Abstract
Information that is processed with reference to the self (i.e., self-referential processing, SRP) is generally associated with better remembering than information processed in a semantic condition. This benefit of self on memory performance is called self-reference effect (SRE). In the present study, we assessed changes in the SRE and SRP-related brain activity in patients diagnosed with mild cognitive impairment or early Alzheimer’s disease (MCI/AD). Fifteen patients with confirmed amyloid-β deposits (positive florbetapir-PET scan) and 28 healthy controls (negative florbetapir-PET scan) were included. Participants either had to judge personality trait adjectives with reference to themselves (self condition) or to a celebrity (other condition), or determine whether these adjectives were positive or not (semantic condition). These adjectives were then presented with distractors in a surprise recognition task. Functional MRI data were acquired during both the judgment and recognition tasks. The SRE was observed in controls, but reduced in patients. Both controls and patients activated cortical midline structures when judging items with reference to themselves, but patients exhibited reduced activity in the angular gyrus. In patients, activity at encoding in the angular gyrus positively correlated with subsequent recognition accuracy in the self condition (self accuracy). This region also exhibited significant hypometabolism and Aβ burden, both related to self accuracy. By contrast, there were no differences in brain activity during recognition, either between the self and semantic conditions, or between groups. These results highlight SRE impairment in patients with MCI/AD, despite intact activity in cortical midline structures, and suggest that dysfunction of the angular gyrus is related to this impairment.
INTRODUCTION
Self refers both to a set of complex and multidimensional mental representations about ourselves and to the associated flow of self-consciousness.It is the cornerstone of our individual identity, and helps to maintain a feeling of continuity over time [1]. Self and memory processes are known to be closely interrelated. From an experimental standpoint, this interaction between self and memory is attested by better memory performances for material processed with reference to the self, rather to another person or in a semantic judgment condition. This is known as the self-reference effect (SRE) [2]. Functional MRI studies conducted in healthy young adults have revealed that the processing of items with reference to the self (i.e., self-referential processing, SRP) is subserved by several cortical midline structures, including the medial prefrontal cortex (MPFC), the anterior and posterior cingulate cortices (ACC and PCC), and the precuneus [3, 4]. Only a few studies have examined the neural substrates of SRE (i.e., brain activity associated with the retrieval of items encoded with reference to the self) [5–7]. These reported PCC and MPFC involvement when healthy young participants retrieved items that had been encoded with reference to the self, compared with a control condition (see [3] for a meta-analysis).
Alzheimer’s disease (AD) is characterized by early memory deficits. In addition, the key structures subserving the SRE (i.e., cortical midline structures) are affected early on the course of the disease [8]. Investigating the SRE and SRP in patients with AD is therefore particularly relevant. Several behavioral studies have reported an absence of SRE in patients. When Lalanne et al. [9] administered an incidental learning task featuring personality trait adjectives in three conditions (perceptual, semantic and self), followed by a free recall task and a yes/no recognition task, they did not observe any SRE in patients with AD. Similarly, when Genon et al. [10] compared the recognition performances of patients and healthy older adults after two encoding conditions (self-referential versus other-referential), they found an impairment of SRE in AD patients. Using the same paradigm as in the present study, Leblond et al. [11] examined the effects of material (positive or negative personality trait adjectives encoded with reference to the self, or a distant other, or processed semantically) and identity valence in amnestic mild cognitive impairment (MCI). The SRE was only observed when the material was positive. In contrast, when adjectives were negative, MCI patients’ performances depended on the valence of their self-representations, with negative self-representations correlated with poor recognition of negative adjectives. A group of AD patients was also included in this study, but patients’ performances were too low to be subjected to similar analyses.
Some authors have also investigated the benefits of self-processing for recollection-based processes, termed the self-reference recollection effect (SRRE [12]). The SRRE is assessed by means of the Remember/Know procedure, which allows distinguishing between recollection-based and familiarity-based recognition. Using this procedure, Genon et al. [10] found an alteration of SRRE in AD patients. However, a few studies have reported preservation of the SRRE in AD patients when the emotional valence of items is taken into account, although this effect is not consistent across studies [9, 14].
