Content uploaded by Raluca Petrican
Author content
All content in this area was uploaded by Raluca Petrican on Feb 14, 2020
Content may be subject to copyright.
Copyright © 2020 Petrican et al.
This is an open-access article distributed under the terms of the Creative Commons Attribution
4.0 International license, which permits unrestricted use, distribution and reproduction in any
medium provided that the original work is properly attributed.
Research Article: Confirmation | Cognition and Behavior
The Neural Dynamics of Individual Differences
in Episodic Autobiographical Memory
https://doi.org/10.1523/ENEURO.0531-19.2020
Cite as: eNeuro 2020; 10.1523/ENEURO.0531-19.2020
Received: 16 December 2019
Revised: 30 January 2020
Accepted: 31 January 2020
This Early Release article has been peer-reviewed and accepted, but has not been through
the composition and copyediting processes. The final version may differ slightly in style or
formatting and will contain links to any extended data.
Alerts: Sign up at www.eneuro.org/alerts to receive customized email alerts when the fully
formatted version of this article is published.
Episodic Memory and Spontaneous Neural Dynamics 1
1
Running head: EPISODIC MEMORY AND SPONTANEOUS NEURAL DYNAMICS 1
2
The Neural Dynamics of Individual Differences in Episodic Autobiographical Memory 3
Raluca Petrican, Daniela J. Palombo, Signy Sheldon and Brian Levine 4
Raluca Petrican, School of Psychology, Cardiff University, CF10 3AT; Daniela Palombo, 5
Department of Psychology, University of British Columbia, V6T 1Z4; Signy Sheldon, 6
Department of Psychology, McGill University, H3A 1G1; Brian Levine, Rotman Research 7
Institute and Departments of Psychology and Neurology, University of Toronto, M6A 2E1. 8
Corresponding authors: petricanr@cardiff.ac.uk; blevine@research.baycrest.org. 9
Number of pages: 55 10
Number of tables: 3 11
Number of figures: 7 12
Abstract: 250 words 13
Introduction: 739 words 14
Discussion: 2,345 words 15
16
Conflict of interest. None declared. 17
Acknowledgments. Data were partially provided by the Human Connectome Project, WU-Minn 18
Consortium (Principal Investigators: David Van Essen and Kamil Ugurbil; 1U54MH091657) 19
funded by the 16 NIH Institutes and Centers that support the NIH Blueprint for Neuroscience 20
Research; and by the McDonnell Center for Systems Neuroscience at Washington University. 21
Episodic Memory and Spontaneous Neural Dynamics 2
2
22
Abstract 23
The ability to mentally travel to specific events from one’s past, dubbed episodic 24
autobiographical memory (E-AM), contributes to adaptive functioning. Nonetheless, the 25
mechanisms underlying its typical inter-individual variation remain poorly understood. To 26
address this issue, we capitalize on existing evidence that successful performance on E-AM tasks 27
draws on the ability to visualize past episodes and reinstate their unique spatiotemporal context. 28
Hence, here, we test whether features of the brain’s functional architecture relevant to perceptual 29
versus conceptual processes shape individual differences in both self-rated E-AM and 30
laboratory-based episodic memory for random visual scene sequences (visual EM). We propose 31
that superior subjective E-AM and visual EM are associated with greater similarity in static 32
neural organization patterns, potentially indicating greater efficiency in switching, between rest 33
and mental states relevant to encoding perceptual information. Complementarily, we postulate 34
that impoverished subjective E-AM and visual EM are linked to dynamic brain organization 35
patterns implying a predisposition towards semanticizing novel perceptual information. Analyses 36
were conducted on resting state and task-based fMRI data from 329 participants (160 women) in 37
the Human Connectome Project who completed visual and verbal EM assessments, and an 38
independent gender diverse sample (N = 59) who self-rated their E-AM. Inter-individual 39
differences in subjective E-AM were linked to the same neural mechanisms underlying visual, 40
but not verbal, EM, in general agreement with the hypothesized static and dynamic brain 41
organization patterns. Our results suggest that higher E-AM entails more efficient processing of 42
temporally extended information sequences, whereas lower E-AM entails more efficient 43
semantic or gist-based processing. 44
Episodic Memory and Spontaneous Neural Dynamics 3
3
Keywords: Episodic Memory, Autobiographical Memory, Functional Networks, Dynamic 45
Connectivity 46
Episodic Memory and Spontaneous Neural Dynamics 4
4
47
Significance Statement 48
The ability to revisit specific events from one’s past is key to identity formation and optimal 49
interpersonal functioning. Nonetheless, the mechanisms underlying its typical inter-individual 50
variation are yet to be fully characterized. Here, we provide novel evidence that, among younger 51
adults, dispositional variations in subjective mental time travel draw on the same dynamic and 52
static features of the brain’s architecture that are uniquely implicated in memory for 53
spatiotemporal contexts. Specifically, the subjective sense of being able to revisit one’s past 54
relates to neural mechanisms supporting serial mental operations, whereas difficulties in 55
accessing past experiences may be traced back to a predisposition towards gist-based processing 56
of incoming information.57
58
59
Episodic Memory and Spontaneous Neural Dynamics 5
5
60
The Neural Dynamics of Individual Differences in Episodic Autobiographical Memory 61
Mental travel to specific past personal events is key to adaptive lifespan development 62
(Fivush, 2011). The contribution of visual imagery to such episodic autobiographical memory 63
(E-AM) feats is well-documented (D’Argembeau & Van, 2006; Daselaar et al., 2008; Greenberg, 64
Eacott, Brechin, & Rubin, 2005; Vannucci, Pelagatti, Chiorri, & Mazzoni, 2016). Nonetheless, a 65
link is yet to be established between dispositional variations in E-AM ability (i.e., ability to 66
recollect majority of previously experienced events) and individual differences in the capacity to 67
visually reconstruct past spatiotemporal contexts in one’s mind. To take a step towards 68
addressing this issue, here, we test whether individual differences in self-rated E-AM draw on 69
some of the neural architecture that supports memory for unique spatiotemporal contexts, 70
specifically, memory for random visual scene sequences (henceforth referred to as visual EM). 71
We thus investigate whether features of the brain’s static and dynamic functional architecture 72
relevant to conceptual versus perceptual processing similarly predict individual differences in 73
visual EM and subjective E-AM. 74
Current literature suggests that conceptual/meaning extraction processes foster episodic 75
memory (EM) formation (Griffiths et al., 2019; Renoult, Irish, Moscovitch, & Rugg, 2019; 76
Staresina & Wimber, 2019), although there is little research on their relative contribution as a 77
function of task demands and individual differences in E-AM. For example, semantic processes 78
may facilitate episodic memory for words, specifically, memory for whether a word was 79
presented in a given context or not (henceforth referred to as verbal EM) because their present 80
contextual correlates can be integrated within an existing knowledge base. In contrast, semantic 81
processes are less likely to foster recall of random visual scene sequences (i.e., visual EM) 82
