<\body> <\with|par-right|0cm> <\with|par-left|0cm> <\with|par-hyphen|professional> <\with|font-base-size|12> <\with|par-sep|1fn> <\with|par-line-sep|0.025fns> <\with|par-mode|center> \; NEURODYNAMICS OF DECLARATIVE MEMORY\ CONSOLIDATION \; By \; Jeremy Lee Smith \; \; A DISSERTATION\ \; Submitted to\ Michigan State University\ in partial fulfillment of the requirements\ for the degree of \; DOCTOR OF PHILOSOPHY\ \; Program in Neuroscience\ 2006 <\with|par-mode|center> \ \; \; NEURODYNAMICS OF DECLARATIVE MEMORY\ CONSOLIDATION \; By \; Jeremy Lee Smith \; \; <\with|par-left|1.5in.> The present study endeavors to implement FMRI and an innovative approach to temporal synchrony-type binding processes -- Rosenni bootstrapping -- to assess the brain's ability to optimize memory network assemblies with consolidation. Rosenni bootstrapping follows from iEEG/FMRI studies by Logothetis et al. and is used to analyze power and frequency covariance between distributed brain regions, permitting a systems-level description of dynamic neural assemblies formed in the encoding and retrieval of a memory. In this dissertation the evidence for a binding problem, or collection of problems, is presented from the phenomenological, cognitive, neuroscience, and neuropathological perspectives, and temporal synchrony-type binding processes are discussed with respect to the concepts of temporal structure and cellular assemblies. These concepts are then related to neural oscillators and their role in the encoding, consolidation, and retrieval of memories, and the anatomy and functional architecture of declarative and other memory systems are outlined. Past and present FMRI and electrophysiological methodologies are also discussed and related to the study of changes in connectivity and binding processes as a whole. We utilized a paired-associate task to examine changes in functional modularity and connectivity with declarative memory consolidation. This task required subjects to learn to associate visually presented words with abstract pictures, then identify correctly-matched, incorrectly-matched, and control pairings during a testing session. Improvements in behavioral performance were exhibited in post-consolidation subjects relative to their pre-consolidation counterparts, particularly with respect to reaction time for evaluation of presented stimuli and retrieval of correct pairings. Additionally, these improvements in performance were correlated with increased recruitment of known attentional and executive areas, such as the anterior cingulate, superior parietal lobule, and prefrontal cortex and decreased recruitment of working memory areas, lower visual areas, and portions of the medial temporal lobe. We also investigated changes in neural recruitment (with respect to activation volume and the number of VROD seed voxels isolated in each region), metabolism, mean correlation between activated voxels and the paired-associate task, and BOLD signal information load (voxel Shannon entropy). Close correlations were found between information load, clustering, and voxel recruitment, but not metabolism or voxel-task correlation, suggesting that the latter two measures are driven by another explanatory variable mediating the effect of network optimization through memory consolidation. Furthermore, we report putatively consolidation-related decreases in recruitment in visual, motor, attentional, executive, and long-term memory regions; global increases in voxel-task correlation; increases in information load in long-term memory and attentional-executive areas; decreased metabolism in visual areas; and decreases in clustering in visual, motor, working memory, and attentional and executive regions, but not long-term memory regions, with consolidation. These findings are discussed in the context of basic scientific research in memory network dynamics, as well as the methodological contributions to neuroscience. \; \; \; \; \; \; \; \; \; \; <\with|par-mode|right> Copyright by JEREMY LEE SMITH 2006 \; \; \; \; <\with|par-mode|center> ACKNOWLEDGEMENTS \; \; The author wishes to thank Mr. William Smith, Ms. Barbara Parker Smith, Mr. and Mrs. Donald and Marcia Brown, Mrs. Katey Smith, and Mr. Kenneth M. Cobb for their ongoing support during the writing of this dissertation; Drs. Alessandra M. Passarotti, Laura Symonds, Sue Barman, Christine Larson, and Cheryl Sisk for indispensable advice and guidance; Dr. David Zhu, Mr. Jeremy Grounds, and Mr. Dave McFarlane for unremitting patience and assistance with daily technical issues; Dr. Craig Branch, for critical feedback on this manuscript; and the students and faculty of the Neuroscience Program, without whose encouragement and helpful comments