Neuroimaging studies investigating the SRE and SRP in patients with AD or MCI are rare, and have provided discrepant results. Ries et al. [15], for instance, reported similar levels of PCC activity in patients with MCI and healthy controls during a SRP paradigm. However, these results were not replicated in another study, conducted in MCI patients, in which the same authors tried to relate brain activity associated with self-appraisal and anosognosia [16]. Zamboni et al. [17] found that patients with AD, but not MCI, exhibited decreased activation, compared with controls, in the MPFC and anterior temporal areas when asked questions about themselves versus a partner. By contrast, Genon et al. [10] reported similar levels of MPFC activity in both healthy controls and patients with AD during self-relevance versus other-relevance (famous person) judgments. Differences in experimental paradigms, including the degree of closeness with the other person, may partly explain these discrepancies. However, a specific study addressing this issue is warranted to understand how the degree of closeness with the other person may modulate brain activity, and how this effect could be modified in AD.
As for the retrieval of information encoded with reference to the self, Genon et al. [10]’s study with patients with AD failed to find any difference in brain activity associated with the recognition of items that had been processed with reference to the self, versus another person. Paradoxically, despite the fact that the SRE was significantly impaired in AD patients, there was no difference in brain activity between these patients and controls. The authors suggested that their SRE impairment might be related to subtle functional changes, probably in terms of connectivity [10]. However, the reported absence of any difference in brain activity during recognition could also reflect the fact that the control condition (processing items with reference to another person, who happened to be a celebrity) also engaged self-reference processes [5, 18].
Up to now, research has therefore yielded mixed results concerning the preservation or impairment of the SRE in AD, and the neural substrates of the SRE and SRP have seldom been explored. In this context, the present study was intended to determine whether the SRE is preserved or not in MCI and early AD, and to unravel the neural substrates of SRP using fMRI. In addition, we also collected structural MRI, FDG-PET, and florbetapir PET data to determine whether the functional disorders potentially revealed using the fMRI task were associated to biomarkers of neuronal injury and amyloid-β (Aβ) deposits in AD.
MATERIAL AND METHODS
Participants
All the participants were enrolled in a larger, multimodal imaging study of early-stage AD (“Imagerie Multimodale de la maladie d’Alzheimer à un stade Précoce”, IMAP) conducted in Caen (France), and some of them had been included in previous publications from our team [19–24].
Participants were all right-handed, had at least 7 years of education, and had no history of alcoholism, drug abuse, head trauma, or psychiatric disorder. The IMAP study was approved by a regional ethics committee (Comité de Protection des Personnes Nord-Ouest III) and registered with http://clinicaltrials.gov (no. NCT01638949). All participants gave their written informed consent to the study prior to the investigation.
A group of patients with MCI or AD was recruited from local memory clinics and selected according to internationally agreed criteria. Clinical diagnoses were assigned by consensus under the supervision of senior neurologists (VdLS and SB) and neuropsychologists (AP and SE). The patients with AD met the standard clinical criteria of the National Institute of Neurological and Communicative Disorders and Stroke, and the Alzheimer’s Disease and Related Disorders Association (NINCDS-ADRDA) for probable AD [25]. The patients with MCI were recruited according to Petersen’s [26] criteria (memory complaint, objective episodic memory deficits, not demented).
As the aim of this study was to assess the SRE and the neural bases of SRP in early AD, we selected patients according to four additional criteria. First, they had to be at the early stage of the disease, in order to be able to perform the fMRI task. Second, they had to have had a positive florbetapir-PET scan, that is with a high probability of AD etiology, within 18 months of the fMRI scan. Third, they had to provide at least eight answers in every condition during both encoding and recognition. Fourth, their EPI data had to pass a quality control to check for the absence of artifacts, followed by a visual examination (see below). We collected data from 21 patients with MCI/AD, but only 15 patients (7 AD and 8 MCI, including two patients who developed AD during the 18-month follow up) met the criteria described above and were thus included in the study. All six patients were excluded after the quality check on functional MRI data using the TSDiffana routines (http://imaging.mrc-cbu.cam.ac.uk/imaging/DataDiagnostics) followed by visual examination performed by the same operator (MG) to detect large movements and/or artifacts.