Episodic Memory and Spontaneous Neural Dynamics 6
6
because these are less readily mapped onto an existing conceptual template. In the latter scenario, 83
access to high fidelity perceptual representations may be required. Put differently, greater 84
reliance on conceptual as opposed to perceptual features at encoding will be disadvantageous on 85
visual, but not verbal, EM tasks. 86
We propose that a similar relationship between conceptual and perceptual processes may 87
underlie individual differences in subjective E-AM. Our conjecture is based on theory that 88
distinguishes the state of awareness linked to retrieving conceptual information versus re-89
experiencing the unique spatiotemporal context of specific past events (noetic versus autonoetic 90
consciousness, Tulving, 1985, 2002). Thus, we argue that individuals who claim superior E-AM 91
are those who show relatively weaker reliance on conceptual, relative to perceptual, processes 92
when encoding new information (including personal events), meaning that they form and 93
subsequently retrieve a perceptually rich memory trace which incorporates relatively few 94
conceptual “sequelae” (i.e., at retrieval, they tend to enter into a state of autonoetic, rather than 95
noetic, consciousness, Tulving, 2002; Gurguryan & Sheldon, 2019). 96
Using network analysis of functional brain imaging data, we tested two hypotheses 97
focused on how the brain’s stable and time-varying functional architecture relevant to perceptual 98
versus conceptual processes may impact individual differences in visual EM and subjective E-99
AM. First, with respect to stable brain architecture, we examine whether visual EM and 100
subjective E-AM are linked to neural patterns indicative of more efficient perceptual, but less 101
efficient conceptual, processing (i.e., increased [for perceptual]/ reduced [for conceptual] 102
similarity in neural connectivity patterns between rest and the respective task, cf. Gold et al., 103
2013; Heinzel et al., 2014; Neubauer & Fink, 2009; Petrican & Levine, 2018; Schultz & Cole, 104
2016). 105
Episodic Memory and Spontaneous Neural Dynamics 7
7
Second, with respect to dynamic organization, we probe whether superior visual EM and 106
subjective E-AM are both associated with fewer spontaneous (i.e., resting state) transitions from 107
perceptual to meaning extraction mental states. We view a higher number of such transitions as 108
indicating a predisposition towards mapping novel perceptual information onto existing 109
conceptual templates (i.e., preferential reliance on semantic features when encoding new 110
information), which, as argued above, would interfere with both visual EM and subjective E-111
AM. Our argument is based on proposals that resting state architecture reflects behavioral history 112
and on recent demonstrations of the correspondence in temporal structure between resting state 113
and task-evoked neural dynamics (Farooq, Sibille, Liu, & Dragoi, 2019; Wig, Schlaggar & 114
Petersen, 2011). 115
This report is organized as follows. Part 1 focuses on a sample of healthy adults from the 116
Human Connectome Project (HCP) with the goal of testing our proposal regarding the role of 117
perceptual versus conceptual processes in visual versus verbal EM. Part 2 focuses on a separate 118
sample of healthy adults who self-rated their E-AM. Its purpose is to determine whether the 119
functional brain organization patterns uniquely linked to visual EM in Part 1 predict dispositional 120
variations in self-reported E-AM. 121
Part 1: HCP Sample 122
Method 123
Participants. This sample included 329 unrelated participants, whose data had been 124
released as part of the HCP 1200 subjects data package in March 2017. This sample represented 125
the largest number of participants from the HCP 1200 subjects data release who were unrelated 126
to one another and who had available data on all the behavioral and fMRI assessments of 127
interest. 128
Episodic Memory and Spontaneous Neural Dynamics 8
8
The majority of participants (N = 296) were right-handed. The sample included 169 129
younger men (51 between 22 and 25, 69 between 26 and 30, and 49 between 31 and 36 years of 130
age) and 160 younger women (50 between 22 and 25, 48 between 26 and 30, and 62 between 31 131
and 36 years of age). Although age is presented here in the range format, as advocated by the 132
HCP team (see Van Essen et al., 2012 for the rationale behind this age reporting strategy in HCP 133
data releases), all our brain-behavior analyses used participants’ actual age in years, as available 134
in the HCP restricted data release. 135
All participants were screened for a history of neurological and psychiatric conditions 136
and use of psychotropic drugs, as well as for physical conditions or bodily implants that may 137
render their participation unsafe. Diagnosis with a mental health disorder and structural 138
abnormalities, as revealed by the MRI structural scans, were also exclusion criteria. Participants 139
provided informed consent in accordance with the HCP research ethics board. 140
Episodic Memory (EM): Behavioral measures. A visual scene EM task assessed 141
individual differences in the ability to recollect the unique temporal flow associated with 142
perceptually rich information (i.e., temporally ordered visual scenes). A verbal EM task gauged 143
dispositional variations in the ability to recall information likely to draw on the existing 144
knowledge base (words). Performance on the two EM tasks was significantly positively 145
correlated, r(327) = .28, p = .0001. Only the population-normed scores were available for the 146
visual EM task. Nonetheless, scores on the visual and verbal EM tasks showed comparable 147
coefficients of variation (.082 [verbal EM] vs. .115 [visual EM], rendering it unlikely that the 148
observed results were due to restricted range in the verbal EM scores. 149
Visual scene sequences. The NIH Toolbox Picture Sequence Memory Test, completed 150
on Day 1 of the participants' HCP schedule, was used to assess EM for temporally ordered visual 151
Episodic Memory and Spontaneous Neural Dynamics 9
9
scenes (Barch et al., 2013). Participants were required to recall increasingly lengthier series of 152
illustrated objects and activities presented in a specific order on a computer screen. Sequence 153
length varied from 6 to 18 pictures. Participants were given credit for each pair of adjacent 154
pictures correctly recalled up to the maximum value for each sequence, which was one less than 155
sequence length. 156
Verbal. Form A of the Penn Word Memory Test (Gur et al., 2001, 2010), a non-NIH 157
Toolbox measure, was completed on Day 1 of the participants' HCP schedule and was used to 158
measure participants’ verbal EM abilities. Participants were presented with 20 words and asked 159
to memorize them for a subsequent test. On the recall trials, they were shown the 20 previously 160
learned words together with 20 new words matched on memory-related characteristics. 161
Participants had to decide whether they had previously seen the word by selecting among the 162
following response options: “definitely yes”, “probably yes”, “probably no” and “definitely no”. 163
fMRI tasks. The tasks described below were selected with an eye towards ensuring a 164
representative repertoire of spontaneous neurocognitive states likely to be observed during rest. 165
