et al., 2004; Pfefferbaum et al., 2005), across the adult life span (Bartzokis et al., 2003), although all brain regions show alterations with healthy ageing in humans (Head et al., 2004; Madden et al., 2004a; O'Sullivan et al., 2001). Increased apparent diffusion coefficients in frontal white matter have been found (Abe et al., 2002). Taken together, the volume of the human PFC has been found to decline linearly with rising age (Jernigan et al., 2001; Raz et al., 1998; 2005; Resnick et al., 2003; Tisserand et al., 2002). Age-related detrimental impact on the number of neurons (Smith et al., 2004), the extent of dendritic branching (de Brabander et al., 1998), and on the synaptic density (Bourgeois et al., 1994; Liu et al., 1996) of the PFC subregions has been reported for humans. Further, decreased white matter integrity in the PFC in elderly people has been observed (Bartzokis et al., 2003; Head et al., 2004; Madden et al., 2004a; O'Sullivan et al., 2001). In comparison, much less is known about the thalamus, which shows a linear volumetric decline in age (Sullivan et al., 2004; Van Der Werf et al., 2001; Walhovd et al., 2005), although observations of no or little volumetric change have also been made (Jernigan et al., 1991; 2001). There was a significant increase in mean diffusity, decrease in white matter integrity and volume in the human thalamus with advancing age (Abe et al., 2002; 2008). To further elucidate the issues of the potential dissociability of RM and NRM (section 1.1.2) and the hippocampal role in spatial RM vs. general RM (section1.1.3), our first study linked the available volumetric observations on ageing-related changes in the brain regions relevant to RM and NRM (HC, PRC, PFC and thalamus) with a narrow-age cohort design covering the adult life span in four consecutive age groups. PART 1: GENERAL INTRODUCTION State of the art 17 1.2.1.3 Caveats of previous experimental approaches All of the above studies used a cross-sectional approach, which might overestimate age-related changes due to possible cohort differences due to historical influences like education, culture and socioeconomic status (Hofer and Sliwinski, 2001), but they are the most labour-efficient comparisons (Hedden and Gabrieli, 2004). They can be classified into extreme-group comparisons and assessment across the adult life span: On the one side, most of the behavioural studies on age-related impact on memory used extreme-group comparisons (Driscoll et al., 2003; Mark and Rugg, 1998; Naveh-Benjamin, 2000; 2003; Naveh-Benjamin et al., 2004; Prull et al., 2006; Schacter et al., 1997; Toth and Parks, 2006). On the other side, many of the anatomical and neuroimaging studies on age-related impact on the neural correlates of memory were assessed across the adult life span (Abe et al., 2002; 2008; Bartzokis et al., 2003; de Brabander et al., 1998; Hanks and Flood, 1991; Jernigan et al., 1991; 2001; Raz et al., 1997; 1998; 2004a; 2005; Schuff et al., 1999; Simic et al., 2005; Small et al., 2002; Sullivan et al., 2005; Tisserand et al., 2002; Van Der Werf et al., 2001; Walhovd et al., 2005; West, 1993; West et al., 1994). A better approach to infer age-related impact is the longitudinal assessment, which does not suffer from cohort differences, but in return may underestimate agerelated changes, because they are potentially confounded by selective attrition at older ages. Specifically, behavioural studies may be confounded by practice effects at younger ages and/or short retest intervals (Hedden and Gabrieli, 2004). Due to the labourconsuming nature, the rare longitudinal behavioural studies also do not tap RM and NRM in greater detail (Beason-Held et al., 2008; Singer et al., 2003), and the few anatomical longitudinal studies cover less than ten year retest intervals (Raz et al., 2004b; Resnick et al., 2003). While both approaches have their flaws, it has been proposed to use a sequential narrow-age cohort approach as a useful means for evaluating associations between ageing-related changes (Hofer and Sliwinski, 2001). Therefore, we employed a narrowage cohort design covering the adult life span from 20-76 years in 106 healthy adults, allowing comparison of the course of progression of RM and NRM with published structural data, elucidating not only the extreme but also two intermediate age groups (Soei and Daum, 2008). PART 1: GENERAL INTRODUCTION State of the art 18 1.2.2 Electrophysiology of spatial and non-spatial relational and non-relational memory As outlined above, the evidence for the notion that dissociable neuronal networks are engaged in RM and NRM stems from several lines of research: Dissociable behavioural sequelae after damage of relevant areas, dissociable effects of experimental manipulations, differential correlating activation patterns in neuroimaging studies, and distinct impact of healthy ageing. The second study focused on the electrophysiological correlates of novelty detection in spatial and non-spatial RM and NRM. In the following, the scope is extended to findings on ERPs of novelty detection in general, of RM and NRM, and finally the caveats of existing experimental approaches. 