The data from these 15 patients were compared with those obtained from 28 healthy controls. Data of 7 MCI and 6 AD patients were already included in our previous behavioral study conducted in a whole group of 20 MCI and 20 AD patients [11]. Controls were recruited following a clinical examination intended to exclude participants with a major neurological or psychiatric disorder, past or present. They also underwent a neuropsychological assessment, including tests of episodic and working memory, language, executive functions and visuospatial abilities, and FDG and florbetapir PET scans. Controls with a positive florbetapir PET scan were excluded from the analyses (see [21] for the method used to determine the positivity threshold).
Participants’ main demographic and clinical data are reported in Table 1. Patients were slightly older than controls, but the difference between the two groups failed to reach significance (p = 0.08). The proportion of women and the level of education were comparable across the two groups. As expected, the Mini-Mental State Examination (MMSE) score was lower in patients than in controls.
Self-reference task
The fMRI self-referential paradigm we used (illustrated in Fig. 1) was inspired by the literature [6, 27–30], and had previously been used in our laboratory with a group of young healthy individuals [31].
The material consisted of a list of 204 personality trait adjectives, selected from 463 adjectives taken from a French language dictionary (http://atilf.atilf.fr/). These adjectives were selected according to familiarity and valence ratings provided in a pre-experiment by young and older individuals with low and high education levels. For the purpose of the fMRI experiment, we used these 204 adjectives to construct six lists of 24 adjectives for two functional runs per condition in the encoding session, and two lists of 30 adjectives to be used as distractors in the retrieval session. The adjectives in these eight lists were counterbalanced for familiarity, valence, and number of letters, so that these parameters did not differ between conditions. In the experiment, participants were shown the selected adjectives one by one, and asked either to indicate whether or not the adjective described either themselves (self condition) or a celebrity (other condition), or to judge whether the adjective was positive or not (semantic condition).
After a training session outside the scanner, participants performed two judgment sessions lasting about 7 min each and including 72 items (24 self, 24 other, and 24 semantic). Each adjective was displayed on a screen for 3500 ms, along with a brief instruction about the nature of the judgment to be performed: “Myself?” for the self condition, “J. Chirac?” (former French president), or “J. Hallyday?” (famous French singer) for the other condition, and “Positive?” for the semantic condition. Each item was followed by a fixation cross for 1000–3000 ms (mean: 2000 ms). Participants had to answer “Yes” or “No” with their left or right index finger on a two-button keypad.
Then came a surprise recognition task featuring the items that had been incidentally encoded during the judgment phase. This task was divided into two functional runs, each lasting about 8 min and including 84 adjectives (18 old self, 18 old other, 18 old semantic, and 30 new adjectives as distractors, with the same numbers of positive and negative items in each category). Each adjective was displayed on the screen for 3500 ms, coupled with the brief instruction “Already seen?”, and was followed by a fixation cross for 1000–3000 ms (mean: 2000 ms). Participants once again had to answer “Yes” or “No” with their right or left index finger on a two-button keypad. Adjectives seen in the first judgment session were presented during the first recognition run, and adjectives from the second judgment session were presented during the second recognition run. The order of presentation of the conditions was optimized using a genetic algorithm, in order to enhance the detection of fMRI differences between experimental conditions in the subsequent SPM statistical analyses [32]. The lists of adjectives used for each condition were also counterbalanced across participants, as was the side of the “Yes” versus “No” answer on the keyboard. Items were displayed using E-Prime software (Psychology Software Tools, Pittsburgh, PA) implemented within IFIS System Manager (Invivo, Orlando, FL).
Neuroimaging data acquisition
MRI
MRI data were acquired on a 3T Achieva scanner (Philips). All participants first underwent high-resolution T1-weighted anatomical volume imaging using a 3D fast field echo (FFE) sequence (3D-T1-FFE sagittal; TR = 20 ms, TE = 4.6 ms, flip angle = 20, 180 slices, slice thickness = 1 mm, no gap, FoV = 256×256 mm2, matrix = 256×256, in-plane resolution = 1×1 mm2). This acquisition was followed by a high-resolution T2-weighted spin echo anatomical sequence (2D-T2-SE sagittal) and a non-echoplanar imaging (non-EPI) T2-star volume (2D-T2*-FFE axial) sequence. The functional sequences were acquired using an interleaved 2D T2-star EPI sequence (2D-T2-star-FFE-EPI axial; SENSE factor = 2, TR = 2382 ms, TE = 30 ms, flip angle = 80°, 44 or 42 slices after sequence adjustment, slice thickness = 2.8 mm, no gap, matrix 80×80, FoV = 224×224 mm2, in-plane resolution = 2.8×2.8 mm2 , 172 volumes per run for judgments, and 199 for recognition).