this work would likely have suffered exceedingly. <\with|par-sep|1fn> <\with|par-mode|center> AIMS AND HYPOTHESES In the present dissertation, mnemonic neurodynamics are investigated with respect to static, or , differences in behavior and brain activity between pre-consolidation and post-consolidation subjects, as well as differences in inter-regional signal correlations, or differences, at pre- and post-consolidation time points. \ In particular, these differences were evaluated in two phases: in the first phase, conventional FMRI statistics -- percent signal change from baseline and activation volume -- were applied in order to calculate and the on a delayed match-to-sample variant (paired-associate) task, and two other statistical assessments -- voxel-task cross-correlation (VTCC, the temporal correlation between stimulus presentation and voxel BOLD signal response) and BOLD signal information load (the Shannon entropy of a voxel time series, or voxel TSE) -- were employed to aid in the interpretation of the data returned by the conventional metrics, as discussed below. \ In the second phase, a novel seed voxel extraction method, Rosenni bootstrapping, was applied to the BOLD signal-derived activity clusters isolated in phase I in order to find voxels that putatively represented the loci of neural activity underlying the spatially diffuse activity clusters (see the chapter). \ The BOLD signals from these seed voxels were then subjected to pairwise coherence and correlation analyses in order to calculate the putatively binding-related changes in temporal correlation () between the BOLD signals of pairs of seed voxels with memory consolidation. \ Furthermore, the number of Rosenni voxels -- which, again, are postulated to represent neural activity loci () -- were calculated for each cytoarchitectonically-defined Brodmann area so as to quantify the spatial reorganization of activity with consolidation, as discussed below. \ <\with|par-first|2fn> In this study, a paired-associate task (described below) was utilized to examine changes in functional modularity and connectivity with declarative memory consolidation. This task required subjects to learn to associate visually presented words with abstract pictures, then identify correctly-matched, incorrectly-matched, and control pairings during a testing session. Here, the two specific aims and associated hypotheses of this dissertation will be examined, based on the introductory material presented in the preceding chapters. \ A discussion of the specific aims will be followed by summary tables (Tables and ) with the relevant expectations. The present study was broadly divided into two specific aims with the goal of assessing changes in behavior, functional architecture, and connectivity in pre- and post-consolidation subjects. Subjects were given a computer-based training session, during which they learned to associate pairs of words and abstract pictures. This training session was followed by an FMRI session, during which the subject was asked to perform a simple match/mismatch task based on the pairings learned in the training session. Two FMRI time points were established: a pre-consolidation time point (TP) immediately after the training session and a post-consolidation time point (TP) seven days after the initial training session. <\with|par-first|2fn> . \ Two hypotheses were proposed under Specific Aim 1. \ First (SA. I., H. 1.), it was hypothesized that , such that subjects' reaction time and accuracy would improve significantly over the consolidation period, and second (SA. I., H. 2.), that (defined as the extent, or , and localization, or , of task-correlated voxels) >. Consequently, it was predicted that the number of task-correlated voxels (activation volume) would decrease in long-term memory areas with consolidation due to more efficient encoding of the memory trace, and, furthermore, that the localization of activity would change in long-term and attentional areas due to changes in, for example, orienting of attention to more global visuospatial stimulus features. In particular, expected changes in locality were categorized as:\ <\enumerate-alpha> from local features to global features with consolidation.