Abnormalities in activation of the medial temporal lobe memory system in AD and MCI
With respect to memory, a number of fMRI studies in patients with clinically diagnosed AD, using a variety of visually presented stimuli, have identified decreased activation in hippocampal and parahippocampal regions compared to control subjects during episodic encoding tasks (Small et al., 1999
; Rombouts et al., 2000
; Kato et al., 2001
; Machulda et al., 2003
; Sperling et al., 2003b
). Neocortical abnormalities in AD have also been demonstrated using fMRI, including decreased activation in temporal and prefrontal regions. In addition to AD-related differences in task-related blood-oxygen level dependent (BOLD) signal amplitude or spatial extent, the temporal dynamics of activation appear to be altered in patients with AD (Rombouts et al., 2005
). And as has been observed in other types of tasks, increased
activation in prefrontal and other regions has also been found in AD patients performing memory tasks (Sperling et al., 2003b
Our fMRI studies of MCI and AD have employed primarily two memory tasks, one item-based and one associative task. These paradigms focus on the encoding of stimuli using three conditions grouped in blocks: “Novel” stimuli, seen only once during the scanning procedure; “Repeated” stimuli, shown to subjects prior to the scanning procedure and then shown repeatedly during specific blocks; and “Fixation” on a crosshair as a relatively passive baseline condition. The primary comparison of interest is the Novel vs. Repeated contrast, which holds the visual complexity of the stimuli constant, and provides information about new learning (i.e., processing relevant to the encoding of stimuli for later memory testing as well as processing relevant to novelty). Subjects are instructed explicitly to try to remember the stimuli for later testing. All of the scanning is performed during the encoding phase, and subjects are tested with a post-scan recognition test.
Depending on the task, different components of the MTL memory system have been activated. In our studies of the encoding of complex indoor and outdoor scenes, the caudal parahippocampal cortex and body of the hippocampal formation were activated, while in our studies of the encoding of face-name paired associates, the rostral hippocampal formation and entorhinal cortex were activated. In addition, both paradigms activate ventral temporal and inferior prefrontal cortices. Given that AD pathology is thought to progress along a rostro-caudal gradient in the MTL, there has been surprisingly little study of the relative sensitivity (to disease effects) of fMRI paradigms that activate rostral vs. caudal MTL regions. Little comparison has been made of tasks that activate different regions of the MTL memory system (hippocampal vs. entorhinal vs. perirhinal) in MCI/AD. The paucity of this sort of data is a result, in large part, of the notorious technical difficulties involved in obtaining fMRI data in the ventromedial regions of the brain due to susceptibility artifacts and distortions. Advances in fMRI technology will be critical for the field to achieve the goal of robustly testing hypotheses about the activity of MTL subregions (Small et al., 2001
; Zeineh et al., 2003
; Dickerson, 2007
In a study of mild AD using the face-name paradigm, the mild AD subjects showed lesser activation in the hippocampal formation bilaterally compared to cognitively intact control subjects (Sperling et al., 2003b
). Several neocortical regions, including frontal cortices, showed increased activation in the mild AD patients compared to controls. These findings are consistent with other recent reports demonstrating a relative lack of MTL activation in patients with clinical AD dementia, and suggest that additional regions, not typically activated in the task in young and older controls, may be recruited during performance of this task in individuals with AD dementia.
