Specific and disease stage-dependent episodic memory-related brain activation patterns in Alzheimer’s disease: a coordinate-based meta-analysis

Abstract

Episodic memory is typically affected during the course of Alzheimer’s disease (AD). Due to the pronounced heterogeneity of functional neuroimaging studies on episodic memory impairments in mild cognitive impairment (MCI) and AD regarding their methodology and findings, we aimed to delineate consistent episodic memory-related brain activation patterns. We performed a systematic, quantitative, coordinate-based whole-brain activation likelihood estimation meta-analysis of 28 functional magnetic resonance imaging (fMRI) studies comprising 292 MCI and 102 AD patients contrasted to 409 age-matched control subjects. We included episodic encoding and/or retrieval phases, investigated the effects of group, verbal or image stimuli and correlated mean Mini-Mental-Status-Examination (MMSE) scores with the modelled activation estimates. MCI patients presented increased right hippocampal activation during memory encoding, decreased activation in the left hippocampus and fusiform gyrus during retrieval tasks, as well as attenuated activation in the right anterior insula/inferior frontal gyrus during verbal retrieval. In AD patients, however, stronger activation within the precuneus during encoding tasks was accompanied by attenuated right hippocampal activation during retrieval tasks. Low cognitive performance (MMSE scores) was associated with stronger activation of the precuneus and reduced activation of the right (para)hippocampus and anterior insula/inferior frontal gyrus. This meta-analysis provides evidence for a specific and probably disease stage-dependent brain activation pattern related to the pathognomonic AD characteristic of episodic memory loss.

Introduction

Episodic memory has been defined as memory for information (‘what’) paired with a temporal (‘when’) and a spatial (‘where’) aspect (Tulving 1972; Jonides et al. 2002). It can roughly be divided into encoding and retrieval phases. Encoding presents an automatic and effortless process (Dere et al. 2010; Pause et al. 2013), whereas retrieval is a more active cognitive function. Functional neuroimaging studies have identified several encoding-related brain activation patterns including the dorsolateral and ventrolateral prefrontal cortices, intraparietal sulcus/superior parietal lobule, medial temporal lobe (MTL) and adjacent occipito-temporal areas (Spaniol et al. 2009). The MTL encompasses the hippocampal formation which is essential for long-term storage of memory (Scoville and Milner 2000; Squire et al. 2004), the parahippocampal gyrus (PHG) which plays an important role in episodic encoding and spatial navigation (Kühn and Gallinat 2013), the perirhinal and entorhinal cortices and the amygdala. In contrast, the retrieval phase, has been shown to engage the posterior parietal cortex including the inferior and posterior lateral parietal cortices and precuneus, the left dorsolateral prefrontal cortex, the anterior cingulate cortex, the left ventrolateral prefrontal cortex, the left PHG, bilateral insula and bilateral caudate (Spaniol et al. 2009). Importantly, brain activation patterns have been reported to be rather left-hemispheric for verbal stimuli while bilateral or preferably right-hemispheric for images (Buckner et al. 1996; Kelley et al. 1998; Golby et al. 2001).

In the psychopathological continuum of Alzheimer’s Disease (AD)—the world’s most common neurodegenerative disorder—episodic memory is one of the first cognitive functions impaired (Welsh et al. 1991). For the early stage, the term “mild cognitive impairment” (MCI), and specifically the amnestic subtype (aMCI), has been widely used in previous studies (Petersen 2011), especially before the revision of the definition of early or prodromal AD by Dubois et al. (2010). MCI describes a clinical condition of objective cognitive (MCI) and/or memory (aMCI) impairment without interference in daily activities and high likelihood of transition to AD. The hippocampal formation is typically affected already in early stages of AD pathology (e.g. MCI) (Braak and Braak 1991). Previous functional neuroimaging research on memory encoding and episodic retrieval in MCI reported attenuated hippocampal activation in line with findings in AD (Hanseeuw et al. 2011; Hampstead et al. 2011; Jin et al. 2012). Paradoxical results of increased encoding hippocampal activation in MCI patients (Celone et al. 2006; Kircher et al. 2007; Hämäläinen et al. 2007; Trivedi et al. 2008; Clément et al. 2010) have been discussed along with longitudinal findings of increased hippocampal activation in earlier stages of AD which later progress to attenuation in more severe stages (O’Brien et al. 2010).

Along with the hippocampus, the precuneus has recently attracted a lot of attention. It has been hypothesised to be deactivated rather than activated during memory encoding tasks (Huijbers et al. 2012). Morphological studies have reported AD-like findings (e.g. atrophy) in MCI patients (Nickl-Jockschat et al. 2012) while neuropathological studies found increasing amyloid plaque deposition over the course of the AD continuum (Braak and Braak 1991; Buckner et al. 2005). The precuneus might be prognostic to distinguish between high and low risk of conversion to AD (Schroeter et al. 2009).

The present study aimed at identifying patterns of aberrant activation during episodic memory processes that are common to MCI and/or AD using a systematic, quantitative comparison of neural correlates in patients compared to non-affected controls. Accordingly, a meta-analysis was carried out using an advanced activation likelihood estimation (ALE), which is a powerful tool to quantitatively converge different studies and to identify consistent patterns of group differences (Eickhoff et al. 2009). Two ALE meta-analyses have previously been published in the field of episodic memory in AD (Schwindt and Black 2009; Browndyke et al. 2012). However, the present study used different methodologies and provides several further insights: in contrast to work by Schwindt and Black (2009), we included only studies reporting differences between patient and control groups, rendering the results more specific. In contrast to the second meta-analysis (Browndyke et al. 2012), we chose to restrict our meta-analysis to whole-brain studies rather than including work derived from smaller fields of view, which is biased by a hypothesis-driven analysis. Following a joint analysis of AD and MCI patients, we also probed episodic memory activation in MCI and AD separately. The present work is furthermore aimed at extending the existing literature not only by (1) increasing the number of studies included, but also by (2) putting a particular emphasis on the differentiation of episodic memory for verbal and image stimuli, (3) investigating encoding as well as retrieval phases and (4) probing the relationship between altered brain activation and cognitive performance.

We hypothesised that the convergence of episodic memory-related brain regions determined by functional neuroimaging would detect aberrant brain activation patterns in brain areas affected in patients with MCI or AD. We further investigated the MCI and AD phenotypes in verbal as well as image episodic memory tasks to unravel differential stimuli dependency. Finally, we conjectured that reduced cognitive performance predicts dysfunctional brain activation.