PET data
Participants also underwent FDG- and florbetapir-PET scans. These data, used for complementary analyses, were acquired on a Discovery RX VCT 64 PET-CT scanner (General Electric Healthcare) with a resolution of 3.76×3.76×4.9 mm (FoV = 157 mm). Forty-seven planes were obtained with a voxel size of 1.95×1.95×3.27 mm. A transmission scan was performed for attenuation correction prior to the PET acquisition. For FDG-PET acquisition, participants were fasted for at least 6 h before scanning. After a 30-min resting period in a quiet and dark environment, ≈180 MBq of FDG were intravenously injected as a bolus. A 10-min PET acquisition scan began 50 min post-injection. For florbetapir-PET acquisition, each participant underwent a 20-min PET scan, beginning 50 min after the intravenous injection of ≈4 MBq/kg of florbetapir.
One control participant could not perform the FDG-PET scan, and one patient was excluded from the FDG analyses owing to poor data quality. For two patients and seven controls, the FDG scan was done at the 18-month follow-up, whereas the fMRI scan took place at inclusion.
Data analysis
As previous studies conducted in healthy volunteers showed that the processes and the neural networks involved for judgments with reference to the self or another person are very close [3, 5], we chose to use the semantic condition as a reference, rather than the other condition.
Behavioral data
Recognition accuracy was calculated for both conditions (self and semantic) as follows: (number of items correctly recognized in a given condition/number of items presented in that condition) – (number of false alarms/number of distractors presented). This score was then entered in a repeated-measures analysis of variance (ANOVA) with condition (self, semantic) as a within-participants factor and group (patients versus controls) as a between-participants factor. In order to assess the benefits of self-reference encoding for memory performance (SRE), we also computed a self-reference benefit index for each participant, by subtracting recognition accuracy in the semantic condition from accuracy in the self condition. For each group, this index was tested in a t test comparison with zero, to confirm or refute the presence of a self-related benefit. Statistical analyses were performed using Statistica software (StatSoft Inc.,Tulsa, OK).
Neuroimaging data: Preprocessing and analysis
MRI data
Structural MRI data were segmented, normalized and modulated (non-linear only) with the unified segmentation algorithm (“Segment” procedure) using Statistical Parametric Mapping 12 (SPM) software (Wellcome Trust Centre for Neuroimaging, London, UK) to obtain maps of local gray matter (GM) volume. Finally, data were smoothed with a 10-mm full width at half maximum (FWHM)kernel.
fMRI data
The first six volumes of the scanning session were discarded, to deal with T1 equilibration effects. The preprocessing procedure for fMRI data, detailed in [33], was performed using SPM8 (Wellcome Trust Centre for Neuroimaging, London, UK). Briefly, after slice timing correction, data were realigned on the first volume of each run and normalized to the MNI template. Images were then quantitatively normalized to the cerebrospinal fluid (CSF) signal from the lateral ventricles, in order to take account of AD-related brain atrophy [34]. Finally, the data were smoothed with an 8-mm FWHM kernel.
Functional MRI data were analyzed using SPM8 in a two-step analysis, taking intra- and interindividual variance into account. First, for each participant, we applied two general linear models, one for the judgment session and the other for the recognition one. For the former, a simple model was used by modeling the following four conditions: responses in the self, other, and semantic conditions, plus no responses. For the latter, nine conditions were modeled, depending on the condition in which the item had been presented during the judgment (self, other, or semantic) and the response given by the participant (old items correctly recognized or not). Correct rejections of distractors (CR), false alarms (distractors identified as items already seen), and no responses were added (see Fig. 1). Motion parameters (translations and rotations in the three axes), calculated during the realign step of the preprocessing, were also included in these models. All these conditions allowed us to calculate δ functions, which we convolved with a canonical hemodynamic response function to obtain the hemodynamic response. The response time and valence of each adjective were also modeled, to gain an accurate and reliable measure of first-level noiseestimates.