> \ These were predicted to be evinced by bilateral activation in pre-consolidation subjects and right-hemisphere dominance (fewer left-hemisphere and more right-hemisphere clusters: Sergent, 1982; Brown and Kosslyn, 1993; Ivry and Robertson, 1998; Yamaguchi, Yamagata, and Kobayashi, 2000; but . Johannes, Wieringa, Matzke, and Munte, 1996), particularly in components of the ventral visual stream (Knierim and van Essen, 1992; Polat, Mizobe, Pettet, Kasamatsu, and Norcia, 1998; Haxby, Gobbini, Furey, Ishai, Schouten and Pietrini, 2001; Lerner, Hendler, Ben-Bashat, Harel, and Malach, 2001) and the parahippocampal gyrus (Hellige, 1996; Brown and Kosslyn, 1993) in post-consolidation subjects;\ s (referred to in the present dissertation as ; see also Tulving, 1983). These were predicted to be evinced by \ a decrease in the number of clusters, analogous to the number of discrete processors, in lower visual areas. \ In contrast, the number of clusters in the left (Farah, 1989) and right (Bruyer, 1988; Brown and Kosslyn, 1993) parahippocampal gyrus and the ectorhinal, perirhinal and entorhinal cortices, all of which appear to be involved in visual long-term memory (Egorov, Hamam, Fransen, Hasselmo, and Alonso, 2002; Frank and Brown, 2003; Clavagnier, Falchier, and Kennedy, 2004; Naya, Yoshida, and Miyashita, 2003; Yoshida, Naya, and Miyashita, 2003) and association of pair codings (Miyashita, Morita, Naya, Yoshida, and Tomita, 1998; Sakai and Miyashita, 1991), were predicted to remain constant between pre- and post-consolidation time points. \ Furthermore, a shift in activity from the visuospatial-oriented ventral (temporal) to the recall-oriented dorsal (septal) hippocampus (Thinus-Blanc, Save, Poucet, and Buhot, 1991; Moser and Moser, 1998; Kesner, Lee, and Gilbert, 2004; Matus-Amat, Higgins, Barrientos, and Rudy, 2004) was predicted. , corresponding to a reallocation of attentional ``resources'' after consolidation due to decreased retrieval effort. Both aspects of Specific Aim 1 -- behavioral performance improvement and changes in activation locality -- follow from the assumption that the phenomenon of consolidation is, in essence, a process of optimization of the mnemonic net. \ One of these effects of network optimization, the reallocation of computational resources, would be evinced in a mnemonic net by the elaboration of synaptic arbors or Hebbian reinforcement through long-range horizontal fibers (Gabbot, Martin, and Whitteridge, 1987; Hubbard and Marshall, 1994; see also the discussion in the chapter) in order to perform more parallel computations, a phenomenon called . \ Such multiplexing could occur within a single cytoarchitectonically-defined (``local'') area, such as a high-level visual area in which visual features are abstracted and become task-specific identification ``rules'' with pattern recognition learning; in such cases, neural reallocation might be reflected in a split from a single BOLD cluster into two (a ) or more () BOLD clusters. Alternatively, multiplexing may occur over sparse, distributed networks composed of neural circuits that occupy several cytoarchitectonically-defined ``local'' areas which become generalized with consolidation in order to encode, for example, visual or auditory features that are conserved across a learned stimulus set. \ In these cases, neural reallocation might be represented by an incorporation of several local BOLD clusters into fewer local clusters (or even a single local cluster). \ In the present study, these presumptive changes in resource allocation are assessed qualitatively by examining changes in cluster centers of mass (Specific Aim 1) and quantitatively with the number of Rosenni seed voxels (Specific Aim 2; see below). The extent of within-area BOLD signal activation might also be expected to change due to internalization, as discussed above, or due to the adaptation and specialization of neural circuits to task-specific contexts (a phenomenon referred to as in the present dissertation) as a consequence of consolidation-related optimization effects and concomitant resource reallocation. \ That is, the present study assumes that the number of neurons required to accomplish an encoding or computational function decreases as the encoding or computational efficiency of a neuron or population of neurons increases, an assumption that is supported by molecular, electrophysiological, and neuroanatomical evidence that repeated neuronal activity drives changes in proximal and distal dendritic morphology (Bailey, Chen, Keller, and Kandel, 1992; Zhang, Endo, Cleary, Eskin, and Byrne, 1997; Verzi, 2004), which in turn effect changes in synaptic integration (Yuste and Denk, 1995; Segev, 2006) and therefore the computational functions of the neuron (Shepherd, 1976, 1996; Shepherd and Brayton, 1987; Zador, Koch, and Brown, 1990; Holmes and Levy, 1990; Mallot, 1995; Hausser, Spruston, and Stuart, 2000; London and Hausser, 2005). \ Thus, if memory consolidation is a process of mnemonic network optimization, the number of recruited neurons (or voxels, in the present case) would be expected to decrease. \ In the present study, therefore, the spatial extent of BOLD signal activation is