Methods

Literature search and study selection

A systematic search was carried out for functional studies investigating MCI and AD subjects within the PubMed (http://www.ncbi.nlm.nih.gov/pubmed/; search-strings: [{(magnetic resonance imaging) OR PET} Alzheimer OR (Mild Cognitive Impairment) episodic memory] and BrainMap databases (http://brainmap.org/). We further screened the reference lists of published reviews on MCI and AD as well as of all obtained studies to identify additional studies. We selected studies according to the following seven criteria:

1. Functional magnetic resonance imaging (fMRI) datasets

2. Subjects had to be diagnosed with (a)MCI or AD (Table 1). As most studies were performed prior to the revision of diagnostic research criteria for AD and prodromal as well as preclinical stages in 2010 (Dubois et al. 2010), studies on MCI mainly used the proposed criteria by Petersen (Petersen et al. 1999; Petersen 2011). For the same reason, AD patients fulfilled diagnostic criteria proposed by the National Institute of Neurological and Communicative Disorders and Stroke and the Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA, McKhann et al. 1984). When inspecting the diagnostic criteria of the four studies labelling their subjects as “MCI”, we found relevant episodic memory deficit characteristics. Three studies (Celone et al. 2006; Heun et al. 2007; Kircher et al. 2007) particularly described objective memory deficits as inclusion criteria in line with an aMCI diagnosis. The fourth study (Kochan et al. 2010) comprised 25 aMCI and only seven naMCI patients. In general, naMCI patients have been shown to also convert to AD, although at lower rates than aMCI patients (Jungwirth et al. 2012; Espinosa et al. 2013). However, compared to aMCI, naMCI patients also have a higher rate of progression to other dementia syndromes, such as vascular dementia or dementia with Lewy bodies.

3. An episodic memory task had to be performed by the subjects while undergoing functional scanning. Specifically, episodic memory was defined as memory for information that is stored regarding the time and location of occurrence (Jonides et al. 2002) (e.g. encoding word lists). For encoding, we included intentional and incidental memory from block design studies as well as encoding phases from working memory if they had been reported separately from retrieval based on an event-related design (Peters et al. 2009). Given that differential neural correlates have been proposed for episodic retrieval and working memory (Cabeza et al. 2002), we aimed at selectively including episodic retrieval. Therefore, we included only event-related and block design retrieval experiments comprising several distinct stimuli (minimum included: seven) in the encoding block, which were later retrieved. Based on this criterion, we rejected three retrieval experiments which had only one or two stimuli per condition (Gould et al. 2005; Peters et al. 2009; Bokde et al. 2010).

4. Direct comparison of study population and a group of age-matched controls.

5. Stereotactic whole-brain coordinates were reported in a standardised reference space [Talairach or Montreal Neurological Institute (MNI) space] for the location of activation differences between MCI or AD, respectively, and controls.

6. Independent study samples without overlap

7. Studies solely focusing on drug effects were excluded.

Table 1.

Summary of clinical and demographic data of included articles with memory task characteristics