Based on these conditions, all individual contrasts were entered in second-level analyses (flexible factorial designs) to compare the two groups. In order to assess SRP during encoding and retrieval, we computed the following contrasts of interest: “self – semantic” for the judgment session; and “self correctly recognized – semantic correctly recognized” for the recognition session. A third “items correctly recognized – CR” contrast was added and used as a mask for the “self correctly recognized – semantic correctly recognized” contrast. The fMRI data of one of the participants were excluded from the recognition analysis because of a technical problem during scanning.
For the direct comparison between groups, we performed analyses using a mask that corresponded to the logical disjunction (SPM global conjunction) of the groups’ brain activity for each contrast.
Importantly, we created two binary analysis masks. The first one, used in first-level analyses, was calculated for each participant from the conjunction of the GM voxel-based morphometry (VBM) segmentation of T1 and T2* volumes (only including voxels with values greater than 0.15 and 0.05, respectively). The second one, used in second-level analyses, was calculated in two steps. First, we calculated the means for GM (meanGM), white matter (meanWM), and CSF (meanCSF) from the T1 VBM segmentation of all controls and patients, then we applied the following formula: (meanGM >meanWM) ∩ (meanGM >meanCSF) ∩ (meanGM >0.3).
Moreover, in order to remove the effect of age in our analyses, this variable was added as a confounding factor in the two models, as well as in the complementary analyses.
The statistical threshold was set at p < 0.001 uncorrected, with a cluster size of at least 20 voxels. All coordinates are reported in MNI space.
PET data
Both FDG and florbetapir PET data were corrected for partial volume effects using PMOD (PMOD Technologies Ltd, Adliswil, Switzerland), coregistered to their corresponding MRI, normalized to MNI space using the deformation fields calculated during the VBM procedure applied to the PET-related T1-MRI sequence, and finally quantitatively scaled using cerebellar GM as a reference. An overall neocortical florbetapir-PET standardized uptake value ratio (SUVr) value was then computed for each participant from the florbetapir-PET data, using a neocortical GM mask corresponding to the conjunction between the neocortex defined in the AAL atlas [35] (all regions except for the cerebellum, hippocampus, amygdala and subcortical gray nuclei) and a whole-brain GM mask computed from our sample of normalized GM images. Finally, images were smoothed with a 12-mm FWHM kernel.
RESULTS
Behavioral results
Recognition accuracy in the self and semantic conditions is reported in Table 1. A repeated-measures ANOVA conducted on recognition accuracy revealed a significant main effect of group (F(1, 41) = 31.7, p < 0.001), with poorer performances by patients with MCI/AD, compared with controls, and a significant effect of condition (F(1, 41) = 10.52, p < 0.01). Thus, SRP led to better recognition accuracy than semantic processing (p < 0.001). The Group x Condition interaction was not significant (F(1, 41) = 0.8, p > 0.38). The lack of significant interaction between group and condition could indicate that SRE is preserved in MCI/AD patients. However, the comparison of the self-reference benefit index (i.e., recognition accuracy in the self condition minus recognition accuracy in the semantic condition) to zero revealed that controls benefited from SRP (p < 0.01), whereas patients did not (p = 0.12).
Neuroimaging results
Brain activity associated with self-referential processing (SRP) during encoding
In healthy controls, SRP during incidental encoding (i.e., “self – semantic” contrast during the judgment phase) elicited activity in a broad neural network that included several cortical midline structures (MPFC, ACC and PCC, precuneus), as well as the angular gyrus bilaterally, left hippocampus, left and right middle temporal gyri, and left orbitofrontal gyrus (p < 0.001; Table 2; Fig. 2A). The same pattern of brain activity was observed in the patients with MCI/AD, with additional activation of the right putamen and left lingual gyrus (Table 2;Fig. 2A).
Between-group comparisons revealed that the left angular gyrus was more activated in controls than in patients (p < 0.001). No brain area was significantly more activated in patients than in controls (Table 2; Fig. 2B).