1.2.2.1 Electrophysiological correlates of relational and non-relational memory Research on the ERPs of RM and NRM during retrieval has yielded topographically, functionally, and temporally dissociable potentials: The RM-associated late old/new effect typically peaks at left posterior electrode sites from 400-800 ms, whereas the NRM-associated early old/new effect (also dubbed FN400) typically peaks at mid-frontal sites from 300-500 ms post-stimulus onset (Allan et al., 1998; Donaldson and Rugg, 1998; Friedman and Johnson, Jr., 2000; Johnson et al., 2008; Paller et al., 1999; Rugg and Curran, 2007). The RM-associated late old/new effect was found to be more positive for hits relative to correct rejections (Sanquist et al., 1980; Warren, 1980), for correct relative to incorrect RM judgements (Johnson et al., 2008; Rugg et al., 1998), and for successful relative to unsuccessful source memory classifications (Senkfor and Van Petten, 1998; Van Petten et al., 2000; Wilding and Rugg, 1996). The same ERP effect was found for Remember relative to Know responses (thought to index the participant’s subjective judgement whether the quality of the memory is rather RMor NRM-based, respectively) (Curran, 2004; Duarte et al., 2006; Paller et al., 1999; Woodruff et al., 2006). Further, this effect has been observed to be larger for interthan for intra-item associations (Jager et al., 2006), and to be topographically and functionally dissociated from parietal late effects related to response confidence and stimulus probability (Curran, 2004; Curran and Hancock, 2007; Rugg and Curran, 2007; Woodruff et al., 2006). The NRM-associated early old/new effect has been shown to be more negative for correct rejections relative to hits (Curran, 2000; Curran and Cleary, 2003; Rugg et al., 1998), for Know relative to Remember responses (Curran, 2004; Woodruff et al., 2006), and to increase with confidence strength (Woodruff et al., 2006). Further, this effect has PART 1: GENERAL INTRODUCTION State of the art 19 been shown to be larger for intrathan for inter-item associations (Jager et al., 2006). NRM has also been observed to be dissociated from conceptual priming in terms of distinct ERPs (Rugg et al., 1998) and activated cortical networks (Voss et al., 2008). RM and NRM have further been shown to be topographically dissociable in terms of subsequent memory effects during encoding, although the number of reports is minimal and neither temporally nor topographically consistent (Rugg and Curran, 2007). Subsequent NRM-based judgements were associated with a left-lateralised enhanced positivity at frontal electrode sites from 300 to 450 ms, whereas subsequent RM was associated with a right-lateralised positivity at frontal electrode sites from 300 to 450 ms and bilateral activity from 450 to 600 ms post-stimulus onset (Duarte et al., 2004). Another study yielded that subsequent face NRM was predicted by right-sided neural activity at parietal electrode sites from 600-800 ms, while subsequent face RM was predicted by bilateral slow-wave potentials at parietal electrode sites from 300 ms poststimulus onset (Yovel and Paller, 2004). However, the latter study reported associations of both RM and NRM with bilateral, parietal-maximum brain potentials at retrieval, albeit smaller amplitudes and shorter durations for NRM. Taken together, the ERPs of RM and NRM have been shown to be topographically, functionally, and temporally dissociable during retrieval and encoding and thus add further support to the idea of dissociable neuronal networks for RM and NRM (Eichenbaum et al., 2007). 1.2.2.2 Electrophysiological correlates of novelty detection Novelty detection is important for updating established knowledge in the face of change and utilises previously formed memory representations (O'Keefe and Nadel, 1978). Concerning the mediating neural regions, a distributed network has been implicated for novelty processing (Halgren et al., 1998; Knight and Nakada, 1998; Nyberg, 2005; Ranganath and Rainer, 2003). Using fMRI, the HC and parahippocampal region have been proposed to be pivotal for relational and non-relational novelty detection, respectively (Duzel et al., 2003; Kohler et al., 2005; Kumaran and Maguire, 2007; Pihlajamaki et al., 2004). Recorded potentials from intracerebral electrodes prior to surgery for intractable epilepsy have revealed further regions critical in novelty processing (Baudena et