Study	Patients				Controls			Task and contrast	Stimuli	Memory process
	MCI/AD	n (f)	Age (years, mean ± SD)	MMSE (mean ± SD)	n (f)	Age (years, mean ± SD)	MMSE (mean ± SD)
Belleville et al. (2011)	aMCI	15 (11)	70.13 ± 7.34	27.73 ± 1.87	15 (10)	70 ± 7.26	29.1 ± 0.74	Encoding > resting	Wordsa	Encoding
								Old–new recognition judgement > resting	Wordsa	Retrieval
Bokde et al. (2010)	aMCI	8 (2)	70.8 ± 5.3	26.6 ± 1.3	8 (3)	66.6 ± 3.9	30 ± 0.0	Successful encoding > resting	5 Consonantsa	Encoding
Celone et al. (2006)	MCIb (low SB)	15 (7)	75.1 ± 7.1	29.3 ± 0.9	15 (8)	75.5 ± 6.0	29.5 ± 0.5	Encoding > fixation (Positive Task-related component; ICA)	Face-name pairsa, c	Encoding
Clément and Belleville (2010)	aMCI (higher cognition)	13 (8)	68.62 ± 10.3	28.85 ± 1.57	14 (8)	67.21 ± 6.8	29.29 ± 1.14	Encoding > fixation	Unrelated word pairsa	Encoding
Clément et al. (2010)	aMCI	12 (9)	67.83 ± 7.49	27.83 ± 1.59	10 (8)	71.7 ± 7.62	29.1 ± 0.74	Encoding > resting	Wordsa	Encoding
								Old–new recognition judgement > resting	Wordsa	Retrieval
Clément and Belleville (2012)	aMCI (higher cognition)	13 (8)	68.62 ± 10.30	28.85 ± 1.57	14 (8)	67.21 ± 6.8	29.29 ± 1.14	Correct associated or rearranged judgement > fixation	Wordsa	Retrieval
								old-new recognition judgement > fixation	Wordsa	Retrieval
Clément and Belleville (2012)	aMCI (lower cognition)	13 (7)	67.08 ± 6.29	26.46 ± 1.56	14 (8)	67.21 ± 6.8	29.29 ± 1.14	Old–new recognition judgement > fixation	Words	Retrieval
Dannhauser et al. (2008)	aMCI	10 (5)	72 ± 13.5	24.5 ± 1.5	10 (6)	68 ± 13.5	28.3 ± 1.6	Reading out stimulus and memorising > reading out “wait”	Wordsa	Encoding
Giovanello et al. (2012)	aMCI	12 (7)	75.2 ± 4.3	27.8 ± 1.7	12 (7)	72.6 ± 5.9	29.5 ± 0.9	Relational > item recognition	Wordsa	Retrieval
								Item > relational recognition	Wordsa	Retrieval
Golby et al. (2005)	AD	7 (2)	69	20.8	7 (3)	66	29.4	Novel > repeated	Scenesc	Encoding
Gould et al. (2005)	AD	12 (7)	77.3 ± 4.9	26.33 ± 2.06	12 (7)	77.3 ± 4.8	29.08 ± 0.9	Successful encoding > resting	Object-location pairsc	Encoding
								Correct > incorrect encoding	Object-location pairsc	Encoding
Grön et al. (2002)	AD	12 (7)	61.7 ± 5.0	25.9 ± 3.5	12 (5)	59.8 ± 2.6	30 ± 0.0	High > low familiarity recall	Abstract patternsc	Retrieval
Hämäläinen et al. (2007)	aMCI	14 (10)	72.4 ± 7.3	25.6 ± 3.1	21 (17)	71.2 ± 4.9	27.7 ± 2.0	Encoding > fixation	Word-picture pairsa, c	Encoding
Hämäläinen et al. (2007)	AD	15 (10)	73.1 ± 6.7	21.7 ± 3.7	21 (17)	71.2 ± 4.9	27.7 ± 2.0	Encoding > fixation	Word-picture pairsa, c	Encoding
Hampstead et al. (2011)	aMCI	18 (n.a.)	71.2 ± 8.5	26.7 ± 2.3	16 (n.a.)	72.1 ± 7.3	27.8 ± 1.97	Novel > repeated	Object-location pairsc	Encoding
Hanseeuw et al. (2011)	aMCI	16 (5)	72.6 ± 7.9	27.3 ± 1.6	15 (9)	69.4 ± 4.8	28.7 ± 1.5	Encoding > fixation	Item-cue paira, c	Encoding
Heun et al. (2007)	MCI	20 (9)	69.7 ± 7.1	26.6 ± 1.5	28 (11)	67.5 ± 5.4	28.9 ± 1.1	Correct recognition > fixation	Wordsa	Retrieval
								Correct rejections > fixation	Wordsa	Retrieval
Jin et al. (2012)	aMCI	8 (3)	60.9 ± 3.2	28.1 ± 1.1	8 (4)	60.6 ± 8.3	29.6 ± 0.5	Simple encoding > control task	Picturesc	Encoding
								Associative encoding > control task	Face-occupation pairsa, c	Encoding
								Old–new recognition judgement > control task	Picturesc	Retrieval
								Old–new recognition judgement > control task	Face-occupation pairsa, c	Retrieval
								Old–new recognition judgement > control task	Wordsa	Retrieval
Kircher et al. (2007)	MCI	21 (12)	69.7 ± 7.0	26.6 ± 1.4	29 (17)	67.8 ± 5.4	28.8 ± 1.2	Successful encoding > fixation	Wordsa	Encoding
Kochan et al. (2010)	MCI	35 (21)	77.97 ± 3.88	27.94 ± 1.56	22 (12)	77.16 ± 3.31	29.32 ± 0.95	High > low load interaction	Object-location pairsc	Encoding
Machulda et al. (2009)	aMCI	29 (15)	75 ± 7	n.a.	19 (8)	73 ± 7	n.a.	Encoding > matching picture decision	Picturesc	Encoding
								Old–new recognition judgement > matching picture decision	Picturesc	Retrieval
Mandzia et al. (2009)	aMCI	14 (7)	68.6 ± 7.4	27.7 ± 1.1	14 (7)	72.2 ± 6.4	28.6 ± 1.1	‘Deep’ incidental encoding > scrambled picture	Picturesc	Encoding
								‘Shallow’ incidental encoding > scrambled picture	Picturesc	Encoding
								‘Deep’ old–new recognition judgement > scrambled picture	Picturesc	Retrieval
								‘Shallow’ old-new recognition judgement > scrambled picture	Picturesc	Retrieval
Pariente et al. (2005)	AD	12 (8)	70.9	25.1	17 (4)	76.5	29	Correct > incorrect encoding	Face-name pairsa, c	Encoding
								Correct encoding	Face-name pairsa, c	Encoding
								Correct > incorrect associative recognition	Face-name pairsa, c	Retrieval
								Correct associative recognition	Face-name pairsa, c	Retrieval
Peters et al. (2009)	AD	16 (13)	77.1 ± 6.6	23.4 ± 1.7	16 (14)	76.0 ± 6.1	n.a.	Encoding > control task	Wordsa	Encoding
Petrella et al. (2006)	aMCI	20 (8)	75.0 ± 7.6	26.7 ± 1.5	20 (11)	71.2 ± 4.5	28.4 ± 1.4	Novel > repeated	Face-name pairsa, c	Encoding
								Novel > repeated	Face-name pairsa, c	Retrieval
Petrella et al. (2007)	AD	13 (5)	71.37 ± 6.8	24.62 ± 2.43	28 (14)	71.96 ± 4.94	28.25 ± 1.4	Novel > repeated	Face-name pairsa, c	Encoding
Rémy et al. (2005)	AD	8 (7)	72.2 ± 10.8	21.2 ± 6.4	11 (6)	65.9 ± 5.7	29.4 ± 0.5	Encoding > reading	Wordsa	Encoding
								Old–new recognition judgement > reading	Wordsa	Retrieval
Sperling et al. (2003)	AD	7 (6)	80.6 ± 6.9	22.6 ± 2.2	10 (8)	74.1 ± 7.3	n.a.	Novel > fixation	Face-name pairsa, c	Encoding
								Novel > repeated	face-name pairsa, c	Encoding
Trivedi et al. (2008)	aMCI	16 (11)	77.0 ± 8.4	26.3 ± 2.3	23 (12)	73.1 ± 5.5	28.8 ± 1.2	Encoding > reading ‘push’	Picturesc	Encoding
								Remembered > forgotten	Picturesc	Encoding
								Remembered > forgotten	Picturesc	Retrieval
								Correct rejections > false alarms	Picturesc	Encoding
								Correct rejections > hits	Picturesc	Encoding
Van der Meulen et al. (2012)	aMCI	13 (9)	69.2 ± 8.2	26.7 ± 2.3	15 (9)	68.1 ± 7.2	29.5 ± 0.8	Encoding > instructions	Picture pairsc	Encoding
								Familiarity > recollection	Picture pairsc	Retrieval
								Recollection > familiarity	Picture pairsc	Retrieval

Values represent numbers of subjects or mean ± SD

F female, (a)MCI (amnestic) mild cognitive impairment, AD Alzheimer’s disease, MMSE mini-mental state examination, n.a. not available

^aVerbal stimuli

^bMCI diagnosis according to clinical dementia rating with at least 0.5 in the memory box

^cImage stimuli

We contacted authors for missing data if the publications were otherwise suitable for our meta-analysis. This resulted in the inclusion of one additional study (Hampstead et al. 2011). To avoid omitting valuable information on spatial distribution of functionally altered brain activation patterns, we included several experiments/contrasts per study (i.e. different stimuli types). All reported clusters have at least two independent studies contributing to it. To gather all available information on spatial distribution, we decided to also include second level contrasts, for instance ‘remembered > forgotten’ or ‘high > low demand’. This meta-analysis was performed according to the ethical standards of the Medical Faculty of RWTH Aachen University and the latest version of the Declaration of Helsinki.