In a post-hoc analysis, we extracted the mean activity for the cluster located in the angular gyrus that had been found to be significantly more activated in controls than in patients in the “self – semantic” contrast. Data were extracted in both the self and semantic conditions, for each group of participants. Statistical analysis (t test comparison with zero) failed to reveal any significant activity in this cluster for patients in the self condition (mean activity (±SD) = –0.06 (±1.1), p = 0.83), whereas this region was significantly activated in controls (mean activity (±SD) = 0.6 (±0.8), p < 0.001). In the semantic condition, this region was deactivated in both groups (mean activity (±SD): patients = –0.49 (±0.84), p < 0.05; controls = –0.49 (±0.81), p < 0.01). Between-group analyses (t tests with age as confounding factor) revealed a significant group difference in the self condition (p < 0.05), with greater activity in controls than in patients, but not in the semantic condition (p = 0.64; Fig. 2B.1).
Next, we conducted a correlation analysis in patients between self-related activity in the angular gyrus and recognition accuracy in the self condition (with age as confounding factor). This analysis revealed a significant positive correlation between the two variables (r = 0.56, p < 0.05; Fig. 2B.2), indicating that the greater the activity in the angular gyrus, the better the recognition of self-processed items. We also added the MMSE score as confounding factor. In this case, we only observed a trend toward signification (p = 0.084). For the control group, the same analysis revealed no significant correlation between the two variables (r = –0.07, p = 0.75).
In order to address the specificity of the association between self recognition accuracy and activity in the angular gyrus, we conducted a similar analysis using activity extracted in the cluster located in the PCC (region obtained in the contrast “Self-Semantic” in both groups; MNI coordinates of the peak: [–4 –56 28]), one of the key structure of SRP, and correlated these values with Self recognition accuracy. This analysis failed to reveal any significant correlation between the two variables neither in patients (r = 0.3, p = 0.3) nor in controls (r = 0.1; p = 0.6).
We then carried out complementary analyses to determine whether this area showing reduced activation in the self condition and a correlation with recognition of self-processed items was also affected by atrophy, hypometabolism or Aβ deposition. To do so, we looked at GM volume changes, metabolism and Aβ burden in this region by extracting mean values from the structural MRI, FDG-PET and florbetapir-PET data, respectively. Age was also added as confounding factor. These analyses revealed no significant atrophy of this region (p = 0.42; Fig. 3A), but significant hypometabolism and greater Aβ deposits, in patients compared with controls (p < 0.01; Fig. 3B, C). Moreover, recognition accuracy in the self condition correlated positively with metabolism (r = 0.38, p < 0.05; Fig. 3B, right panel) and negatively with Aβ burden (r = –0.54, p < 0.001; Fig. 3C, right panel) in this region in both patients and controls, although no significant relationships were found when each group was considered separately, except a positive correlation between glucose metabolism in the angular gyrus and Self recognition accuracy (r = 0.64, p < 0.05) in patients. There was no significant correlation with GM density (r = –0.03, p = 0.85; Fig. 3A, right panel), whichever group was considered.
Brain activity associated with SRP during recognition
In healthy controls, comparing brain activity associated with recognition of incidentally encoded items (irrespective of condition) versus CR revealed increased activity in the angular gyrus, middle frontal and middle temporal gyri, precuneus and PCC, medial frontal cortex, lingual gyrus and caudate nucleus (all p values <0.001; Table 3), all clusters being located in the left hemisphere.
The same contrast in the patients with MCI/AD revealed a similar pattern of activity, albeit less extensive, with activation in the left superior and middle temporal gyri, left angular and middle frontal gyri, and left precuneus and PCC (Table 3). Between-group comparisons failed to reveal any region that was more activated in patients than in controls, and vice versa.
In order to highlight the pattern of brain activity associated with the recognition of self-processed items, the “self recognized - semantic recognized” contrast was masked by the network associated with the recognition of previously seen items versus distractors (i.e., “correctly recognized items irrespective of condition – CR” contrast; inclusive masking at p < 0.05). This analysis failed to reveal any cluster with significant activation in either controls or patients (p < 0.001; Table 3), indicating that there was no difference in brain activity between the recognition of self- and semantically processed items.