We reasoned that this sampling strategy would help us identify our hypothesis-relevant 166
neurocognitive states with greater accuracy (e.g., in a comparison involving only the perceptual 167
and semantic processing conditions, a predominantly motor state could be mis-classified as 168
reflecting perceptual processing just because of its greater similarity with the perceptual, rather 169
than the semantic, processing state). This is why we included tasks not directly linked to our 170
hypotheses (e.g., the motor task), but which captured mental states highly likely to occur in the 171
scanner (i.e., mental states relevant to body movement). 172
Episodic Memory and Spontaneous Neural Dynamics 10
10
Perceptual processing and online maintenance. As a measure of their ability to process 173
perceptual information and temporarily keep in their minds mental representations based on rich 174
percepts, participants completed two runs of an n-back task, which included as targets four 175
categories of stimuli: faces, places, tools and body parts. In the present report, we focused on the 176
zero-back condition, in which a stimulus was presented at the beginning of each block and the 177
participants had to respond "target" whenever the respective stimulus was encountered during the 178
block. We considered this task condition, which is analogous to a delayed match to sample 179
procedure, to best exemplify basic processing of perceptual information, including the creation 180
of the relevant mental images. Each run of the zero-back task encompassed 4 task blocks (27.5 s 181
each), with each comprising all four stimulus categories, presented in separate blocks. Each 182
block began with the 2.5 s presentation of a cue indicating task type and, for the 0-back task 183
only, target stimulus, followed by 10 trials of 2.5 s each (2 s stimulus presentation and 500 ms 184
interstimulus interval) for a total block duration of 27.5 s (see Barch et al., 2013). 185
The two-back condition, which assesses both perceptual processing and updating of 186
online mental contents, was not included in the present report because preliminary classifier 187
analyses revealed that, in individual-to-group mappings of task architecture, the two-back 188
condition could not be reliably differentiated from the zero-back condition. Instead, the serial 189
math task, described below, was employed as a measure of the participants’ ability to manipulate 190
online mental contents. 191
Meaning extraction and manipulation of online mental representations. Brief fables 192
were employed to assess meaning extraction from rich narrative information. Temporally 193
extended manipulation of mental representations was assessed with a math task (serial arithmetic 194
operations). Participants thus completed two runs of a task, adapted from Binder et al. (2011), in 195
Episodic Memory and Spontaneous Neural Dynamics 11
11
which aural presentation of brief stories alternates with aural presentation of math problems. On 196
each run, participants are presented with four story and four math blocks, which are matched in 197
duration. On the story blocks, participants are presented with short adaptations of Aesop’s fables 198
(5-9 sentences), which involve animal and human characters interacting in easily understandable 199
social situations. Subsequently, participants are required to answer a two-alternative forced 200
choice question, which tests their understanding of the story topic. On the math blocks, 201
participants are asked to solve serial addition and subtraction problems. Each series of arithmetic 202
operations (e.g., “Four plus twelve minus two plus nine”) ends with the word “equals”, followed 203
by two alternatives (e.g., “thirty-two or twenty-three”). The math task is adapted on an individual 204
basis, so that a similar level of difficulty is maintained across subjects. The story task was 205
designed to tap participants’ ability to extract meaning from incoming perceptual information 206
(i.e., aurally presented stories) (Binder et al., 2011). Complementarily, the math task was 207
designed to gauge the participants’ ability to engage in similarly effortful, temporally extended 208
cognitive processes (i.e., aurally presented arithmetic operations), which, however, do not 209
involve meaning extraction processes (Binder et al., 2011). The aforementioned dissociation 210
between the cognitive processes hypothesized to be underlying performance on the story versus 211
the math task is supported by their associated brain activation patterns, as reported in the initial 212
study (Binder et al., 2011) and in the HCP sample (Barch et al., 2013). 213
Motor processing. This task was included in order to account for mental states relevant to 214
actual and/or planned/desired movement, which was expected to occur in the scanner, including 215
movement pertaining to ongoing mind wandering. It was adapted from the one developed by 216
Buckner and colleagues (Buckner et al., 2011; Yeo et al., 2011). In response to visual cues, 217
participants are required to tap their left or right fingers, squeeze their left or right toes, or move 218
Episodic Memory and Spontaneous Neural Dynamics 12
12
their tongue. Each block, corresponding to a movement type, lasts 12 s (10 movements) and is 219
preceded by a 3 s cue. In each of the two task runs, there are two tongue, four finger (two left, 220
two right) and four toe (two left, two right) movement blocks, respectively, as well as three 15 s 221
fixation blocks. 222
fMRI data acquisition. Images were acquired with a customized Siemens 3T 223
“Connectome Skyra” scanner housed at Washington University in St. Louis (32-channel coil). 224
Pulse and respiration were measured during scanning. T1-weighted anatomical scans were 225
acquired with a 3D MP-RAGE sequence (TR = 2400 ms, TE = 2.14 ms, FOV = 224 mm, 320 x 226
320 matrix, 256 slices of 0.7 mm isotropic voxels). The high-resolution structural scan preceded 227
the acquisition of functional scans. 228
Functional images were acquired with a multiband EPI sequence (TR=720 ms, TE=33.1 229
ms, flip angle=52°, FOV = 208 mm, 104 × 90 matrix, 72 slices of 2 × 2 mm in-plane resolution, 230
2 mm thick, no gap). For each task, two runs of equal duration were obtained, one collected with 231
a L-R, and the other, with a R-L, EPI phase coding sequence. For rest, four different scans were 232
acquired in two different sessions (two collected with a L-R and two collected with R-L EPI 233
phase coding sequence). In the present study, we used the L-R and R-L resting state scans 234
collected from both sessions (i.e., four runs in total). The length of one run (in minutes) was as 235
follows: 14:33 (rest), 5:01 (perceptual processing), 3:57 (story/math), and 3:34 (motor). Details 236
on the duration of each resting state epoch and task condition, used in the connectivity analyses, 237
are included in the section on fMRI data analysis. 238
Individual L-R and R-L scans exhibit distinct regions of complete signal loss, but it has 239
been verified that the preprocessed datasets are anatomically well-aligned with one another, even 240
in areas of complete signal loss (cf. Smith et al., 2013). Because it is only the dropout that differs 241
Episodic Memory and Spontaneous Neural Dynamics 13
13
between the two scan types, it has been recommended that connectivity analyses based on HCP 242
data aggregate the respective metrics from the L-R and R-L scans (cf. Smith et al., 2013). 243