Study characteristics

In the first step, we screened the abstracts of 1,425 individual studies published up until January 2013. Twenty-eight fMRI studies matched our inclusion criteria and were subjected to an ALE meta-analysis. They comprised 473 foci from 56 experiments from 102 patients with AD, 292 with MCI and 409 healthy age-matched controls.

All demographic and clinical data along with descriptions of employed memory tasks and contrasts are listed in Table 1. In the following paragraph, we will briefly outline the diverse episodic memory paradigms employed. As stimuli, word lists and face-name pairs were most commonly used. Instructions for encoding phases explicitly required the subjects to intentionally memorise the stimuli with the exception of “incidental” memory (Mandzia et al. 2009), whereby the subjects performed a distraction task (i.e. decision whether picture is man-made or natural) and later received a surprise retrieval task. For retrieval tasks, we found most frequently the so-called “old-new recognition judgement”, which involves the presentation of either a formerly presented or an unknown stimulus along with an according response by the subject. Other studies employed cued recall (“Which stimulus was presented with this cue?”) (Hämäläinen et al. 2007) or associative recognition (e.g. “Have these two stimuli been presented together earlier?”). Most experiments directly compared the encoding or retrieval phases to respective control tasks, thus presented a “first-level contrast”. Often an undemanding referential condition was used as baseline (e.g. fixation cross). Some studies included second-level contrasts such as ‘correct > incorrect encoding’ which, for instance, highlighted successful memory. Finally, Pariente et al. (2005) directly compared the encoding or recognition phases of AD patients to controls without subtracting any referential conditions before.

To guarantee comparability of MCI and AD studies in the respective contrasts (encoding, retrieval), we statistically compared mean age and MMSE scores. We used the statistical package SPSS 19 (Chicago, IL, USA) for Levene’s Tests for Equality of Variances (p < 0.05) and t tests (p < 0.05). MCI compared to AD patients did not differ significantly for age in either task. In contrast, AD patients had significantly lower MMSE scores than MCI patients (encoding: MCI 27.28 ± 1.38 vs. AD 23.22 ± 1.99; retrieval: MCI 27.34 ± 0.83 vs. AD 24.58 ± 2.30). This finding was rather expected since significant cognitive problems are a discriminative criterion between MCI and AD. We then checked for comparability of encoding and retrieval studies within each patient group. MCI patients did not differ significantly for age or MMSE scores. AD patients were matched for MMSE scores for both tasks and were significantly older in the encoding (mean 73.95 ± 3.96) than in the retrieval studies (mean 66.89 ± 5.43). However, we are not convinced that age solely explains the reported differential brain activation patterns. The difference in age might be a result of the small selection of only four AD retrieval studies.

ALE algorithm

A revised version (Eickhoff et al. 2012) of the ALE approach for coordinate-based meta-analysis of neuroimaging data (Turkeltaub et al. 2002; Laird et al. 2009; Eickhoff et al. 2009) was used to test for significant convergence of episodic memory-associated brain activation in MCI and AD. The central data processing step in ALE meta-analysis is to model each focus’ suspected real location as a 3D Gaussian probability distribution. Larger sample sizes are believed to provide data closer to the real value, and the belonging Gaussian distribution is thus smaller (Eickhoff et al. 2009). The probabilities of all foci reported in a given experiment were then combined for each voxel, resulting in a modelled activation (MA) map (Turkeltaub et al. 2012). Taking the union across these MA maps yielded voxel-wise ALE scores describing the convergence of results at each particular location of the brain. To distinguish between ‘true’ convergence studies and random convergence (i.e. noise), ALE scores were compared to an analytically derived null-distribution reflecting a random spatial association between experiments (Eickhoff et al. 2012). The yielded non-parametric p values for each meta-analysis were then thresholded at a cluster-level corrected threshold of p < 0.05 (cluster-forming threshold at voxel-level p < 0.001) and transformed into Z scores for display. The extent-threshold necessary to control the cluster-level family-wise error (cFWE) rate was derived from a Monte-Carlo simulation of the excursion-set above cluster-forming threshold based on the analysis of randomly distributed foci under otherwise identical settings.

All resulting clusters of the relative activation differences were labelled using the SPM Anatomy Toolbox v1.8 (Eickhoff et al. 2005, 2007). The Anatomy Toolbox provides maximum probability maps, which represent a summary of different histological maps and allows attribution of each voxel of the reference space to the most likely cytoarchitectonic area (Eickhoff et al. 2005). Additionally, to define subregions in the hippocampus in more detail, clusters in the cornu ammonis (CA) were visually allocated by experienced raters to the subregions CA1, CA2, CA3 or CA4 by manually drawing labels according to ‘The Human Hippocampus’ (Duvernoy et al. 2005) (Fig. 1).

Fig. 1.

Hippocampal clusters and (manual) cytoarchitectonic allocation. a Axial section of increased anterior hippocampal activation (mainly CA2-4) in MCI patients during encoding (all stimuli). b Coronary section of reduced left-hemispheric and hippocampal activation (mainly CA1) extending to fusiform gyrus in MCI patients during image retrieval. c Axial section showing decreased hippocampal activation (SUB, CA1) in AD patients during retrieval (all stimuli). All clusters are family-wise error cluster corrected at p < 0.05. MCI mild cognitive impairment, AD Alzheimer’s disease, R right hemisphere, L left hemisphere, x, y, z display Montreal Neurological Institute (MNI) coordinates, CA cornu ammonis, DG dentate gyrus, SUB subiculum, PHG parahippocampal gyrus

Contrasts and subgroup-analyses

Encoding and retrieval phases were analysed separately. First, co-localisations between the four contrasts published by our included studies were calculated (‘MCI > Controls’, ‘Controls > MCI’, ‘AD > Controls’ or ‘Controls > AD’). Assuming MCI to be part of the continuum from healthy ageing to AD (Buckner et al. 2005), we searched for convergent clusters within MCI and AD patients compared to controls by generating encoding and retrieval maps for the MCI and AD groups pooled together for the published contrasts [‘MCI > Controls’ AND ‘AD > Controls’] and [‘Controls > MCI’ AND ‘Controls > AD], further referred to as “MCI + AD > Controls” and “Controls > MCI + AD”, respectively. To address the hypothesis of a common pathophysiology of MCI and AD, we computed conjunction analyses (cFWE p < 0.05). However, the main results (Table 2) for each contrast in the MCI vs. controls group revealed entirely different regions than the AD vs. controls group in the respective contrasts. Hence, a quantitative conjunction analysis could and did not yield any results at the same statistical threshold.