quantitatively assessed through statistical analysis of the (in mm) per region of interest. Activation volume is an excellent indicator of neural or voxel recruitment, but it is not informative with respect to the nature or degree of information processing in signals from the neural populations () that remain correlated with the task after consolidation. \ Moreover, interpretation of observed decreases in activation volume is further complicated by the indirect relationship between neuronal activity and the BOLD signal which underlies the definition of ``activated'' voxels: a change in activation volume may be due to such phenomena as internalization, multiplexing, or increased circuit computational efficiency, as described previously, but in the absence of data indicating that these changes in recruitment are accompanied by increases in (or stability of) the task-specificity and magnitude of neuronal activity, it is equally justifiable to conclude that such decreased recruitment is a consequence of -- for example -- subject inattentiveness or task habituation. If, however, a region of interest was shown to exhibit a higher information load, suggesting an increase in computational efficiency, and/or a higher covariance between the regional BOLD signal and stimulus presentation, suggesting task attunement, a stronger argument could be made that network optimization had occurred in the region. \ In the present study, computational efficiency and task attunement are quantitatively assessed through statistical analysis of the information load (voxel Shannon time series entropy, or voxel TSE, a novel application to FMRI) and voxel/task cross-correlation (VTCC), respectively. \ The particulars of both metrics will be discussed in the chapter. \ Additionally, activation volume, voxel TSE, and VTCC findings will be compared to assessments of regional signal change from baseline, a conventional measure of FMRI activity -- namely, the intensity of oxidative glycolysis -- in order to determine whether the neurodynamic metrics discussed here occur with the same directionality as changes in metabolism and are therefore closely correlated (., a strong correlation: r>>.85, as a stipulative definition), mutually informative (a moderate correlation: .65 r>\ .85), or mutually exclusive (weak correlation: r \>.65). \ An analysis of the correlation between task-related signal change and voxel TSE will also serve to support or refute an inference that remnant voxels exhibiting a post-consolidation increase in information load are as metabolically active as (or more metabolically active than) significantly task-correlated voxels at pre-consolidation. \ This issue is of interest because, again, the assessment of BOLD signal entropy is a novel application for an FMRI-based study of human neurodynamics, and it is as yet unclear how faithfully BOLD signal entropy corresponds to neuronal signal entropy; a statistical concurrence of task-correlated increases in metabolic intensity and voxel TSE would suggest that remnant voxels both carry more information and are more metabolically active after consolidation, thereby supporting the assumption that the metrics discussed above are in actuality assessing the degree of synaptic integration (and therefore the computational intensity) within the region of interest. It should be noted that activation volume, regional signal change from baseline, voxel TSE, and VTCC will only be assessed and discussed for brain activity that was significantly correlated with the presentation of >() in order to minimize the effect of BOLD signals corresponding to stable task-related sensory and lexical networks (, responses to control stimuli) and task-related but generalized attentional, executive, and search-and-retrieval mechanisms (, responses to mismatched stimulus pairs). \ Thus, the findings discussed in the present dissertation should be related exclusively to the recognition and processing of matched associates and the related memory traces putatively stored in the subject's cortex. Finally, the relationship between consolidation, changes in BOLD signal locality (activation volume in Specific Aim 1 and the number of Rosenni seed voxels in Specific Aim 2), and subject performance on the paired-associate task are assessed in the present study by (a) statistical analysis of pre- and post-consolidation subjects' reaction times and accuracies for matched, mismatched, and control stimuli, and (b) an examination of the correlation between matched-associate-correlated performance and locality effects at pre- and post-consolidation time points. \ The objective in case (b) was to determine which brain regions were significantly correlated with decreases in