Neurocognitive aging data release with behavioral, structural and multi-echo functional MRI measures

Central to understanding human behavior is a comprehensive mapping of brain-behavior relations within the context of lifespan development. Reproducible discoveries depend upon well-powered samples of reliable data. We provide to the scientific community two, 10-minute, multi-echo functional MRI (ME-fMRI) runs, and structural MRI (T1-MPRAGE), from 181 healthy younger (ages 18–34 y) and 120 older adults (ages 60–89 y). T2-FLAIR MRIs and behavioral assessments are available in a majority subset of over 250 participants. Behavioral assessments include fluid and crystallized cognition, self-reported measures of personality, and socioemotional functioning. Initial quality control and validation of these data is provided. This dataset will be of value to scientists interested in BOLD signal specifically isolated from ME-fMRI, individual differences in brain-behavioral associations, and cross-sectional aging effects in healthy adults. Demographic and behavioral data are available within the Open Science Framework project “Goal-Directed Cognition in Older and Younger Adults” (http://osf.io/yhzxe/), which will be augmented over time; neuroimaging data are available on OpenNeuro (https://openneuro.org/datasets/ds003592).

www.nature.com/scientificdatawww.nature.com/scientificdata/Magnetic resonance imaging.Neuroimaging data were acquired from two sites with a 3T GE Discovery MR750 and 32-channel head coil at the Cornell Magnetic Resonance Imaging Facility or on a 3T Siemens TimTrio MRI scanner with a 32-channel head coil at the York University Neuroimaging Center in Toronto.

T2-FLAIR.
A subset of 110 older adults and 148 younger adults have T2-FLAIR images.T2-weighted FLAIR sequences were acquired on a GE (TR = 12000 ms; TE = 95 ms; TI = 2712 ms; 160° flip angle; 42 slices of 1x1x3 mm; 2m36s) and Siemens (TR = 12000 ms; TE = 95 ms; TI = 2759.4ms; 160° flip angle; 44 slices of 0.8 × 0.8 × 3 mm; 3m38s).12 participants had FLAIR images with 46 slices acquired due to technician error, detailed in the README and participants.tsv.For 233 participants scanned on the GE, pulse and respiration were monitored continuously during scanning using an integrated pulse oximeter and respiratory belt.Note that due to a software upgrade, physiological sampling varies between participants at 50 Hz or 40 Hz, as indicated in the Data Records.

Data collection quality assurance and control.
A number of measures were taken to ensure reliable high quality behavioral and neuroimaging data collection.
In lab behavioral data were collected by a trained psychometrist.All behavioral data were then quality controlled before and after compilation.Paper and pencil measures were digitized by two researchers to ensure accuracy.Online data collection included multiple attention checks.Participants with incorrect responses to 4 or more checks were excluded.Participants who answered 3 or more different questionnaires uniformly (i.e., without variation in the Likert ratings) were excluded completely, along with participants missing more than 15% of total behavioral data.Participants missing more than 15% of any given measure were not included in any analysis with that measure.
Prior to undergoing MRI scanning, all participants were informed about the importance of staying still during the MR scan.All scans at the Cornell Magnetic Resonance Imaging Facility and York University Neuroimaging Center were performed by a trained MR technician working with a standardized protocol.This ensured consistent data acquisition procedures, including visual checks for coverage, ongoing quality assessment, and confirmation of participant wakefulness between runs.Five participants were initially excluded after visual inspection of anatomical scans revealed anomalies.One younger participant had a hyperintensity in the posterior lateral ventricle.One younger and four older adults had extended basal ganglia lesions of unknown etiology resembling perivascular spaces 34,35 .As the cognitive consequences of perivascular spaces in cognitively healthy adults remain unknown (e.g. 36,37), these participants were excluded from our sample.Image quality assessment was then performed on each functional run following preprocessing to exclude participants with unsuccessful coregistration, residual noise (framewise displacement > 0.50 mm coupled with denoised data with DVARS > 1 38 ), a temporal signal-to-noise ratio <50, or an insufficient amount of data retained after denoising (<10 BOLD components; see BOLD dimensionality below).Only participants with two functional runs that met these criteria were included in our final sample.9 younger and 24 older adults were excluded on this basis.
Quality metrics on the final sample are discussed below.

Data Records
All demographic information, in addition to cognitive, behavioral and personality variables, is available within the Open Science Framework project "Goal-Directed