Table 2.

Significant clusters of convergence in the ALE meta-analyses

Contrast	Task	Volume (n)	MNI coordinates			R/L	Macroanatomical and cytoarchitectonic region	Computed experiments	Main contributing studies
			x	y	z
MCI > Controls	Encoding all	66	24	−16	−14	R	Hippocampus (maximum: CA) Distribution: 82 % in CA, 6 % in SUB CA manually: CA2, CA3, CA4	10	Celone et al. (2006), Trivedi et al. (2008)
	Encoding verbal							8
	Encoding images	88	25	−15	−14	R	Hippocampus (maximum: CA) Distribution: 75 % in CA, 8 % in amygdala (CM), 6 % in SUB CA manually: CA2, CA3, CA4	5	Celone et al. (2006), Trivedi et al. (2008)
	Retrieval all							13
	Retrieval verbal							10
	Retrieval images							4
Controls > MCI	Encoding all							19
	Encoding verbal							7
	Encoding images							14
	Retrieval all							14
	Retrieval verbal	94	32	24	−4	R	Anterior insula lobe, inferior frontal gyrus	8	Petrella et al. (2006), Giovanello et al. (2012)
	Retrieval images	92	−32	−12	−22	L	Fusiform gyrus, hippocampus (maximum: CA)	8	Petrella et al. (2006), Trivedi et al. (2008), Jin et al. (2012)
							Distribution: 75 % in CA
							CA manually: CA1, CA4
AD > Controls	Encoding all	140	12	−60	38	R	Precuneus	9	Sperling et al. (2003), Pariente et al. (2005), Petrella et al. (2007)
	Encoding verbal	187	12	−60	38	R	Precuneus	7	Sperling et al. (2003), Pariente et al. (2005), Petrella et al. (2007)
		116	−14	−64	32	L	Precuneus	7	Sperling et al. (2003), Petrella et al. (2007)
	Encoding images	140	12	−60	38	R	Precuneus	8	Sperling et al. (2003), Pariente et al. (2005), Petrella et al. (2007)
		85	−14	−66	32	L	Precuneus	8	Sperling et al. (2003), Petrella et al. (2007)
	Retrieval all							4
	Retrieval verbal							3
	Retrieval images							3
Controls > AD	Encoding all							10
	Encoding verbal							7
	Encoding images							8
	Retrieval all	156	26	−18	−20	R	Hippocampus (maximum: SUB)	4	Grön et al. (2002), Pariente et al. (2005), Rémy et al. (2005)
							Distribution: 50 % in CA, 47 % in SUB
							CA manually: CA1
	Retrieval verbal							3
	Retrieval images	155	26	−18	−20	R	PHG, hippocampus (maximum: SUB)	3	Grön et al. (2002), Pariente et al. (2005)
							Distribution: 42 % in CA, 40 % in SUB, 10 % in EC, 7 % in amygdala (LB)
							CA manually: CA1

n number of voxels, MNI Montreal Neurological Institute, R right, L left, MCI mild cognitive impairment, AD Alzheimer’s disease, CA cornu ammonis, SUB subiculum, CM centromedial, PHG parahippocampal gyrus, EC entorhinal cortex, LB laterobasal, DG dentate gyrus

We also performed subgroup analyses of either verbal or image stimuli (cf. Table 1). Any stimulus with letters or words was depicted as being verbal. Photographs, pictures, patterns or line drawings were categorised in the “images” subgroup. Some stimuli combined verbal and image aspects, i.e. face-name pairs and were consequently included in both subgroup analyses.

Correlation between MMSE and likelihood of altered activation

The Mini-Mental State Examination (MMSE) (Folstein et al. 1975) is still the most frequently used screening test and short neuropsychological exam in fMRI literature to asses a subject’s global cognitive performance. We aimed to investigate the dependence of the likelihood of changes revealed by functional neuroimaging and cognitive performance using mean MMSE scores for Spearman rank correlations, thresholded at a p value of p < 0.05.

Results

Table 2 summarises all significant clusters and provides volume, peak coordinates, macroanatomic as well as cytoarchitectonic regions (if available) and contributing studies. We provide more details in Table 2, including the most probable cytoarchitectonic region of the peak coordinate, spatial distribution of the cluster by giving the percentage in different sub-regions of the whole volume, as well as manual allocation to CA1-4 and finally, number of experiments included in each contrast. The remaining analyses, which are not mentioned in the following results part, failed to reach the significance threshold (cf. Table 2). Furthermore, the pooled analyses mainly replicated the group-specific comparisons. The results are shown in supplementary Table S1. They are essential to assure power of the correlational analyses.

Aberrant episodic memory-related brain activation in MCI patients compared to controls

Compared to non-affected controls, MCI patients showed consistently increased right anterior hippocampal activation (CA2-4, subiculum = SUB) (Fig. 1a) during encoding. A similar cluster in the hippocampus (CA2-4, SUB) extending to the amygdala yielded significance in the image subgroup.

During verbal retrieval, MCI patients showed decreased activation in the right anterior insula (Fig. 2), while attenuated activation was observed in the left anterior hippocampus (CA1, CA4) extending to the left fusiform gyrus (Fig. 1b) for the image stimuli subgroup.

Fig. 2.

Decreased insula and inferior frontal gyrus activation during verbal retrieval tasks in patients with mild cognitive impairment. Patients with mild cognitive impairment showed less activation during verbal retrieval in the right anterior insula extending to the inferior frontal gyrus when compared to controls. All clusters are family-wise error cluster corrected at p < 0.05. R right hemisphere, x, z in Montreal Neurological Institute (MNI) coordinates

Aberrant episodic memory-related activation in AD patients compared to controls

Across all encoding paradigms (verbal and images), AD patients activated the right precuneus stronger than control subjects. During encoding of either verbal or image stimuli, the bilateral precuneus also showed stronger activation (Fig. 3).

Fig. 3.