Consequently, in the present report, we concatenated the L-R and R-L runs for rest and each task 244
(see section on fMRI data analysis for further details on the concatenation of the resting state 245
scans). 246
fMRI data preprocessing. A schematic representation of our preprocessing pipeline is 247
depicted in Figure 1. In short, the present report used the preprocessed rest and task (i.e., 248
perceptual processing, story/math, and motor processing) data from the HCP 1200 subjects data 249
release. These data all have been preprocessed with version 3 of the HCP spatial and temporal 250
pipelines (Smith et al., 2013; for specification of preprocessing pipeline version, see 251
http://www.humanconnectome.org/data). Spatial preprocessing involved removal of spatial and 252
gradient distortions, correction for participant movement, bias field removal, spatial 253
normalization to the standard Montreal Neurological Institute (MNI)-152 template (2 mm 254
isotropic voxels), intensity normalization to a global mean and masking out of non-brain voxels. 255
Subsequent temporal preprocessing steps involved weak high-pass temporal filtering with the 256
goal of removing linear trends in the data. 257
Task fMRI data: Regression of condition effects. Our goal was to isolate task-related 258
functional coupling from mere co-activation effects corresponding to the beginning and end of a 259
task block (i.e., two regions that are both activated at the beginning of a task block and de-260
activated at the end of a task block, although they do not “communicate” with one another 261
throughout the task block). Consequently, following existing guidelines in the literature, we 262
regressed out condition effects from each task block by applying to the BOLD timeseries of each 263
ROI a regressor, obtained by convolving a boxcar task design function with the hemodynamic 264
Episodic Memory and Spontaneous Neural Dynamics 14
14
response function, and its first temporal order derivative (cf. Braun et al., 2015; Vatansever et al., 265
2015; Westphal, Wang, & Rissman, 2017; Whitfield-Gabrieli & Nieto-Castanon, 2012). The 266
regression of the condition effects was implemented by using the Denoising step in the CONN 267
toolbox (see paragraph below for additional regressors implemented in this step). All the task-268
related functional brain organization analyses were conducted only on the task blocks (i.e., the 269
between-task block rest periods were eliminated from the analyses). 270
Task and resting state fMRI data. Because motion can significantly impact functional 271
connectivity measures (Power et al., 2012; Van Dijk et al., 2012), we implemented several 272
additional preprocessing steps to address this potential confound in both the task and resting state 273
data (Figure 1-b). In line with prior studies that compared functional brain organization in the 274
task and resting state HCP data (Bolt, Nomi, Rubinov, & Uddin, 2017), these denoising steps 275
were identical for the task and rest data. First, after extracting the BOLD time series from our 276
regions-of-interest (ROIs, see below), but prior to computing the ROI-to-ROI correlations, we 277
used the Denoising step in the CONN toolbox (version 17c; Whitfield-Gabrieli & Nieto-278
Castanon, 2012) to apply further physiological and rigid motion corrections. Specifically, linear 279
regression was used to remove from the BOLD time series of each ROI the BOLD time series of 280
the voxels within the MNI-152 white matter and CSF masks, respectively (i.e., the default 281
CONN option of five CompCor-extracted principal components for each, Behzadi, Restom, 282
Liau, & Liu, 2007), the 6 realignment parameters, their first-order temporal derivatives and their 283
associated quadratic terms (24 regressors in total, cf. Bolt et al., 2017). For the task fMRI data 284
only, regression of the task effects was applied to the ROI timeseries corresponding to each task 285
block (see preceding section for details). The residual BOLD time series for both task and rest 286
were bandpass filtered (0.008 Hz< f < 0.09 Hz), linearly detrended and despiked (all three are 287
Episodic Memory and Spontaneous Neural Dynamics 15
15
default CONN denoising steps). Following these corrections (which did not include global signal 288
regression), an inspection of each subject’s histogram of voxel-to-voxel connectivity values for 289
each scrutinized condition (rest, task) revealed a normal distribution, approximately centered 290
around zero, which would suggest reduced contamination from physiological and motion-related 291
confounds (cf. Whietfield-Gabrieli & Castanon, 2012). Nonetheless, in supplementary analyses, 292
accompanying all the brain-behavior tests, we confirmed that all the reported effects were not 293
driven by individual differences in motion, as they remained unchanged after controlling for the 294
average relative (i.e., volume-to-volume) displacement per participant, a widely used motion 295
metric (Power et al., 2012, 2015; Satterthwaite et al., 2013). 296
fMRI data analysis. A schematic representation of our analysis pipeline is depicted in 297
Figure 2. The specific steps are detailed below. 298
ROI time series. 229 nodes for 10 networks (i.e., default [DMN], frontoparietal [FPC], 299
cingulo-opercular [CON], salience [SAL], dorsal attention [DAN], ventral attention [VAN], 300
somatomotor [SM], subcortical [SUB], auditory [AUD] and visual [VIS]) were defined for each 301
participant as spherical ROIs (radius 5 mm) centered on the coordinates of the regions reported 302
in Power et al. (2011) and assigned network labels corresponding to the graph analyses from this 303
earlier article. The Power et al. atlas was selected because it was created by taking into account 304
both the task-related activation (derived meta-analytically) and the resting state connectivity 305
patterns of the component voxels for each ROI. Thus, this atlas provided an optimal parcellation 306
scheme for comparing resting state and task-related functional brain architecture. 307
The ROIs were created in FSL (Smith et al., 2004), using its standard 2 mm isotropic 308
space, with each ROI containing 81 voxels. These template space dimensions were selected 309
because they yielded the most adequate spatial representation of the Power atlas. The 229 ROIs 310
Episodic Memory and Spontaneous Neural Dynamics 16
16
represent a subset of the 264 putative functional areas proposed by Power et al. (2011). The 229 311
ROIs were selected because, based on Power et al.’s analyses, they showed relatively 312
unambiguous membership to one of the ten functional networks outlined above. 313
For each participant, we used the CONN toolbox to compute pairwise bivariate 314
correlations among all 229 ROIs during each scrutinized condition (see Figure 2-a). Thus, for 315
each participant, we computed 1552 correlation matrices encompassing the pairwise correlations 316
among the 229 ROIs in each condition of interest: rest (1548 matrices, see below for details), 317
perceptual processing, story, math, and motor processing. For all analyses, the pairwise 318
correlations among all the ROIs were expressed as Fisher's z-scores. 319
Consistent with existing practices aimed at maximizing interpretability of results in 320
neural network studies of individual or group differences (e.g., sex or age, Betzel et al., 2014; 321