DISCUSSION
The aim of the present study was to investigate the neural substrates of self during both encoding (i.e., processing information with reference to oneself, SRP) and recognition (i.e., memory advantage for items encoded with reference to the self, SRE) in patients with MCI or AD patients with a high likelihood of AD etiology (confirmed Aβ deposits). The SRE was observed in healthy controls, but not in patients. From a functional standpoint, similar patterns of activity were found in healthy controls and patients with MCI/AD during SRP, including cortical midline structures, temporal areas, and the hippocampus, but patients showed reduced activity in the left angular gyrus, compared with controls. Interestingly, activity in this region was positively correlated with patients’ recognition accuracy in the self condition. Finally, our data failed to reveal any difference in the brain activity associated with the recognition of self-encoded items, as opposed to items that had been processed semantically, either in patients or incontrols.
The memory performances of the healthy older adults were better for personality trait adjectives they had judged with reference to themselves, rather than in the semantic condition, in line with previous studies [18, 36]. By contrast, performances of the patients with MCI/AD were lower than those of controls, and almost equivalent in self and semantic conditions, indicating that the SRE is altered in early AD [9, 10].
We confirmed the alteration of SRE in early AD that had previously been reported by Genon et al. [10], using a retrieval procedure similar to the one used in the present study, but with a different control condition (judgment of famous people, i.e., other condition). Several studies have reported preservation of the SRE in the early stages of AD, but methodological differences may explain at least some of the discrepancies between results. Kalenzaga et al. [13], for instance, reported better performances by patients with AD in the self condition, but only for emotional adjectives. Other studies reported that this effect was restricted to positive [9, 11] or negative [14] words. In the present study, we could not perform analyses involving the adjectives’ emotional valence as we did in a previous publication in MCI patients [11], owing to the limited number of correct answers per condition (often fewer than ten hits per valence in each condition). Nevertheless, this suggests that further studies are needed to pinpoint the conditions that allow the SRE to emerge in the early stages of AD.
From a functional standpoint, the self relies on a neural network that has been extensively described in the literature, composed of several cortical midline structures, including the MPFC, ACC and PCC [3, 4]. This network was activated in healthy older adults and in the patients with MCI/AD. Thus, despite the patients’ SRE impairment, the core structures on which self-processing relies appeared to be relatively spared. This was confirmed by the group comparison, which failed to show any difference between the two groups in terms of activity in cortical midline structures, in line with previous studies [10, 15]. In particular, there was no difference between patients and controls on neural activity in the MPFC, which is regarded as the core region for the self. This result stands at odds with previous fMRI studies showing that MPFC engagement during self-referential judgments is not significant in patients with AD [17, 37]. This discrepancy may be explained by methodological issues. In these studies, activity during self-judgment was compared with activity during the judgment of another person (close relative or caregiver), whereas our comparison was with a semantic condition. This choice was guided by results from healthy volunteers showing that the neural networks activated for judgments with reference to the self or another person are very close. For example, Denny et al. performed a meta-analysis of 108 neuroimaging studies of self- and other-related judgments and found that both conditions relied on the same network, including the ventral and dorsal parts of the MPFC, the left temporoparietal junction, and the PCC. In another meta-analysis, Araujo et al. [4] also suggested that the involvement of cortical midline structures is not specific to the judgment of trait adjectives with reference to the self, but is also involved when judging another person’s traits.
By contrast to the cortical midline structures, the angular gyrus was found to be less activated in patients than in controls during SRP. Further analyses revealed that the functional deficit in this region was due to impaired activity during self-reference judgments, but not during semantic judgments. Although the angular gyrus is not one of the core regions of the self, it has been associated with self-related processes. Two articles [3, 39] found that there is significant activity in this region when adjectives are judged with reference to the self. Seghier et al. [39] reported that the angular gyrus is involved in a variety of cognitive functions, including attention, semantic processing and memory retrieval. Binder et al. [40] proposed that the angular gyrus probably plays a role in the integration of complex information and knowledge retrieval, in which case patients exhibiting reduced activity in this region would not have access to this high-order processing hub, leading to impairment of SRP. For their part, Ciaramelli et al. [41] and Vilberg and Rugg [42] reported an activation of the angular gyrus during episodic memory retrieval, particularly during successful recollection (see also [43, 44]). These results are consistent with the positive correlation we observed in patients with MCI/AD between activity in the angular gyrus during SRP and recognition accuracy in the self condition. Thus, patients exhibiting high activity in the angular gyrus were more accurate in their recognition of self-encoded adjectives. In other words, insufficient activation of the angular gyrus may lead to inaccurate recognition of items encoded with reference to the self.