Stronger precuneus activation and correlation with performance. A convergent finding was stronger bilateral precuneus activation in AD patients compared to controls during image encoding (blue). We furthermore report significant voxel-wise correlation with mini-mental state examination (MMSE) scores in the right precuneus in MCI + AD patients compared to controls during image encoding (magenta). All clusters are family-wise error cluster corrected at p < 0.05. R right hemisphere, x, y, z display Montreal Neurological Institute (MNI) coordinates

During episodic retrieval of all stimuli, the right anterior hippocampus (SUB, CA1) was significantly less active in AD compared to controls (Fig. 1c). Decreased activation remained significant in the image retrieval subgroup in the PHG, the hippocampus (SUB, CA1, entorhinal cortex) as well as the amygdala.

Correlation of cognitive performance with aberrant brain activation in all patient groups

A voxel-wise analysis investigating whether patients’ aberrant brain activation correlated with MMSE scores highlighted four clusters (Table 3). Specifically, lower MMSE scores, which reflect more severe cognitive impairments, are associated with a higher probability of finding altered activation in several regions. Decreased activation in the right anterior insula/inferior frontal gyrus during verbal retrieval was linked to lower MMSE scores in MCI, while AD patients with lower MMSE scores exhibited decreased right hippocampal (SUB, CA1) activation during image retrieval. Stronger right precuneus activation during encoding of images was associated with lower MMSE scores of the pooled group of MCI + AD (Fig. 3). Finally, lower right (para)hippocampal (SUB, CA1, CA3, CA4) and amygdala activation was correlated with lower MMSE performance during image retrieval in the pooled MCI + AD group. Pooling MCI and AD data focused on the continuum of MCI to AD. Given the regional diversity of brain activation patterns reported on group level, we found effects that correlate well with MMSE scores. The method confirmed the disease stage dependency of memory processing network alterations.

Table 3.

Significant correlation of lower MMSE scores with altered activation in specific brain regions

Contrast	Task	r	R/L	MNI coordinates			Macroanatomical and cytoarchitectonic region
				x	y	z
Controls > MCI	Retrieval verbal	−0.8	R	32	28	−2	Anterior insula lobe, inferior frontal gyrus
(MCI + AD) > Controls	Encoding	−0.51	R	10	−58	36	Precuneus
(MCI + AD) > Controls	Encoding images	−0.61	R	10	−58	36	Precuneus
Controls > (MCI + AD)	Retrieval images	−0.85	R	26	−18	−20	PHG, hippocampus (maximum: SUB)
							Distribution: 42 % in SUB, 40 % in CA, 10 % in EC, 7 % in amygdala (LB)
							CA manually: CA1, CA3, CA4

r Spearman correlation coefficient, R right, L left, MNI Montreal Neurological Institute, MCI mild cognitive impairment, AD Alzheimer’s disease, PHG parahippocampal gyrus, EC entorhinal cortex, SUB subiculum, CA = cornu ammonis, LB laterobasal

Discussion

Using a quantitative meta-analysis approach, we characterised episodic memory-related alternations in brain activation patterns in patients with MCI or AD. Our study yielded four main results. First, we identified increased encoding-related brain activation in the hippocampus (mainly CA2-4, also SUB) in early disease stages (MCI) as well as decreased activation in other hippocampal areas (mainly CA1, also CA4) during retrieval. Second, we report that MCI and AD are characterised by decreased PHG activation during episodic retrieval. Third, we found encoding-related stronger activation in the precuneus in AD only. Fourth, higher probability of finding decreased right (para)hippocampal and right anterior insula/inferior frontal activation as well as stronger precuneus activation was associated with worse cognitive performance measures (MMSE scores). The results highlight both the neuropathological role of the MTL and precuneus and their association to cognitive performance and importantly verify specific and probably disease stage-dependent brain activation patterns.

Hippocampal formation

As revealed by our meta-analysis, individuals with MCI showed elevated activation during encoding of all or image stimuli in the anterior hippocampus, mainly in CA2-4, and also in the SUB. Activation in the hippocampal formation in general during encoding has been shown to be essential for long-term storage of memory (Scoville and Milner 2000; Squire et al. 2004) and has also been associated with encoding success (Spaniol et al. 2009). In MCI patients, hippocampal activation has been contradictorily described as both elevated (Celone et al. 2006; Kircher et al. 2007; Hämäläinen et al. 2007) and reduced (Hanseeuw et al. 2011; Hampstead et al. 2011; Giovanello et al. 2012) compared to age-matched healthy controls. These discrepancies may be explained by a longitudinal study suggesting that elevated hippocampal activation only occurs in early stages of the disease and precedes decline in cognition and hippocampal activation (O’Brien et al. 2010). Understanding of the underlying neuropathological processes can be augmented by studies which reported neurofibrillary tangles within the earliest stages of the disease as well as amyloid β protein (Aβ) depositions in later stages (Braak and Braak 1991; Braak et al. 2006). Neurofibrillary tangles lead to dysfunctions and disconnections accompanied by impaired function of the affected regions already in MCI patients (Bokde et al. 2006). It is hypothesised that hippocampal hyperactivation in MCI might represent an early response to AD pathology, which may be of compensative nature or predicts impending hippocampal failure and memory decline. This period of hippocampal hyperactivation is then followed by loss of hippocampal activation as the disease progresses (Fig. 4). It is suggested that the hyperactivation may be an indicator of neuronal stress or directly related to the pathophysiologic process of AD. For instance, hyperactivation was associated with aberrant sprouting of cholinergic fibres (Hashimoto and Masliah 2003), excitotoxicity from the excessive stimulation of NMDA receptors (Hynd et al. 2004) or stress from Aβ-induced neuronal alterations (e.g. excitatory activation triggers compensatory inhibitory mechanisms, which jointly contribute to network dysfunction) (Palop et al. 2007). Postmortem studies have furthermore implied that the entorhinal cortex is prominently affected, the SUB and CA1 severely involved, whereas CA3 and the DG are relatively preserved (Mueller et al. 2010; Small et al. 2011). This may explain why our observed cluster in CA2-4 presumably represents compensatory elevated activation in early stages of the disease (MCI). This hippocampal hyperactivation failed to correlate with consistent MMSE score trends, indicating that it is too rare or simply not related to specific MMSE scores.

Fig. 4.