Satterthwaite et al., 2015), we used both positive and negative z-scores to compute the indices of 322
interest for all connectivity analyses. We reasoned that such an approach would be particularly 323
well-justified in our present case since global signal regression, an artefact removal technique 324
that generates negative correlations whose interpretation is still controversial, was not part of our 325
preprocessing pipeline (for further discussion on the validity of the negative correlations 326
obtained with the CONN toolbox, see Whitfield-Gabrieli & Nieto-Castanon, 2012). 327
Task-related connectivity analyses. Pairwise coupling among the 229 ROIs was 328
estimated in CONN, separately for each task condition. The task-relevant connectivity matrices 329
were based on durations ranging from 220 s (i.e., perceptual processing) to approximately 230 s 330
(i.e., motor processing). We used such durations because we sought to characterize the stable 331
core of the brain’s functional organization during the task modes under scrutiny (Telesford et al., 332
Episodic Memory and Spontaneous Neural Dynamics 17
17
2016), which we subsequently compared against the stable and transient aspects of the brain’s 333
functional organization observed during rest. 334
Resting state connectivity analyses. To characterize individual differences in stable and 335
dynamic network structure, we broke down the resting state runs into 1548 windows of 30 s 336
each. This window length was selected in light of prior evidence that it both maximizes detection 337
of individual differences in dynamic network reconfiguration and enables identification of a 338
stable functional core (Leonardi & Van De Ville, 2015; Preti, Bolton, de Ville, 2017; Telesford 339
et al., 2016; for similar window sizes in dynamic connectivity analyses of HCP data, see also 340
Chen et al., 2016). Thus, pairwise coupling among the 229 ROIs was estimated in CONN using a 341
sliding window of 30 s in length (~41 volumes) with a three-TR gap in-between windows and a 342
"hanning weighting" (i.e., greater weight to the scans in the middle of the window relative to the 343
ones at the periphery) applied to all the time points within a window. The use of a hanning 344
weighting was intended to reduce the autocorrelation in the fMRI data series and, thus, maximize 345
the opportunity to detect differences in functional brain organization between adjacent windows. 346
Each window was created so that it would contain only scans acquired with a LR or only scans 347
acquired with a RL encoding sequence. We thus opted to slide the window separately within the 348
LR and RL runs, respectively, in order to eliminate noise that could result from having a window 349
that contained a different proportion of LR and RL scans, which differ with respect to areas of 350
complete signal loss (cf. Smith et al., 2013). This issue was not applicable to the task data, which 351
contained a single window made of an equal number of LR and RL scans. 352
Network-level analyses. All the network-level metrics for both task and rest were 353
computed using the Brain Connectivity Toolbox (BCT, Rubinov & Sporns, 2010) and the 354
Episodic Memory and Spontaneous Neural Dynamics 18
18
Network Community Toolbox (Bassett, D.S. [2017, November]. Network Community Toolbox. 355
Retrieved from http://commdetect.weebly.com/), as described below (see Figure 2-b). 356
Community detection. Rather than being computed directly, the degree to which a 357
network can be fragmented into well-delineated and non-overlapping communities is estimated 358
using optimization algorithms, which sacrifice some degree of accuracy for processing speed 359
(Rubinov & Sporns, 2010). Here, for both task-related and resting state connectivity analyses, 360
the optimal whole-brain division into constituent communities was estimated using a Louvain 361
community detection algorithm implemented in the BCT. This algorithm partitions a network 362
into non-overlapping groups of nodes with the goal of maximizing an objective modularity Q 363
function (Rubinov & Sporns, 2011; Betzel & Bassett, 2017). There are multiple strategies for 364
estimating community structure based on sliding window data, as was the case of our resting 365
state data. Specifically, multilayer modularity algorithms (Bassett et al., 2011; Braun et al., 2015; 366
Mucha et al., 2010) can provide important insights into community dynamics at multiple time 367
scales. Nonetheless, such algorithms require estimation of additional free parameters (e.g., the 368
temporal coupling parameter between two adjacent temporal windows). Since we feared that 369
estimation of the temporal coupling parameter could act as a potential confound when comparing 370
task-related and resting state connectivity results, particularly given the multiple samples 371
included in the analysis, we used the same procedure to estimate community structure 372
independently in each task condition and each of the 1548 resting state time windows (see also 373
Chen et al., 2016), as described below. 374
For signed networks, such as the ones investigated in our study, optimization of the Q 375
function can be achieved by either placing equal weight on maximizing positive within-module 376
connections and minimizing negative within-module connections or by putting a premium on 377
Episodic Memory and Spontaneous Neural Dynamics 19
19
maximizing positive connections, which have been argued to be of greater biological 378
significance (Rubinov & Sporns, 2011). Although we verified that all the reported results emerge 379
with either formula, for the sake of simplicity and because we agree with their argument 380
regarding the greater importance of positive weights in determining node grouping into 381
communities, we report here the results based on Rubinov and Sporns’s modularity formula (cf. 382
Chen et al., 2016; Rubinov & Sporns, 2011). To account for the near degeneracy of the 383
modularity landscape (Good et al., 2010) and for changes in community structure due to 384
variations in the estimation parameters, for both task-related and resting state connectivity 385
analyses, the community detection algorithm was each initiated 100 times for three values of the 386
spatial resolution parameter, centered around the default value of 1 (cf. Betzel & Bassett, 2017; 387
Braun et al., 2015; Chen et al., 2016). 388
Based on the results of these analyses, run separately for each of the three spatial 389
resolution values, a consensus partition (i.e., whole-brain division into constituent communities) 390
was estimated for each participant in each task condition (cf. Bassett et al., 2013; Lancichinetti & 391
Fortunato, 2012). Based on the participant-specific consensus partitions, a group-level consensus 392
partition was estimated for each task mode under scrutiny (see Figure 3 for the consensus 393
partitions corresponding to each scrutinized task mode at the default value of the spatial 394
resolution parameter and Table 1 for indices of relative similarity in functional brain architecture 395
across the four tasks). For rest, we followed a similar procedure in two steps (cf. Braun et al., 396
2015). First, we derived a consensus partition for each time window and each participant. Each 397
participant’s window-specific consensus partitions were entered in the analyses involving EM-398
relevant neural process sequences. Across participants, the average similarity in functional brain 399