task reaction time and increases in task accuracy. \ (Morris, Bransford, and Franks, 1977; Tulving and Thomson, 1973; Roediger, Weldon, and Challis, 1989, also known as : Wheeler and Buckner, 2003) and an improved capacity for differential processing of features of the presented stimuli so as to >increase the efficacy of comparisons with its stored representation (: Robb and Rugg, 2002; Rugg and Wilding, 2000). \ Based on the hypothesis, discussed above, that consolidation results in increased retrieval efficiency, it was postulated that associates are better-learned after consolidation (Buckner, Koutstaal, Schacter, Wagner, and Rosen, 1998), and therefore retrieval effort was expected to be lower for post-consolidation subjects than for pre-consolidation subjects. \ Thus, the correlation between performance improvements and locality changes, particularly with respect to activation volume, VTCC, and voxel TSE, was expected to be most significant in executive and working memory areas of the left hemisphere dorsolateral prefrontal and frontopolar cortices (corresponding to retrieval -- Rugg, Fletcher, Chua, and Dolan, 1999; Nolde, Johnson, and D'Esposito, 1998; Henson, Shallice, and Dolan, 1999; Nolde, Johnson, and Raye, 1998; Montaldi, Mayes, Barnes, Hadley, Patterson, and Wyper, 2002 -- as well as retrieval : Rugg and Henson, 2002), in the fusiform gyrus (corresponding to retrieval orientation: Homberger, Rugg, and Henson, 2006), and the medial temporal lobe, superior parietal lobule, and posterior cingulate (corresponding to retrieval success: Homberger, Rugg, and Henson, 2006; Montaldi ., 2002). \ A summary table of Specific Aim 1 hypotheses and the relevant predictions that followed from them is presented below. \ \; <\with|par-line-sep|0fn> <\with|par-sep|0fn> ||||||||||||||||||||||||||||||| \ Hypotheses and relevant predictions for Specific Aim 1. \ Abbreviations: , voxel-task cross-correlation; , Shannon entropy of a time series.>>>|>|>>>|>>>||>| <\with|par-line-sep|0fns> <\with|par-par-sep|0.6666fn> Optimization of the mnemonic networks by consolidation enhances retrieval efficiency. > |<\cell> <\with|par-sep|0fn> <\with|font-base-size|10> <\with|par-sep|0fn> <\with|par-line-sep|1fns> <\with|par-par-sep|1pc.> Matched-associate reaction time decreases.*Matched-associate accuracy increases.*The absolute value of the correlation between performance (reaction time, accuracy) and locality (activation volume, voxel-task cross-correlation, and percent signal change) metrics is highest in: <\itemize-dot> Left-hemisphere dorsolateral prefrontal cortex and frontopolar cortex, corresponding to retrieval effort and retrieval success; The right-hemisphere fusiform gyrus, corresponding to retrieval orientation; and The medial temporal lobe, superior parietal lobule, and posterior cingulate, corresponding to retrieval success. >|>>>|r\.85>). \ Thus, these statistics are mutually informative.Reallocation of activity due to changes in the overt orienting of attention is reflected in a shift from bilateral activation to right hemisphere dominance (increased right hemisphere activation volume) in the ventral visual stream and parahippocampal gyrus.Reallocation of activity due to internalization and retrieval orientation is reflected in a decrease in activation volume in the ventral hippocampus and an increase in activation volume in the dorsal hippocampus.Decreases in activation volume in long-term memory areas, as well as attentional and executive areas, due to optimization are accompanied by an increase in the BOLD signal information load in these areas due to increased synaptic integration by constituent neural circuits. \ Moreover, remnant voxels in long-term memory and attentional-executive regions are as metabolically active, or more so, than pre-consolidation voxels, reflecting the increase in synaptic integration; thus, information load (voxel TSE) voxel TSE and glycolytic intensity (percent signal change from baseline) are strongly correlated.> >>>>||*Matched-associate, mismatched-associate, and control-stimuli reaction times and accuracies were computed and will be discussed in the present dissertation, but all further analyses were performed with respect to matched-associates only.