T1-MPRAGE.
Cortical reconstruction and volumetric segmentation was performed with FreeSurfer version 6.0.1 41,42 .All participant surfaces had a Euler number of 2, indicating that no holes or defects were detected across the entire sample.Estimated total intracranial volume (eTIV), grey matter, white matter, hippocampus, BA45, V1, and MT volumes were extracted.These regions were selected for divergent susceptibilities to age-related volume reductions.Heteromodal cortices (prefrontal BA 45) and hippocampus characteristically show significant age-related volume losses, while volumes in unimodal cortices (V1, MT) are comparatively preserved into older age ( [43][44][45][46][47] and see 8 , for a review).Regional volumes were adjusted for head size by using the residuals of a linear regression between each volume, as output by FreeSurfer, and eTIV [48][49][50] .Adjusted volumes for each hemisphere were summed to yield a single adjusted volume for each region.Estimated whole brain  www.nature.com/scientificdatawww.nature.com/scientificdata/volume (eWBV) was also calculated as (grey matter + white matter)/eTIV.A series of ANCOVAs were then conducted to test for age group differences on volume with site, gender, education, and eWBV (regional volumes only) as covariates.Inclusion of eWBV provides additional estimation of specificity, particularly given global atrophy that occurs with aging (e.g. 51,).Education was not recorded for 14 young adult participants.
Volume distributions are plotted by age group in Fig. 1.Younger and older adults had comparable head sizes (Fig. 1A; F(1, 282) = 0.08, p = 0.784, η p 2 = 0.00), but younger adults had higher grey matter volume (Fig. 1B; F(1,282) = 165.58,p < 0.001, η p 2 = 0.37).T2-FLAIR.T2-FLAIR sequences were used to evaluate white matter hyperintensities (WMH) in 105 healthy older adults (57% female; M age = 68.35;age range = 60-83 y).WMH were segmented by the lesion prediction algorithm (Schmidt, 2017, Chapter 6.1) as implemented in the Lesion Segmentation Toolbox version 2.0.15 (www.statistical-modelling.de/lst.html)for Statistical Parametric Mapping.Each participant's raw total lesion volume in cubic millimetres (mm 3 ) was then divided by their eTIV derived from the T1 image in mm 3 to correct for head size.Final total lesion volume and number of lesions data were converted to within-sample z-scores for subsequent analysis.We characterized white matter lesion load, indexed by total lesion volume and number of lesions, and examined the validity of these indices to confirm that estimates of white matter lesion load in our sample demonstrated patterns previously established in the literature 43,[52][53][54][55]

Resting-state ME-fMRI.
To assess fMRI scan quality, quality metrics were calculated for each scan.

Framewise Displacement (FD).
A measure of the frame-to-frame movement, assessed in millimetres.FD was calculated on the second echo image for each resting-state scan using weighted scaling 38 .In younger adults, the average FD was 0.10 mm (SD = 0.05); in older adults, the average FD was 0.13 mm (SD = 0.05 mm).
Temporal Signal to Noise Ratio (tSNR).A measure of signal strength at the voxel level, calculated as the mean signal intensity of a voxel across the timeseries divided by its standard deviation.tSNR was calculated on each run of ME-ICA denoised data in native space (see 17 for a comparison of single-echo and multi-echo).Following Kundu and colleagues 56 , tSNR was quantified within the conjunction of grey matter and functional masks.Skull-stripped anatomical images were resampled to functional resolution and segmented with FSL FAST to create the grey matter mask.AFNI 3dAutomask was applied to the functional data to create the functional mask.
The median of all voxels within this mask is used to characterize each run of resting-state fMRI, where higher tSNR values reflect clearer signal.Median tSNR values for younger adults ranged from 140.10-361.37,and   www.nature.com/scientificdatawww.nature.com/scientificdata/for older adults from 129.59-418.90.For visualization purposes, tSNR spatial maps were separately derived in standard MNI space across the whole brain.Maps were averaged across all participants, thresholded at 50, and plotted in Fig. 3A.The results clearly demonstrate high tSNR throughout the cortical mantle in both cohorts.BOLD dimensionality.A unique advantage of ME-fMRI and the ME-ICA processing framework is that BOLD and non-BOLD signals can be separated into independent components.A novel metric of "BOLD dimensionality" 58 , or the number of BOLD components identified in the ME-fMRI timeseries, may then be examined.We assessed test-retest reliability of BOLD dimensionality across two runs of data and compared BOLD dimensionality between younger and older adults.BOLD dimensionality was stable across resting-state fMRI runs (r(299) = 0.79, p < 0.001, [0.75, 0.83]; Fig. 3B  Connectomics.Whole-brain interregional functional connectivity was computed and compared between younger and older adults.Group mean connectivity matrices are in Fig. 3C.Age-related differences in the 79800 interregional connections (i.e., the lower triangle of the 400 × 400 functional connectivity matrix 59 ; individualized with Group Prior Individualized Parcellations 60 ) were quantitatively assessed with Partial Least Squares 61,62 .A significant latent variable (permuted p = 0.001) revealed a pattern of age differences in RSFC, with increases and decreases observed across the connectome (Fig. 3D).See Setton and Mwilambwe-Tshilobo et al. 7 for in depth assessment.

Fig. 1
Fig. 1 Structural MRI MPRAGE technical validation.(A) Estimated total intracranial volume (eTIV) in younger and older adults.(B) Grey matter volume in younger and older adults.(C) Hippocampal, lateral prefrontal (BA45), primary visual cortex (V1) and Motion Complex (MT) volumes in younger and older adults.Regional volumes were adjusted for eTIV.* indicates significant age group differences as determined by ANCOVAs controlling for site, gender, education, and eWBV (regional volumes only).