Schematic course of differential (para)hippocampal, precuneus and insula/IFG activation during episodic memory tasks over the course of the disease. Retrieval task-induced hippocampal (at first predominantly CA1, later also SUB and PHG; red line) activation declines over the course of the disease from MCI to fully expressed AD compared to controls in hippocampal sub-regions which are specifically and severely affected by AD pathology. This is accompanied by attenuated verbal retrieval task-related activation in the insula and inferior frontal gyrus (IFG) in MCI only (yellow line). Encoding task-induced activation is higher in MCI patients than in controls in hippocampal sub-regions relatively spared from AD pathology (CA2-4), probably compensatory for the loss of memory function and activation (green line). Precuneus activation is stronger during encoding conditions in severely affected AD patients than in controls. CA cornu ammonis, AD Alzheimer’s disease, MCI mild cognitive impairment, SUB subiculum, PHG parahippocampal gyrus, IFG inferior frontal gyrus

Additionally, we found that during functional neuroimaging retrieval experiments using image stimuli, MCI compared to controls showed attenuated left anterior hippocampal (mainly CA1, also CA4) activation extending to the left fusiform gyrus. In addition to the above-mentioned involvement in encoding tasks, the hippocampus has been previously associated with episodic retrieval (Hayes et al. 2004; Kühn and Gallinat 2013), especially retrieval success (Nyberg et al. 1996; Schacter et al. 1999; Hayes et al. 2004). The fusiform gyrus is anatomically and functionally adjacent to the hippocampus and has been highlighted during face recognition (McCarthy et al. 1996; Lim et al. 2011) and colour processing (Wang et al. 2013). The main contributing studies employed pictures, face-name pairs and black-and-white line drawings of real objects. In addition to the above-mentioned histopathological findings in the hippocampus in AD, the CA1 region has shown decreased metabolic responses (Salehi and Swaab 1999). Decreased activation within the hippocampal formation itself is believed to result in decreased activation of consecutive regions (Small et al. 2011) such as the fusiform gyrus. Early AD pathology revealed by histopathological as well as neuroimaging studies might explain decreased activation within the hippocampus (mainly CA1) in MCI.

In our meta-analysis of AD patients compared to controls, the right anterior hippocampus (especially the SUB and CA1) was significantly less activated during retrieval of all kinds of probed stimuli. In the subgroup of image stimuli, we found a similar cluster in the SUB and CA1 extending to the entorhinal cortex as well as the amygdala. In addition to the above-mentioned functions of the hippocampus, retrieval-related activation has been linked to the SUB (Small et al. 2011). This decreased (mainly SUB and CA1) activation during image retrieval was associated with lower MMSE scores. Taken together, reduced hippocampal (SUB, CA1) activation apparently precisely colocalises with AD pathology and is associated with impaired retrieval memory performance in AD.

Correlational analysis of the pooled analysis of MCI and AD revealed that reduced right posterior PHG and hippocampal (mostly SUB and CA) activation during retrieval of image stimuli was associated with lower MMSE scores.

Reduced PHG and hippocampal (mainly CA1 and SUB) activation during episodic retrieval tasks is the most consistent effect of AD in early as well as late stages, also related to impaired performance.

In summary, we report that early (presumably compensatory) encoding-related hippocampal hyperactivation (mainly CA2-4) in MCI patients is followed by retrieval-related declining activation in both MCI and AD patients, especially in the CA1, SUB and PHG—the sub-regions specifically affected by AD pathology.

Notably, all studies that mainly contributed to clusters in the MTL contrasted memory-related brain activation to brain activation during a demanding referential condition (e.g. ‘hits > misses’ or ‘novel > repeated’). It has been shown that undemanding referential conditions (e.g. visual fixation of a cross) involve the MTL, and especially hippocampal activation, comparable to that in memory tasks (Stark and Squire 2001). Thus to explore hippocampal activation during memory tasks, a (demanding) referential condition is needed which does not involve hippocampal activation.

Precuneus

While AD patients were characterised by stronger encoding-related activation in the precuneus, no aberrant precuneus activation was found within the 20 included studies on MCI patients compared to controls. The precuneus is presumably involved in complex behavioural tasks, especially visuo-spatial imagery, self-processing and consciousness (for a review see Cavanna and Trimble 2006). Moreover, the precuneus steadily plays a main role in the so-called ‘default-mode network’ which has the highest metabolic rates and blood perfusion during resting conditions (Gusnard and Raichle 2001; Schilbach et al. 2012). Considering the normal function of the precuneus, activation during memory tasks has been described as an encoding/retrieval flip (Huijbers et al. 2012) implying that the BOLD-signal is negative (lower than during resting baseline conditions) in encoding tasks (“deactivation”). Following this hypothesis, the relatively stronger precuneus activation found during encoding in AD patients compared to controls might reflect “less deactivation”. Nevertheless, the ideas of areal baseline and deactivations remain highly controversial. Stronger precuneus activation might also reflect a compensatory or dysfunctional increase of activation in response to amyloid deposition. In contrast, relatively increased activation compared to baseline is observed during retrieval tasks (“activation”). AD-related histopathological changes in the precuneus have mainly been associated with extracellular Aβ depositions in later stages of the disease, such as Braak IV-VI (Braak and Braak 1991; Braak et al. 2006). In vivo imaging of fibrillary amyloid depositions via Pittsburgh compound B-PET (PiB-imaging) has marked the precuneus and posterior cingulate as regions that are consistently affected rather early. Healthy subjects with elevated levels of Aβ depositions showed highly aberrant patterns of default network fMRI activation (Celone et al. 2006; Sperling et al. 2009). This link between pathology and function has been hypothesised as a result of loss of connectivity and synapses in this region (Scheff et al. 2013). Less deactivation or stronger activation during demanding encoding tasks compared to undemanding tasks in the precuneus has been suggested as prognostic for the development of AD in at-risk stages such as MCI (Petrella et al. 2007). Our observation that MCI patients were spared from precuneus dysfunctions is supported by a study reporting that cholinergic activity was maintained regardless of PiB binding and amyloid burden in postmortem brains (Ikonomovic et al. 2011). This suggests maintained function of the precuneus until severe stages of the disease (Fig. 4). We showed that precuneus dysfunctions during episodic memory tasks do not consistently occur in MCI and, thus, investigation of task-related activation in prodromal AD in the precuneus has to be more differential.

In AD, bilateral precuneus activation was significantly strongly activated in the subgroups of verbal or image stimuli. In a former meta-analysis of functional neuroimaging studies of episodic memory tasks in AD, precuneus activation has been described as both greater and less activated compared to controls, notably during both encoding and retrieval phases (Schwindt and Black 2009). In line with our findings, experimental studies reported stronger activation of the precuneus in AD compared to controls during encoding tasks (Frings et al. 2010; Browndyke et al. 2012).