With respect to MCI, a handful of fMRI studies have been published to date and the results, thus far, have been inconsistent (see for summary). In comparison to older controls, Machulda et al. reported that, during the encoding of novel pictures, MTL activation was decreased to a similar degree in patients with MCI and AD patients (Machulda et al., 2003
). In a face-encoding paradigm, Small et al. reported heterogeneity in MTL activation in memory-impaired subjects, with some showing hypoactivation similar to that of AD patients. Other subjects showed entorhinal and hippocampal activation that was similar to controls, but had decreased activation in the subiculum (Small et al., 1999
). Using a face-name associative paradigm, Petrella et al. (Petrella et al., 2006
) found no differences between MCI and controls in MTL activation during encoding, but observed left hippocampal hypoactivation in MCI vs. controls during the retrieval (forced-choice recognition) condition. Hippocampal hypoactivation in MCI was no longer seen when memory performance accuracy was included as a covariate in the analysis. Using an item-based old/new recognition retrieval paradigm, Johnson et al. found right hippocampal hypoactivation in MCI patients compared to controls (Johnson et al., 2006
). In a separate study, Johnson et al. used a paradigm involving the repetitive presentation of faces to demonstrate that MCI patients do not show the same slope of decreasing hippocampal activation with face repetition that is seen in older controls, suggesting disruption of this “adaptive” response in the medial temporal lobe (Johnson et al., 2004
Three studies have now demonstrated greater MTL activation in MCI patients compared to controls. We used the associative face-name encoding paradigm described above to compare MTL activation in very mild MCI, AD, and controls (Dickerson et al., 2005b
). Compared with controls, MCI subjects showed a greater extent of hippocampal activation and a trend toward greater entorhinal activation (). This group of MCI subjects was very mildly impaired based on Clinical Dementia Rating (CDR) (Morris et al., 1997
) ratings, MMSE, and neuropsychological data, as well as fMRI memory task performance (which was similar to controls). Furthermore, there was minimal atrophy of the hippocampal formation or entorhinal cortex in this MCI group. The AD patients had smaller MTL volumes and a lesser degree of activation in these regions, and performed below controls on the post-scan memory test. Across all the subjects in the three groups, post-scan memory task performance correlated with extent of activation in both the entorhinal cortex and hippocampus.
Figure 2 A phase of compensatory hyperactivation appears to occur in the medial temporal lobe (MTL) in very mild mild cognitive impairment, prior to AD dementia. Representative single subjects from each group, showing normal memory-related MTL activation measured (more ...)
Using a visual object encoding paradigm, Hamalainen et al. found that MCI subjects had greater activation (than controls) of caudal hippocampal formation, parahippocampal gyrus, and fusiform cortex (Hamalainen et al., 2006
). Based on MMSE and neuropsychological data, the MCI subjects in this study were on the relatively more impaired end of the MCI spectrum (although CDR-SB was still mildly impaired), yet the group performed the fMRI memory paradigm relatively well—better than the AD group—although not as well as controls. In the first event-related subsequent memory study of MCI, Kircher et al. used an item-based task with words and found that MCI subjects activated rostral left hippocampal and surrounding cortical regions to a greater degree than controls (Kircher et al., 2007
). MMSE scores from these MCI participants suggested that the group was at the more impaired end of the MCI spectrum, but neuropsychological data indicated milder impairment—in fact, delayed verbal recall scores were minimally impaired relative to controls, with scores for the MCI participants ranging as high as 14 items freely recalled after a 20 minute delay in this 15-item test. In addition, the MCI participants performed similarly to controls on the fMRI memory paradigm.
The variability in fMRI data from MCI subjects probably relates, at least in part, to the complex relationships between the severity of the subjects’ clinical impairment and to their ability to perform the memory task employed as the fMRI paradigm. This hypothesis is based primarily on data from our initial fMRI study of MCI, which was a within-group investigation of individuals who spanned a broad range of MCI, the common feature of which was an overall CDR rating of 0.5 (Dickerson et al., 2004
), as well as a more recent investigation separating the MCI spectrum into two subgroups, one on the milder end and one on the more impaired end (this investigation (Celone et al., 2006
) is discussed below). Because we explicitly sought to study a broad range of impairment within the spectrum of MCI, we did not require subjects to perform below a particular cutoff on neuropsychologic memory tests, and thus some subjects were included who performed relatively well on neuropsychological testing, despite symptoms of memory impairment in daily life.
Thirty-two subjects with a total CDR rating of 0.5 and CDR Sum of Boxes (CDR-SB) scores ranging from 0.5 – 3.0 were studied. We focused on analysis of functional activation in the novel vs. familiar scene-encoding paradigm in the hippocampal formation and the parahippocampal gyrus, and the volumes of these regions were also quantified. As expected, there was an inverse linear relationship between the degree of clinical impairment (CDR-SB) and volume of MTL ROIs, most significant in the left hippocampus, such that subjects with greater clinical impairment had smaller hippocampal volumes. Interestingly, however, there was a direct linear relationship between the degree of clinical impairment and the extent (number of voxels activated) of fMRI activation bilaterally in both the hippocampus and parahippocampus, such that subjects with greater clinical impairment had a relatively larger extent of MTL activation. This “paradoxical” relationship was still apparent after correction for volume, and a multivariate analysis showed that greater clinical impairment (as measured by the CDR-SB) was associated with older age, increased extent of activation in the right parahippocampal gyrus, and decreased volume of the left hippocampus. Furthermore, similarly to the data from the face-name paradigm described above, better performance on the post-scan recognition memory task correlated with greater MTL activation and larger volume.