al., 1995; Clarke et al., 1999a; 1999b; Halgren et al., 1995a; 1995b; 1998): The lateral PFC, inferior temporal, entorhinal, orbitofrontal regions, anterior cingulate cortex, parietal lobes, primary and extrastriate visual areas, and premotor and motor areas. While these observations stem from assessing severe epileptic brains and inference to healthy neural processes thus warrants caution, PART 1: GENERAL INTRODUCTION State of the art 20 standard ERPs and fMRI in healthy participants supported this evidence (Dudukovic and Wagner, 2007; Knight, 1984; Williams et al., 2007). Regarding the ERPs of novelty detection in RM, the P300 potentials (Sutton et al., 1965) have been proposed to reflect the different orienting responses, through which attentional resources are automatically allocated towards the stimulus changing the memorised context (Corbetta and Shulman, 2002; Sokolov, 1963). The P300 potentials typically peak 200-500 ms post-stimulus onset with positive polarity, with the time range depending on stimulus modality, subject age, task conditions etc. Based on studies using variants of the oddball paradigm; in which infrequently presented targets have to be detected among the standard stimuli, the recorded P300 potentials are thought to index neural activities underlying revision or updating of the mental representation induced by the incoming stimuli (Donchin, 1981). After initial sensory processing of the stimuli, an attentional comparison evaluating the representation of the previous event takes place in working memory, a process distinct albeit related to the auditory or visual mismatchnegativity (Heslenfeld, 2003; Kujala et al., 2003). Subsequently, if the representation of the previous event remains unchanged, only sensory evoked potentials are elicited (N100, N200, P200), if a change is detected, attentional processes update the representational context and elicit the P300 (Polich, 2003; 2007). The P300-effects have been observed in many mammalian species (dolphins, rabbits, rats, dogs, cats, squirrel and macaques monkeys, and humans) (Paller, 1994). The P300-effects have been broadly subdivided into the P3a and P3b, with the latter being related to subsequent attentional resource activations promoting memory operations in temporal and parietal regions (Brazdil et al., 2001; 2003; Knight, 1996), and which will not be covered here, as the focus is on novelty detection. The P3a is the potential most directly related to novelty detection (Courchesne et al., 1975; Squires et al., 1975). Polich (2007) has reviewed the available evidence on the P3a, the novelty P3 and no-go P300, and has proposed that these are all variants of the same potential, with the scalp distributions varying as a function of attentional and task demands. The P3a-effect typically peaks 200-350 ms post-stimulus onset with a positive polarity, and presumably originates from stimulus-driven frontal attention mechanisms during task processing related to detection and rapid orientation to novel events and stimuli (Fabiani and Friedman, 1997; Friedman et al., 2001; Knight and Nakada, 1998; Polich, 2007; Ranganath and Rainer, 2003; Soltani and Knight, 2000). Three characteristics of the P3a have been observed (Friedman et al., 2001): P3a responses habituate across successive presentations of novel items, i.e. the more predictable, the smaller the magnitude of the response (Courchesne et al., 1975; Sokolov, 1990). P3a PART 1: GENERAL INTRODUCTION State of the art 21 responses have been shown to be robust even when the detected deviance was task irrelevant or unattended (Courchesne et al., 1975; Friedman et al., 1998; Ranganath and Paller, 1999; Squires et al., 1975). Similar P3a responses have been observed for auditory, visual and somatosensory modalities (Comerchero and Polich, 1999; Knight, 1996; 1984). The question, whether the P3a responses are also similar for different contexts (e.g. spatial vs. non-spatial), was addressed in the second study of this thesis (section 3.2). 1.2.2.3 Limitations of previous paradigms Although ERPs lack the spatial resolution required to localise the neural substrates of different processes, they help to determine whether neural correlates of RM and NRM differ qualitatively (indexed by ERP effects that differ in scalp distribution as opposed to simple magnitude differences), as would be expected if RM and NRM have different neural substrates (Rugg and Yonelinas, 2003). The studies discussed above have not addressed the question, whether the mechanisms underlying spatial and nonspatial RM are dissociable regarding novelty detection effects (e.g. the P3a) and the encoding phase has so far not been analysed in terms of subsequent RM and NRM effects. 1.2.3 The human thalamus in spatial and non-spatial relational and non-