Subjects
A total of sixty-four subjects were recruited from community health screenings. All subjects were right-handed, aged between 50 and 80 years, and underwent a three-step inclusion process (detailed in the Supplemental Material S1). Nine subjects were excluded due to excessive motion artifacts (i.e., during the fMRI scan, head motion exceeded either 2 mm of the maximum displacement in any direction or 2° of angular motion) or incomplete data acquisition. The remaining 55 subjects, comprised of 26 aMCI and 29 healthy control (HC) subjects, were included in the final analysis. This study was approved by the Human Participants Ethics Committee of the Affiliated ZhongDa Hospital, Southeast University and Henan Mental Hospital, the Second Affiliated Hospital of Xinxiang Medical University, China. Written informed consent was obtained from all participants.
Clinical Evaluation
Global cognitive function was assessed using the Mini-Mental State Examination (MMSE) and the Mattis Dementia Rating Scale-2 (MDRS-2). Moreover, a neuropsychological battery was used, which covered EM, visuospatial function, information processing speed, and executive function. Details of the neuropsychological tests are provided in Supplemental Material S1.
Inclusion Criteria and Exclusion Criteria
The inclusion criteria for aMCI subjects were as follows: 1) subjective memory complaints; 2) objective memory impairment measured at 1.5 standard deviations (SD) below the age-adjusted norms in the Auditory Verbal Learning Test 20-minute delayed recall score; 3) minimal or absent impairments in global cognition or daily activities: MMSE score ≥ 24, MDRS-2 score ≥ 120, and activities of daily living score ≤ 25, and 4) the absence of dementia. All HC subjects presented MMSE scores ≥ 26.
In this study, the following exclusion criteria were applied: 1) any history of neurological or psychiatric diseases; 2) major medical illness or severe visual or hearing loss that interfered with cognitive evaluation; 3) contraindications to the MRI scan, and 4) gross brain structural abnormality as revealed by MRI.
Evaluation Procedure and Stimuli
The episodic retrieval paradigm included a study phase (40 trials, approximately 3.1 minutes), followed by a retrieval phase (80 trials, approximately 7.5 minutes). These two phases were separated by a distracter task in which the subjects were asked to silently repeat the numbers “one, two, three” for approximately two minutes (Figure 1A). In the study phase, 40 different nouns, consisting of two Chinese characters were presented, one at a time. These involved nouns that were used with high frequency in speech as defined by the Frequency Dictionary of Modern Chinese. In the retrieval phase, the 40 studied, as well as the 40 unstudied nouns, were presented in a pseudo-randomized order. No significant difference was observed between the frequencies of the studied and unstudied words presented (t78 = -0.075, p = 0.940). The stimuli were displayed in a white color against a black background using the E-Prime 2.0 software (Psychology Software Tools, Pittsburgh, PA, USA). All stimuli were presented in the central vision via MR-compatible goggles (Resonance Technology, Inc., Northridge, CA, USA). Example trials are illustrated in Figure 1B and in Supplemental Material S2.
Simultaneous EEG-fMRI Data Acquisition
In this study, fMRI data were acquired using a Siemens Verio 3.0 Tesla scanner (Siemens, Erlangen, Germany) with a standard 12-channel head coil. A gradient echo planar imaging sequence was used with the following parameters: 34 axial slices, repetition time (TR) = 2700 ms, echo time (TE) = 30 ms, flip angle = 90°, field of view (FoV) = 220 × 220 mm, matrix = 96 × 96, and thickness = 3.4 mm.
EEG data were acquired with an MR-compatible EEG amplifier (BrainAMP MR, Brain Products, Munich, Germany) at a sampling rate of 5,000 Hz, using 64 electrodes in the extended 10-20 montage, plus one extra EEG electrode. An experienced specialist (L.G.) positioned the EEG caps on the subjects. To ensure valid standard positions, the electrode Cz was placed halfway between the nasion and the inion, and was right-left-centered. The reference was set at the mid-frontal position FCz, and the impedances were kept below 10 kΩ. The data were transmitted via fiber optics outside the scanner room. To facilitate the removal of MR-induced artifacts from the EEG data, the sampling clocks of the EEG and MRI systems were synchronized by the Sync box (Brain Products, Munich, Germany).
Data Processing and Analysis
Behavioral Data Analysis
For behavioral data analysis, a total of 4 conditions were identified given each subject’s response in the retrieval trials, including hit (“old” response to an old word), miss (“new” response to an old word), correct rejection (CR, “new” response to a new word), and false alarm (FA, “old” response to a new word). Subsequently, a discrimination index, d-prime (d’), was calculated as the Z-score of the hit rate minus the Z-score of the false alarm rate. Independent two-sample t-tests were employed to compare the hit number, CR number, and the d’ values between HC and aMCI groups. For the demographic and neuropsychological data, independent two-sample t-tests and Chi-square tests were used to compare quantitative and qualitative variables, respectively. The significance threshold was set at p < 0.05.
fMRI Data Preprocessing and Analysis
In this study, fMRI data were preprocessed using the Analysis of Functional NeuroImages (AFNI) software (https://afni.nimh.nih.gov/afni). The preprocessing pipeline included despiking (3dDespike, AFNI), slice timing and motion correction (3dvolreg, AFNI), aligning functional data to structural data and spatial normalization to the Talairach space (3dAline, AFNI), and smoothness with a 6-mm Gaussian kernel (3dmerge, AFNI).