Gutchess et al. [45] observed greater activity in this region but in the contralateral hemisphere (MNI coordinates: [34 – 52 38]) in healthy older versus young adults when they successfully encoded self-reference items (in comparison with another condition). A higher level of activity in this region in older adults could be seen as a compensatory mechanism or, as proposed by the authors, the reflection of a different way of processing items in the self condition. In order to understand these results better, we performed additional analyses to determine whether the angular gyrus was also affected by brain atrophy, Aβ burden or hypometabolism, the three hallmarks of AD. The angular gyrus was not more atrophied in patients than in controls, but exhibited significant hypometabolism and more Aβ deposits. In addition, both hypometabolism and Aβ burden correlated with recognition accuracy in the self condition acrossparticipants.
We also investigated whether brain activity during recognition differs between items processed in reference to the self or in a semantic condition. In the first analysis, we focused on the brain activity associated with the successful recognition versus CR of personality trait adjectives, regardless of condition (“hits - CR” contrast). This contrast revealed activity in lateral parietal and frontal areas, as well as in the PCC, for both groups, in line with previous studies of successful episodic memory retrieval [46, 47].
Comparison of successful retrieval of adjectives processed in the self versus semantic condition did not reveal any significant activation in either group (even at a more lenient threshold of p < 0.01). This absence of specific brain activity during recognition had already been reported in two studies using the same task as ours, one in young participants [31], the other in AD patients [10]. These data suggest that the SRE is subserved by neural processes that predominantly occur during encoding. Moreover, the absence of activity during recognition in the self and semantic conditions in controls suggests that the link between the SRE and episodic memory may be more complex than it appears. The SRE may not only form part of the memory system, but may also rely on specific self-processing. Indeed, Sui and Humphreys [48] reported the case of a patient with both episodic and semantic impairments, but a preserved SRE. This result, together with our data, highlights the complexity of the relationship between self and episodic memory, self being a cognitive function interacting with episodic memory but independent of this memory system.
Our study had several limitations. The first one concerns the neurovascular coupling inherent to fMRI imaging. In this technique, neural activity is measured indirectly, through the blood-oxygen-level-dependent signal, which depends on neurovascular coupling. However, this neurovascularization probably undergoes changes during aging, especially in patients with neurodegenerative diseases [49, 50]. Then again, this drawback may have had only a limited impact, given that our comparison involved older adults versus patients matched for age. Another limitation concerns the small size of our patient group, which affected the statistical power of our analyses. Indeed, fMRI results were reported at uncorrected statistical thresholds that do not protect against false positive. Further studies with larger groups of patients are therefore needed to confirm our results. Given that the number of items in each condition during encoding and retrieval can also influence statistical power, we decided that we would only include in our analyses participants who provided at least eight answers in every condition, and in both the judgment and retrieval phases.
To conclude, our study highlighted an impaired SRE in AD patients that was related to dysfunction of the angular gyrus, but not of the cortical midline structures, whose activity was preserved. The investigation of the interaction between self and memory needs to be taken still further, in order to gain a better grasp of the changes to the self in pathologies such as AD. For instance, future studies could also take the nature of the items into consideration, notably the valence of the information being processed, as a few studies have reported an SRE specifically for negative or positive items [9, 51]. Moreover, the relevance of the items (i.e., whether or not an adjective really describes the participant) during a self-reference task should also be regarded as an important factor to consider.
Footnotes
ACKNOWLEDGMENTS
The authors would like to thank M. Fouquet, R. La Joie, K. Mevel, H. Mirabel, A. Pélerin, J. Dayan, A. Quillard, C. Schupp, C. Lebouleux, M.H. Noël and M.C. Onfroy for their help with the data acquisition, and all the volunteers who took part in this study. The authors are grateful to F. Degeilh and N. Morel for their help various stages of this study and their invaluable comments and support. Finally, the authors would also like to Elizabeth Portier-Willes for reviewing the English style.
This work was supported by the Fondation Plan Alzheimer (Alzheimer Plan 2008-2012), Programme Hospitalier de Recherche Clinique (PHRC National 2011), Agence Nationale de la Recherche (ANR LONGVIE 2007), Région Basse Normandie, and Institut National de la Santé et de la Recherche Médicale (Inserm).