Fig. 2
Fig. 2 Technical validation of FLAIR images and white matter hyperintensities (A) Total lesion volume is associated with number of lesions.Increasing age is associated with (B) Total lesion volume, and, (C) Number of lesions.Fluid-IQ is negatively associated with (D) Total lesion volume, and, (E) Number of lesions.Total lesion volume is corrected for intracranial volume.

Fig. 3
Fig. 3 Functional MRI technical validation.(A) Temporal signal-to-noise map across the full sample.(B) BOLD signal dimensionality across runs (left) and in younger and older adults.(C) Resting-state functional connectivity for younger (left) and older (middle) adults.(D) Age-related differences in connectivity between younger and older adults.Red color indicates significantly greater connectivity in younger adults, and blue color indicates significantly greater connectivity in older adults.(E) Resting-state functional connectivity across runs (sample mean edge-weights).VIS = visual, SOM = somatomotor, DAN = dorsal attention, VAN = ventral attention, LIM = limbic, FPC = frontoparietal control, DN = default, RSFC = resting-state functional connectivity.

Table 1 .
Sample All participants completed two 10m06s resting-state multi-echo BOLD functional scans.Participants were instructed to keep their eyes open, blinking and breathing normally in the dimly lit scanner bay.Resting-state runs were acquired using a multi-echo (ME) EPI sequence on GE (TR = 3000 ms; TE 1 = 13.7 ms, TE 2 = 30 ms, TE 3 = 47 ms; 83° flip angle; matrix size = 72 × 72; field of view (FOV) = 210 mm; 46 axial slices; 3 mm isotropic voxels; 204 volumes, 2.5x acceleration with sensitivity encoding) and Siemens (TR = 3000 ms; TE 1 = 14 ms, TE 2 = 29.96ms, TE 3 = 45.92 ms; 83° flip angle; matrix size = 64 × 64; FOV = 216 mm; 43 axial slices; 3.4 × 3.4 × 3mm voxels; 200 volumes, 3x acceleration and GRAPPA encoding) scanners.One participant (sub-149) had 206 volumes collected instead of 204: This discrepancy is detailed in the README and participants.tsvfile on OpenNeuro Demographics.Note: Episodic Memory, Semantic Memory, and Executive Function are index scores.Processing Speed is a z-score on Symbol Digit Modalities Task, Oral.* significant group differences.Education was not recorded for 14 participants.Age group differences in episodic memory, semantic memory, executive function, and processing speed were tested in 283 participants.Positive T values reflect higher scores in younger adults, negative values reflect higher scores in older adults.Statistical results were nearly identical when including sex, education, site, and estimated whole brain volume as covariates in an ANCOVA.www.nature.com/scientificdatawww.nature.com/scientificdata/Resting-state ME-fMRI.

technical Validation Cognitive, Behavioral and Personality assessment.
Location: sub-<ID> /ses-1/anat/sub-<ID>_ses-1_FLAIR.nii.gzComposite scores of episodic memory, semantic memory, and executive function were created from cognitive measures.Missing cells were first imputed with age group means (36 younger adults and 42 older adults had at least one cell missing).Latency scores on the Trail Making Task were reversed so that higher values on all measures reflected better performance.Scores were then z-scored and averaged for each composite.Episodic memory included scores on Verbal Paired Associates, Associative Recall, NIH Cognition Rey Auditory Verbal Learning, and NIH Cognition Picture Sequence Memory; Semantic memory included scores on Shipley Vocabulary, NIH Cognition Picture Vocabulary, and NIH Cognition Oral Reading Recognition; Executive function included scores on the Trail Making Task, NIH Cognition Flanker Inhibitory Control and Attention, NIH Cognition Dimensional Change Card Sort, and NIH Cognition List Sort Working Memory.Processing speed was additionally measured with the Symbol Digits Modalities Task.Descriptive statistics for all cognitive measures are shown in Table 1.Overall, younger adults had higher episodic memory (t(281) = 17.51, p < 0.001, [1.10, 1.38], Cohen's d = 2.11), executive function (t(281) = 12.67, p < 0.001, [0.71, 0.97], Cohen's d = 1.52), and processing speed (t(281) = 15.03,p < 0.001, [1.17, 1.53], Cohen's d = 1.81) scores than older adults.Older adults had higher semantic memory scores (t(281) = 9.18, p < 0.001, [−1.00, −0.65], Cohen's d = 1.10) than younger adults.These results remained when controlling for site, gender, and education.Table 2 contains descriptive statistics for all individual measures included in the index scores.Table 3 contains descriptive statistics for all self-report measures.

Table 2 .
Descriptive Statistics for Cognitive Measures by Age Group (N = 283).Note.NIH Cognition scores are unadjusted.One-way ANCOVAs were conducted on each measure to test for age group differences with site, gender, and education as covariates.

Table 3 .
Descriptive Statistics for Self-Report Measures by Age Group (N = 253).Note.NIH Emotion scores reflect T-scores.One-way ANCOVAs were conducted on each measure to test for age group differences with site, gender, and education as covariates.BIS/BAS = Behavioral Inhibition System/Behavioral Activation System.