>>>>>> \; . As discussed previously, the primary contributions of this dissertation to neuroscience are the application of an information-theoretic statistic, Shannon entropy, to voxel BOLD signals, and the analysis of consolidation-driven mnemonic neurodynamics in the context of functional MRI. \ These two experimental approaches can be seen as addressing the relatively static, time-independent differences in functional architecture and the time-dependent correlations (functional connectivity), or binding dynamics, between the BOLD signals of voxel pairs, respectively. \ Changes in functional architecture were addressed in Specific Aim 1. \ Thus, Specific Aim 2 primarily elaborates on the ``modular'' data obtained in Specific Aim 1, although there exists some overlap between the two Aims, such as the qualitative and quantitative assessment of differential neurocomputational resource allocation, which requires both static (center of mass) and time-dependent (Rosenni) approaches, as discussed previously. To this end, the Rosenni bootstrapping methodology (see the chapter) was applied to FMRI datasets in order to identify brain areas that were significantly covariant and related to the recall task. Voxel pairwise cohoerence, voxel pairwise cross-correlation (PWCC), and Shannon entropy of a voxel time series (TSE), which serve as tests of the nature of the interaction between signal source pairs, were then calculated and compared for pre- and post-consolidation time points. Two hypotheses were proposed under Specific Aim 2: first (SA. II, H. 1.), that (binding efficiency) between memory-relevant voxel pairs, specifically, voxels in attentional, executive, and long-term memory areas (including visual area V3), but diminishes binding efficiency among memory-irrelevant voxel pairs, specifically, working memory and lower visual areas; and second (SA. II, H. 2.), that More explicitly, it was hypothesized that these changes would reflect diminished retrieval effort, an improved capacity for retrieval orientation, and the stability of memory traces in the medial temporal lobe for transfer-appropriate processing. An overall decrease in the number of Rosenni seed voxels was therefore predicted in the left-hemisphere ventral visual stream and perihippocampal gyrus, as well as in attentional and executive regions and visual areas V1 and V2. \ In contrast, the number of Rosenni seed voxels were predicted to remain constant in area V3, the perirhinal, perihippocampal, and ectorhinal cortices, and the hippocampus. The statistics presented in the and chapters were utilized to assess these changes. \ Binding strength was measured using pairwise voxel BOLD signal cross-correlation and the coherence statistic (as binding strength is directly proportional to coherence, stronger coupling results in a larger coherence value) in order to support or refute the prediction, based on SA. II. H. 1. above, that coherence would increase among voxels in long-term memory areas, reflecting enhanced connectivity strength among memory traces, as well as attentional and executive areas, corresponding to a consolidation-driven specialization of neural circuits involved in retrieval orientation and trace search. \ In contrast, coherence among visual areas, particularly areas V1, V2, and components of the ventral visual stream, with the notable exception of the anterior fusiform gyrus and lateral occipital cortex, which function as high-level global visual feature processing areas (Gross, Rocha-Mirenda, and Bender, 1972; Logothetis and Sheinburg, 1996; Tanaka, 1996; Grill-Spector and Malach, 2001; Haxby, Gobbini, Furey, Ishai, Schouten, and Pietrini, 2001) were predicted to decrease, in accordance with the expectations highlighted in the discussion of internalization above and per Tulving (1983). A summary table of the Specific Aim 2 hypotheses and their associated predictions is presented below. \; <\with|par-line-sep|0fn> <\with|par-sep|0fn> ||||||||||||||||||||||||||||||| \ Hypotheses and relevant predictions for Specific Aim 2. \ Abbreviations: , esponse-nset/tatistical xtrema-based etwork ode solation.>>>|>|>>>|>>>||>| > >>>(binding efficiency) between voxels in attentional, executive, and long-term memory areas but diminishes binding efficiency among voxels in working memory and lower visual areas.> |<\cell> <\with|par-sep|0fn> <\with|font-base-size|10> <\with|par-sep|0fn> <\with|par-line-sep|1fns> <\with|par-par-sep|1pc.> Coherence decreases among visual areas, particularly areas V1, V2, and components of the ventral visual stream, with the exception of the anterior fusiform gyrus and lateral occipital cortex. > : >|>>>reflecting diminished retrieval effort, an improved capacity for retrieval orientation, and the stability of memory traces in the medial temporal lobe for transfer-appropriate processing.>>|The number of Rosenni seed voxels in area V3, the perirhinal, perihippocampal, and ectorhinal cortices, and the hippocampus remains constant.>>>>||*All Specific Aim 2 analyses were performed with respect to matched-associates only.>>>>>> \; <\with|par-mode|center> METHODS [Insert METHODS here] <\initial> <\collection> <\references> <\collection> >