Based on these data, we believe that discrepant findings on memory-related MTL activation in the MCI literature may potentially be explained, at least in part, by differences in the level of clinical impairment and in performance on the fMRI memory task between the subject groups. From the descriptions of the clinical data from the studies above (and detailed in the ), it is clear that these clinical-behavioral measures are not necessarily completely correlated with each other—that is, some samples of MCI subjects may appear relatively more impaired from certain clinical measures (e.g., symptom-based measures such as CDR), less impaired on other clinical measures (e.g., neuropsychological performance), and may or may not be able to perform the fMRI behavioral task used for activation at a level comparable to controls. Further research is needed to clarify these relationships, a deeper understanding of which is critical to our interpretation of imaging data.
There are also a number of other factors that likely contribute to variability in MTL activation, which may vary between studies in a manner that could also explain some of the discrepancies, including differences in the memory tasks themselves (e.g., encoding vs. retrieval; visual vs. verbal material; paired-associate vs. item-based memory, etc.), differences in analysis methods (i.e., voxel-based whole-brain vs. focused region-of-interest approaches) and dependent variables used to determine level of activation (i.e., magnitude vs. extent of activation vs. voxel-based measures that take both magnitude and extent into account), and differences in age, education, and apolipoprotein E genotype (Dickerson et al., 2005b
). Aside from the details of the studies, the table highlights the broad scope of methodologic variability that makes it difficult to compare fMRI studies of MCI. Additional research will be critical to the further elucidation of the contributions of these and other factors to variability in fMRI measures, if such measures are to be translated into biomarkers for clinical trials. It will be important to study structure-function relationships to better understand the relationships between anatomic abnormalities (e.g., hippocampal and entorhinal atrophy, as well as neocortical atrophy) and functional hypo- and hyperactivation. Multimodal investigations including positron emission tomography (PET) measures of metabolism and pathology will likely be helpful to ensure that patients have abnormalities consistent with AD, as well as to determine relationships between the localization and severity of these abnormalities and alterations in functional activation. Longitudinal studies—involving both clinical and imaging follow-up data—will probably shed a great deal of light on the variety of contributors to fMRI abnormalities in MCI. Finally, a multi-center study in which investigators agree on the use of a particular set of behavioral task paradigms and data collection methods would potentially enable these issues to be further clarified, and would enable data analysis methods to be compared (Friedman et al., 2007
). In all of these future investigations, we believe it is important to provide detailed demographic, clinical, neuropsychological, and behavioral performance data to help clarify the similarities or differences between samples of MCI subjects in fMRI studies. In addition, it may be helpful to report hippocampal volumes as well.
Despite all the caveats, there is replicated evidence to support the hypothesis that there may be a phase of increased MTL activation in MCI. This increase, which also may be present in cognitively intact carriers of the APOE-ε4 allele (for review, see (Wierenga and Bondi, 2007
)), may represent an attempted compensatory response to AD neuropathology, given that some MCI individuals with smaller hippocampal volume perform similarly on memory tasks to MCI individuals with larger hippocampal volume but have relatively greater MTL activation (Dickerson et al., 2004
; Hamalainen et al., 2006
) (). Additional studies employing event-related fMRI paradigms (Sperling et al., 2003a
; Dickerson et al., 2007a
; Kircher et al., 2007
) will be very helpful in determining whether increased MTL activation in MCI patients is specifically associated with successful memory, as opposed to a general effect that is present regardless of success (possibly indicating increased effort). It is possible that MTL hyperactivation reflects cholinergic or other neurotransmitter upregulation in MCI patients (DeKosky et al., 2002
). Alternatively, increased regional brain activation may be a marker of the pathophysiologic process of AD itself, such as aberrant sprouting of cholinergic fibers (Hashimoto and Masliah, 2003
) or inefficiency in synaptic transmission (Stern et al., 2004
). It is important, however, to acknowledge that multiple non-neural factors may confound the interpretation of changes in the hemodynamic response measured by BOLD fMRI, such as age- and disease-related changes in neurovascular coupling (Buckner et al., 2000
; D’Esposito et al., 2003
), AD-specific alterations in vascular physiology (Mueggler et al., 2002
), and resting hypoperfusion and metabolism in MCI and AD (El Fakhri et al., 2003
), which may result in an amplified BOLD fMRI signal during activation (Davis et al., 1998
; Cohen et al., 2002
). Further research to determine the specificity of hyperactivation with respect to particular brain regions and behavioral conditions will be valuable to better characterize this phenomenon.