The correlation analysis in the pooled group MCI + AD of MMSE scores and likelihood of finding altered activation showed significant correlation between low MMSE scores and the right precuneus during encoding (all and image subgroups). The contributing studies employed AD patients only. This supports the hypothesis that functional deficits in the precuneus follow the pathological amyloid deposition with delay, only in patients achieving MMSE scores on average of less than 26, and are most consistent during encoding of images. Stronger activation in the precuneus co-localises with AD histopathology and might underlie the memory encoding deficits.

Insula, inferior frontal gyrus

During retrieval of verbal stimuli, individuals with MCI presented attenuated activation in the right anterior insula and inferior frontal gyrus. The insula is a part of the brain believed to integrate perceptions and emotional situations with cognitive processes. The anterior part in particular is connected to the frontal lobe and specifically active during attention and memory retrieval paradigms (here only right-hemispheric) revealed by meta-analysis (Kurth et al. 2010). It is further believed to play an active role in language processing (McCarthy et al. 1993; Price 2000; Kurth et al. 2010), whilst its function was more recently described as a very basic planning of general orofacial functions independent of actual words (Price 2010). Bilateral, rather than one-sided prefrontal activation (including the inferior frontal gyrus) has been observed in older rather than younger subjects, presumably to maintain declining cognitive functions (Grady 2012), e.g. during verbal or memory tasks. Neurofibrillary tangles have been reported in this region in the latest stages IV to VI in AD (Braak et al. 2006). MCI patients showed impaired verbal memory performance (Giovanello et al. 2012) and activation likelihood in the afore-mentioned cluster was correlated with the lowest MMSE scores during retrieval of verbal stimuli. Summing up, decreased anterior insula/inferior frontal gyrus activation is phenotypic for verbal retrieval performance deficits in MCI patients.

Stimulus specificity

It has been shown that tasks with images and verbal stimuli preferably activate right-to-bilateral and left-hemispheric brain regions, respectively. Since AD pathology is asymmetric yet not lateralised (Derflinger et al. 2011), we expected disease-related brain activation alterations to be more obvious in regions of greater task involvement or stimulus dependency. Increased right hippocampal activation in MCI patients compared to controls during encoding was found with most contributions from studies employing objects and face-name pairs as stimuli. Stronger left-hemispheric activation was earlier observed during encoding of verbal stimuli, while images, faces, scenes or patterns have been described as being processed bilaterally or right-hemispherically (Kelley et al. 1998; Schacter et al. 1999; Golby et al. 2001; Rosazza et al. 2009). This right-hemispheric hippocampal hyperactivation is in line with those previous findings. We further found mostly right-hemispheric hippocampal activation during retrieval tasks in MCI and AD patients independent of stimulus material or in the image stimuli subgroups. Although the only left-hemispheric cluster of decreased activation in MCI compared to controls in the hippocampus appeared in the image retrieval subgroup, we identified clear verbal aspects in the material used. Stimuli consisted of face-name-pairs, face-occupation pairs and, as stated in the original publication (Trivedi et al. 2008), “nameable” objects. Reduced right anterior insula/inferior frontal gyrus activation in MCI patients during verbal retrieval tasks seemed to contradict the above-mentioned left-hemispheric dependency of verbal tasks at first. However, according to a recent meta-analysis on insula functions, bilateral verbal processing within the insula seems to be an exception from general left-hemispheric lateralisation in verbal tasks (Kurth et al. 2010). In further support of our findings, unilateral right anterior insula activation has been reported as being specific for memory tasks (Kurth et al. 2010).

Stimulus specificity has not yet been reported for the precuneus during encoding tasks.

Limitations

Although we used the meta-analytic approach to converge previous findings, limitations still exist in the heterogeneity of included studies such as stimuli and diagnosis of patients. First, the considered patient groups exhibit a certain level of heterogeneity as we included studies with single- and multidomain aMCI as well as very few non-aMCI as diagnostic entity. Future studies should examine more homogeneous prodromal AD groups by applying the revised diagnostic criteria by Dubois et al. (2010). It is important to note that none of the included studies reported cerebrospinal fluid (CSF) examination results or amyloid status revealed by PiB-PET. Future investigations combining this important information with task-related fMRI results for patients and controls could help to unravel the possible co-existence or co-localisation of tau and amyloid accumulations and functional impairment. Furthermore, AD patients in contrast to control groups were frequently on medication which itself has been shown to influence memory-related brain activation. Given that published papers do not typically report patient-to-patient medication, we had to ignore the potential effect of medication. Moreover, we summarised different stimuli types, variable memory tasks and time spans between encoding and retrieval phases, varying task complexity and demand and several baseline or reference conditions. Future studies may (1) employ demanding referential conditions that do not depend on MTL activation and (2) check for possible differences of short-term vs. long-term memory-related activation patterns in AD compared to controls to increase the understanding of the relevance of histopathological AD-related findings in the MTL. Finally, in the included studies’ magnetic field strength ranged from 1.5 to 4.0 T while the majority of data stemmed from 3.0 T scanners. In the future more studies using ultra high field fMRI (e.g. MTL during episodic memory tasks) may unravel the nature and exact cytoarchitectonic localisation of hippocampal activation in MCI and AD. While region of interest (ROI) based analyses may offer a higher spatial resolution, this approach requires a strong a priori hypothesis as additional relevant neural correlates might be overseen not using a whole-brain method. An additional problem in that endeavour was the high variety of chosen shapes and sizes of ROIs.

Conclusions

This quantitative meta-analysis provides evidence for a specific and probably disease stage-determined brain activation pattern linked to the pathognomonic AD feature of episodic memory loss. In this meta-analysis, cytoarchitectonic labelling of clusters revealed by neuroimaging studies was substantial for a precise understanding of brain (dys)functions. CA1 and the SUB as well as the PHG are not only relatively selectively and severely affected by AD pathology, but also display declining activation in MCI and AD patients during retrieval tasks. Notably, CA2-4 is relatively preserved from AD pathology and yet reveals (presumably compensatory) hyperactivation in early disease stages (MCI). Decreased MTL activation in AD is accompanied by stronger activation during memory encoding tasks in the precuneus, indicating compensatory or rather dysfunctional mechanisms. These key regions might operate as surrogate markers for further clinical trials.