The BOLD functional response to the four conditions was estimated by the general linear model (GLM) (3dDeconvolve, AFNI). The design matrix contained three types of regressors. First, the hemodynamic response to the onset of each condition was modeled by the canonical hemodynamic response function with the first-order time derivative. Second, six motion parameters, including three translational and three rotational movements, and signals from cerebrospinal fluid (CSF) and white matter (WM), were included as regressors of no interest. Third, baseline detrending was conducted at a polynomial degree of 4, given the 496 seconds of scanning time. The contrast maps of hit versus baseline, CR versus baseline, and hit versus CR were created at individual subject levels. Then, the effects of hit, CR, and hit relative to CR were identified by one-sample t-tests in HC and aMCI groups, respectively. The differences in these effects were estimated by two-sample t-tests between groups (3dttest++, AFNI). For group-level analysis, a voxel-wise threshold of p < 0.005 with 15 contiguous voxels was used. Thus, the contrast map of hit versus CR represented the retrieval success pattern in this study.
EEG Data Preprocessing and ERP Analysis
EEG data were preprocessed using the BrainVision Analyzer software 2.0 (Brain Products GmbH, Munich, Germany), and the preprocessing pipeline included the following steps: 1) removal of MR-induced artifacts from the raw EEG signal with the All Scanned Intervals for Average approach (14); 2) down-sampling the EEG data to 250 Hz; 3) removal of cardioballistic artifacts by restricted infomax independent component analysis (ICA) based correction; 4) visually identifying and spherically interpolating noisy channels; 5) re-referencing to the average of all scalp electrodes; 6) bandpass filtering from 0.5 hertz (Hz) to 30 Hz; 7) ICA-based ocular correction; 8) inspecting raw data for the removal of waveforms with an amplitude greater than +100 µV or less than -100 µV; 9) editing markers to reject trails with incorrect responses, forming epochs from 100 ms of pre-stimulus to 1600 ms of post-stimulus for hit and CR trials, respectively; 10) correcting the baseline using the waveform before the stimulus onset, and 11) averaging waveforms across trials. The “stimulus” at steps 9 and 10 referred to the word shown on the screen but not the button-press response indicating an old or new word.
The components of interest ranged from 350 to 550 ms for the early old/new effect at electrode FCz and from 580 to 750 ms for the late old/new effect at electrodes P2 and P4, which was in accordance with the grand average ERP waveforms, and was consistent with a simultaneous EEG-fMRI study using the word-list retrieval task (15). To determine if the ERP component had a significant effect, repeated-measures analysis of variance (RMANOVA) were used with post hoc simple main effect analysis. Specifically, with regard to the ERP data at FCz, analysis was performed using the stimuli (hit versus CR) as the within-subject factors and the groups (NC versus aMCI) as the between-subject factors. With respect to the ERP data at P2 and P4, analysis was performed using the stimuli (hit versus CR) and electrodes (P2 versus P4) as the within-subject factors, and the groups (NC versus aMCI) as the between-subject factors. Statistical significance was set at p < 0.05. As presented in a previous study, no significant early old/new effects were observed among the frontal electrodes. The mean amplitude at electrode FCz was calculated to inform the fMRI analysis, which was consistent with the data presented in a previous study (15). Our data indicated significant late old/new effects at electrodes P2 and P4 among the parietal electrodes. Therefore, the mean amplitudes were used at the two electrodes to inform the fMRI analysis. The mean amplitudes of these components were extracted for the hit and CR conditions, which acted as amplitude modulators in the subsequent single-trial EEG-informed fMRI analysis. In addition, we calculated the signal-to-noise ratio (SNR) to evaluate the ERP signal quality. The method and result are detailed in Supplemental Material S3.
fMRI-constrained ERP Analysis
For each group, the fMRI-constrained ERP analysis was performed. In brief, source localization analysis was conducted to calculate the source waveforms across the whole brain using Brainstorm 3.0 software (http://neuroimage.usc.edu/brainstorm). The head model was computed using the OpenMEEG Boundary Element Method on the cortical surface of a standard 3D brain model. Subsequently, for each subject, the noise covariance matrix was calculated on the EEG baseline period (-100 ms to stimulus onset). The cortical current maps were computed from the averaged ERP time series using the weighted minimum norm estimate (wMNE) inverse solution for the hit and CR conditions, respectively. Furthermore, brain regions showing significant BOLD activation in the hit versus the CR contrast, which were informed by fMRI results, were used as the regional sources to obtain the source waveforms for each condition. The group-wise source waveforms for the hit and CR conditions (old/new effect) are presented in Figure 4.
Single-trial EEG-informed fMRI Analysis
Single-trial EEG-informed fMRI analysis was performed for the early and late old/new effects. In brief, mean amplitudes of the single-trial EEG data within a 350 to 550 ms time window at electrode FCz were extracted as the early old/new effect, and mean amplitudes of the single-trial EEG data at the P2 and P4 electrodes within the 580 to 750 ms time window were extracted as the late old/new effect. Next, extracted amplitudes from the hit and CR conditions were entered into the GLM as amplitude modulators (AM) with respect to hit and CR regressors. Specifically, the regressors in the GLM included: 1) the hemodynamic response to the onset of hit, miss, CR, and FA, as well as the AM of hit and CR; 2) six motion parameters, including three translational and three rotational movements, and CSF and WM signals, and 3) baseline detrending regressor with a polynomial degree of 4. Then, for each subject, the contrasts (hit AM versus baseline, CR AM versus baseline, and hit AM versus CR AM) were calculated. Finally, one-sample and two-sample t-tests were employed to calculate the statistical differences within each group as well as the differences between the groups.