Figure 3 A greater degree of hyperactivation of MTL regions is present in MCI individuals with a greater degree of MTL (hippocampal) atrophy, supporting the possible compensatory role of hyperactivation for AD pathology. Data shown here are taken from two separate (more ...)
A number of authors have hypothesized that MTL and other cortical hyperactivation during the performance of memory and other cognitive tasks may play, at least in part, a compensatory role for neuropathologic abnormalities in MCI/mild AD (Becker et al., 1996
; Backman et al., 1999
; Stern et al., 2000
). “Compensation” is typically defined as greater regional brain activity (hyperactivation) in an MCI/AD group in the setting of task performance accuracy that is similar to that of a matched control group. Regional hyperactivation may involve greater magnitude of activity in brain regions typically active during performance of the task (when performed by controls), or the recruitment of additional brain regions not normally engaged by controls. However, it is also clear that greater task difficulty may provoke similar alterations in regional brain activity in healthy individuals (Gur et al., 1988
; Grasby et al., 1994
; Grady, 1996
; Rypma and D’Esposito, 1999
). It is challenging to know to what degree MCI/AD groups find memory tasks to be “more difficult” than they would in the absence of disease. This has led some investigators to attempt to match task difficulty between MCI/AD patients and controls (Stern et al., 2000
). It is also possible that different cognitive strategies during memory task performance (e.g., semantic elaborative encoding strategies vs. visualization strategies) may contribute to differences in the recruitment of particular brain regions (Kirchhoff and Buckner, 2006
), and that this may vary between patient and control groups. Further work in this area, including longitudinal studies in MCI/AD patients, ideally including detailed behavioral measures of reaction time as well as accuracy and possibly self report of task difficulty, will be important to better clarify the situations in which activity increases can be reasonably interpreted as compensatory for brain disease.
Despite these caveats, we believe that the accumulating evidence indicates that task-related regional brain hyperactivation may be a universal neural response to insult, as it occurs in sleep deprivation (Drummond et al., 2000
), aging (Cabeza et al., 2002
), and a variety of neuropsychiatric disorders and conditions, including AD/MCI, Huntington’s disease (Rosas et al., 2004
), Parkinson’s disease (Monchi et al., 2004
), cerebrovascular disease (Carey et al., 2002
; Johansen-Berg et al., 2002
), multiple sclerosis (Reddy et al., 2000
; Morgen et al., 2004
), traumatic brain injury (McAllister et al., 1999
), HIV (Ernst et al., 2002
), alcoholism (Desmond et al., 2003
), schizophrenia (Callicott et al., 2003
). In many of these studies, task-related regional brain hyperactivation was associated with the relative preservation of performance on the task, suggesting that hyperactivation may be serving, at least in part, a compensatory role for neurologic insult. The evidence discussed above also indicates that increased MTL activation can be seen in MCI in the setting of minimal MTL atrophy (Dickerson et al., 2005b
), which provides in vivo
support for laboratory and animal data suggesting that physiologic alterations may precede significant structural abnormalities very early in the course of a neurodegenerative disease such as AD (Selkoe, 2002
; Walsh and Selkoe, 2004
) and may represent inefficient neural circuit function (Stern et al., 2004
). Thus, fMRI may provide a means to detect changes in human memory circuit function that underlie the earliest symptoms of AD, and may be useful in identifying groups of subjects at high risk for future cognitive decline prior to a diagnosis of AD.