organization (i.e., AMI [from 0 no similarity to 1 the two partitions are identical], see section 400
Episodic Memory and Spontaneous Neural Dynamics 20
20
below) between two consecutive windows was .61 ± .01. Second, to identify a stable functional 401
core for each participant, we derived a full resting state consensus partition, corresponding to the 402
time scale of the initial sliding windows (~30 s). 403
Task-rest similarity in functional brain organization. Each individual’s whole brain 404
functional architecture during rest was compared against the whole-brain organization that 405
typified each scrutinized task mode at the group level. We opted for this approach because we 406
reasoned that it is whole-brain functional architecture, rather than specific ROI-to-ROI 407
connections, that best characterize specific mental states or task modes (i.e., the meaning of 408
specific ROI-to-ROI connections is likely dependent on the whole brain context in which they 409
occur). We thus used the Network Community Toolbox to compute for each participant two 410
types of similarity indices, based on the adjusted normalized mutual information index [AMI], 411
corrected for chance (Vinh, Epps, & Bailey, 2010). One index type gauged similarity between 412
each participant’s stable functional core, as assessed during rest, and the group-level consensus 413
clustering for each relevant task condition. Specifically, we created four indices reflecting 414
similarity between individual rest and the group-based perceptual processing organization (1), 415
between individual rest and group-based semantic processing organization (2), between 416
individual rest and the group-based math processing organization (3), and between individual 417
rest and group-based motor processing organization (4). 418
The second index type measured similarity between the group-level consensus clustering 419
for each relevant task condition and each participant’s window-specific functional brain 420
organization. Based on the highest task-rest similarity value, a specific window from a given 421
participant was classified as reflecting primarily one of the four task modes under scrutiny 422
(perceptual processing, semantic processing/math, motor processing). Individuals who spent 423
Episodic Memory and Spontaneous Neural Dynamics 21
21
more resting state windows in a given task mode showed greater similarity between the stable 424
core of their resting state architecture and the respective task mode (rs from .29 to .33, all ps < 425
.0001), but not the other task modes (all other rs < .04). 426
For each participant, we counted the number of times a participant expressed the 427
mnemonically relevant sequence (i.e., perceptual [window n]-semantic [window n+1] 428
processing) and its counterpart (i.e., semantic [window n] - perceptual [window n+1] 429
processing). Tables 2 and 3 contain summary statistics for the similarity indices between the 430
stable core of the brain’s intrinsic architecture and each of the four task modes, as well as 431
summary statistics for the number of windows spent in each task mode and the number of all 432
task mode switches across the full HCP sample. Figure 4 contains a histogram showing the 433
distribution of our core task switch variable, i.e., perceptual-to-semantic processing. 434
Reliability analyses. In line with prior studies on graph theoretical metrics derived from 435
task-based and resting state connectivity patterns (Braun et al., 2012; Cao et al., 2014), we used 436
the intra-class correlation coefficient, ICC (2, 1) (Shrout & Fleiss, 1979) to quantify the absolute 437
agreement among the graph theoretical metrics, corresponding to each value of the spatial 438
resolution parameter (see above), computed separately for the Day 1 and Day 2 resting state 439
sessions. Thus, for each neural index of interest, we entered in the reliability analysis six values, 440
corresponding to Day 1—spatial resolution parameter of .95, Day 1—spatial resolution 441
parameter of 1.00, Day 1—spatial resolution parameter of 1.05, Day 2—spatial resolution 442
parameter of .95, Day 2—spatial resolution parameter of 1.00, and Day 2—spatial resolution 443
parameter of 1.05. 444
In line with published criteria (Cicchetti & Sparrow, 1981), as well as prior reliability 445
data for graph metrics derived from task-based and resting state connectivity patterns (Braun et 446
Episodic Memory and Spontaneous Neural Dynamics 22
22
al., 2012; Cao et al., 2014), ICC values ≥ .40 were regarded as reflecting fair to good reliability. 447
Since subject motion can impact such reliability estimates, we present the relevant ICC values, 448
both before and after regressing out subject level average frame-to-frame displacement (please 449
see Preprocessing above for the additional motion effect removal procedures already 450
implemented). 451
Across the two days and three values of the spatial resolution parameter, the task-rest 452
similarity indices relevant to the stable functional core showed ICCs ranging from .60 to .65 (for 453
data metrics from which the subject-level summary motion metric had been regressed out). For 454
indices from which the subject-level summary motion had not been regressed out, the ICCs 455
ranged from .59 to .65. 456
Across the two days and three values of the spatial resolution parameter, number of 457
perceptual-to-semantic transitions showed an ICC of .48, irrespective of whether the subject-458
level summary motion had not been regressed out or not. The number of semantic-to-perceptual 459
transitions showed an ICC of .48 (ICC of .49, if the subject-level summary motion had not been 460
regressed out). 461
For all the brain-behavior analyses reported below, the stable functional core architecture 462
was estimated based on the resting state data from both days. However, in the reliability analyses 463
reported above, the stable functional core of the resting state architecture was estimated 464
separately for Day 1 versus Day 2 in order to verify our assumptions (i.e., that there is a person-465
specific functional core that can be derived from resting state data and that shows some stability 466
across days). The number of state transitions were summed across the two scanning days. In 467
order to obtain more stable estimates of the neural variables of interest, we averaged the 468
homologous indices corresponding to the three values of the spatial resolution parameter for both 469
Episodic Memory and Spontaneous Neural Dynamics 23
23
rest-task similarity metrics (i.e., those relevant to the stable functional core and dynamic neural 470
sequence expression). 471
Validation of the individual to group task-based functional architecture. As outlined 472
above, the analyses herein reported are based on the indices of similarity between group-level 473
task-related and individual-specific resting state functional brain architecture. We opted to do so 474
for two reasons. One was to maximize comparability with the analyses conducted on the SAM 475
sample for which relevant task data were unavailable. The second was to optimize classification 476
of a given resting state window as reflecting primarily one of the four task modes under scrutiny. 477
Specifically, preliminary analyses revealed that in intra-individual comparisons of resting state 478
and task architectures the neural organization within some resting state windows could be 479
equally similar to two or more task modes. Such a pattern of results could emerge even when a 480
window-specific functional brain organization shares the greatest similarity with the key 481
architectural features (i.e., those that are reproduced at the group level) associated with only one 482
task mode. Consequently, we based our analyses on the indices of similarity between the 483
individual-specific resting state architecture and the relevant group-derived task architectures, 484
which yielded unambiguous classification of a resting state window to one of the four task 485
modes under investigation. The group-derived functional task architectures were also used in the 486
comparisons involving the stable functional core within each sliding window. 487
To verify the validity of our approach, we tested the accuracy of our proposed AMI-488
based classifier in correctly linking an individual’s task architecture to the corresponding task 489
architecture of a group that did not include the respective individual (e.g., verify that an 490
individual’s functional architecture during story processing is most similar to the group-based 491
story processing functional architecture rather than the group-based perceptual, math or motor 492
Episodic Memory and Spontaneous Neural Dynamics 24
24
processing architectures). To this end, for each task type (perceptual, story, math and motor 493
processing, respectively), we used the ROI-to-ROI correlation matrices, corresponding to each 494
task block, to define through the graph theoretical analyses outlined above (1) a consensus 495
partition corresponding to each block within a given task type, then, based on (1), define (2) a 496
consensus partition characteristic of each task type, which generalizes across different stimulus 497
categories (i.e., the perceptual and motor processing tasks contained stimulus-specific blocks, see 498
Barch et al., 2013). The consensus partitions obtained at (2) reflected the brain organization 499
specific to each task type on time scales ranging from about 24 s (motor processing) to 27.5 500
(perceptual processing), hence similar to the time scale used in the resting state dynamic analyses 501
(i.e., 30 s windows). Subsequently, we used a leave-one-subject-out cross validation procedure in 502
which the task architectures of the left-out subject (based on less than 30 s of data) are evaluated 503
for how similar they are to the group-based stable task architectures derived from the remaining 504
328 individuals and based on the full task runs of 220 to 240 s. This procedure was repeated until 505
all participants served as the left-out (“test”) subject. Subsequently, for each individual, we 506
evaluated whether his or her architecture for a specific task condition (based on less than 30 s of 507
data) showed the greatest similarity to the corresponding group-based architecture (based on 220 508
to 240 s of data). Across the four task contexts and the three values of the spatial resolution 509
parameter, our AMI-based classifier had an average accuracy of 66% (±2.10 [standard error]) 510
and a positive predictive value (PPV) of 67% (±2.22 [standard error]). 511
Brain-behavior analyses 512
Canonical correlation analysis (CCA). To characterize the relationship of our neural 513
indices of interest with EM, we used canonical correlation analysis (CCA, Hotelling, 1936; see 514
Figure 2-c) with cross-validation procedures (cf. Hair et al., 2008). CCA is a multivariate 515
Episodic Memory and Spontaneous Neural Dynamics 25
25
technique, which seeks maximal correlations between two sets of variables by creating linear 516
combinations (i.e., canonical variates) from the variables within each set. Recently, CCA has 517
been successfully used to investigate brain-behavior relationships in large datasets (see Smith et 518
al., 2015; Tsvetanov et al., 2016; Vatansever et al., 2017). CCA was implemented in Matlab 519
using the canoncorr module. Task-rest similarity indices for math, perceptual, semantic and 520
motor processing, as well as the number of perceptual-to-semantic and semantic-to-perceptual 521
processing sequences were introduced as brain variables. Age, verbal and visual EM were 522
entered as behavioral variables. Age was introduced in the CCA outlined below because the 523
neural architecture underlying higher-order cognitive functions, including EM, shows protracted 524
development, which extends into the third decade of life (Lebel et al., 2012; Petrican & Grady, 525
2017; Toga, Thompson, & Sowell, 2006). In order to obtain reliable estimates of canonical 526
loadings (i.e., correlations between the brain or behavioral variables and their corresponding 527
variates), it is generally recommended that CCA be performed on a sample size at least ten times 528
the number of variables in the analysis (Hair, Anderson, Tatham, & Black, 1998), a criterion 529
which was exceeded in all analyses reported below. 530
The performance of our CCA-derived model of EM was tested by using a 10-fold cross 531
validation procedure. Specifically, the data were broken down into ten folds, all but one 532
containing 30 participants for a total of 329 participants. Discovery CCA was conducted on nine 533
folds of data and the resulting CCA weights were employed to derive predicted values of the 534
brain and behavioral variate in the left-out (“test”) fold. This procedure was repeated until each 535
of the ten folds served as “test” data once. The correlation between the predicted brain and 536
behavioral variates across all testing folds was evaluated using a permutation test with 100,000 537
samples (cf. Smith et al., 2015). To describe the relationship between the behavioral or brain 538
Episodic Memory and Spontaneous Neural Dynamics 26
26
variables and their corresponding variates across all the discovery CCAs, we include canonical 539
loadings (cf. Hair, Black, Babin, & Anderson, 2009), which reflect the raw correlation between a 540
brain or behavioral variable and its corresponding variate, as well as canonical weights, which 541
indicate the unique contribution of a behavioral or brain variable to its corresponding variate (see 542
also Tsvetanov et al., 2016; Vatansever et al., 2017). 543
Code Accessibility 544
The scripts for the graph theoretical analyses outlined above are available in Extended 545
Data 1. 546
Results 547
The discovery CCAs detected only one significant mode, which was validated across all 548
test sets (r= .20, p = .0001 , see Figure 5-a for loadings of each connectivity and cognitive 549
variable on its respective canonical variate across all discovery sets, Figure 5-b for standardized 550
coefficients of each connectivity and cognitive variable on its respective canonical variate across 551
all discovery sets and Figure 5-c for the relationship between the predicted brain and behavioral 552
canonical variates across all test sets). The mode identified indicated that younger individuals 553
with superior visual EM demonstrated reduced expression of the perceptual-to-semantic 554
processing sequence, as well as greater similarity between the stable core of the brain’s intrinsic 555
architecture and the functional architecture common to all scrutinized task contexts (see Figure 556
5-a), but particularly the math/mental manipulation context (i.e., after accounting for the 557
intercorrelations among the brain variables, the rest-math similarity index showed the strongest 558
association with the brain variate, see Figure 5-b). Next, we sought to verify that the association 559
between the brain and behavioral variate is not contaminated by demographic factors or 560
extraneous neural variables. To this end, we first created a residual brain variate by regressing 561
Episodic Memory and Spontaneous Neural Dynamics 27
27
out from the original brain variate the number of windows spent in each of the four scrutinized 562
task states and the number of switches between task states not included in the discovery CCA. 563
Subsequently, we conducted a partial correlation analysis, based on 100,000 permutation 564
samples, in which we verified that the association between the original behavioral variate and the 565
aforementioned residual brain variate remained significant (r of .15, p = .006) after controlling 566
for sex, handedness, years of education and average volume-to-volume displacement during rest. 567
Part 2: SAM Sample 568
Method 569
Participants. The SAM sample included fifty-nine unrelated, neurologically intact adults 570
(mean age: 23.34 ± 4.90 years [median = 22 years]; age range: 18-41 years, 15 males). The 571
majority of participants (N = 50) were right-handed. Data from this sample were also included in 572
Sheldon, Farb, Palombo and Levine (2016), but there is no overlap in the analyses documented 573
in the respective paper and the present report. 574
Self-reported memory capacity. Self-reported memory capacity at the trait level was 575
assessed with the 26-item Survey of Autobiographical Memory (SAM, Palombo, Williams, 576
Abdi, & Levine, 2013). SAM requires participants to rate their E-AM, semantic memory, future 577
thinking, and spatial memory, on a 5-point Likert scale (1 = strongly disagree to 5 = agree 578
strongly). In all the analyses herein reported, we used the weighted sum scores derived from the 579
episodic, future, semantic and spatial subscales described below (cf. Palombo et al., 2013). 580
SAM-Episodic (8 items). This subscale gauges participants’ ability to recall specific 581
event and contextual details (e.g., “When I remember events, in general I can recall people, what 582
Episodic Memory and Spontaneous Neural Dynamics 28
28
they looked like, or what they were wearing.”, “When I remember events, in general I can recall 583
objects that were in the environment.”). This subscale was regarded as a measure of trait E-AM. 584
SAM-Semantic (6 items). This subscale assesses trait-level differences in the 585
participants’ ability to recall factual information (“I can learn and repeat facts easily, even if I 586
don’t remember where I learned them.”, “After I have met someone once, I easily remember his 587
or her name.”). 588
SAM-Future (6 items). This subscale measures trait-level differences in the participants’ 589
ability to imagine specific event and contextual details pertaining to future occurrences (“When I 590
imagine an event in the future, the event generates vivid mental images that are specific in time 591
and place.”, “When I imagine an event in the future, I can imagine how I may feel.”). 592
SAM-Spatial (6 items). This subscale evaluates trait-level differences in the participants’ 593
spatial navigation skills (“In general, my ability to navigate is better than most of my 594
family/friends.”, “After I have visited an area, it is easy for me to find my way around the second 595
time I visit.”). 596
Intercorrelations among the SAM memory subscales. The episodic SAM subscale 597
showed a significant positive correlation with the semantic subscale (r of .38, p = .003) and a 598
trending positive association with the future subscale (r of .23, p = .074). No other correlations 599
reached statistical significance (all other ps > .40). 600
fMRI Data Acquisition. Images were acquired with a Siemens 3T Trio scanner housed 601
at the Rotman Research Institute (32-channel coil: 35 participants; 12-channel coil: 24 602
participants). Coil type was introduced as a covariate in all brain-behavior analyses reported 603
below. T1-weighted anatomical scans were acquired with a 3D MP-RAGE sequence (TR = 2000 604
Episodic Memory and Spontaneous Neural Dynamics 29
29
ms, TE = 2.63 ms, FOV = 256 mm, 256 x 256 matrix, 160 slices of 1 mm isotropic voxels). The 605
high-resolution structural scan preceded the acquisition of functional scans. 606
Functional images were acquired with a T2*-weighted EPI sequence (TR=2000 ms, 607
TE=32 ms, flip angle=70°, FOV = 200 mm, 64 × 64 matrix, 32 axial slices of 3.1 × 3.1 mm in-608
plane resolution, 4.5 mm thick, no gap). Acquisition of the resting state scan preceded 609
acquisition of the functional task scans, which are not discussed in this report. During their 610
resting state scan, which lasted approximately 6.5 min, participants were asked to allow their 611
minds to wander, while keeping their eyes open and focused on a black fixation cross presented 612
on a white background. 613
fMRI data preprocessing. We performed image processing in SPM12 (Wellcome 614
Department of Imaging Neuroscience, London, UK). Specifically, we corrected for slice timing 615
differences and rigid body motion (which included unwarping) and spatially normalized the 616
images to the standard Montreal Neurological Institute (MNI)-152 template (2 mm isotropic 617
voxels). 618
Because motion can significantly impact functional connectivity measures (Power et al., 619
2012; Van Dijk et al., 2012), we used the Denoising step in the CONN toolbox to implement 620
several additional preprocessing steps, which were also applied to the data from the HCP sample, 621
in order to address this potential confound (see step 2 in Figure 1). Following these corrections 622
(which did not include global signal regression), an inspection of each subject’s histogram of 623
voxel-to-voxel connectivity values revealed a normal distribution, approximately centered 624
around zero, which would suggest reduced contamination from physiological and motion-related 625
confounds (cf. Whietfield-Gabrieli & Castanon, 2012). Nonetheless, same as we did for the HCP 626
data, in supplementary analyses, accompanying all the brain-behavior tests, we confirmed that all 627
Episodic Memory and Spontaneous Neural Dynamics 30
30
the reported effects were not driven by individual differences in motion, as they remained 628
unchanged after controlling for the average relative (i.e., volume-to-volume) displacement per 629
participant, a widely used motion metric (Power et al., 2012, 2015; Satterthwaite et al., 2013). 630
fMRI data analysis. For all analyses, we followed the same steps as the ones outlined 631
for the HCP sample (see Figure 2, steps 1 and 2). Due to the duration of the resting state scan in 632
the SAM sample, all analyses were based on 166 sliding windows with each window being 633
moved in increments on one TR (i.e., 2 s, a duration similar to the one used in the HCP data [3 634
TRs = 2.16 s]). All other parameters were identical to the ones used with the HCP data. Across 635
participants and across the three values of the spatial resolution parameter, the average AMI 636
between consecutive windows was .63 ± .01. 637
Brain-behavior analyses. The goal of these analyses was to test the hypothesis that the 638
neural profile significantly linked to visual EM in the HCP sample would be linked to E-AM 639
abilities, but not the other mnemonic traits assessed by the SAM. To identify the brain variables 640
that make the most reliable contribution to the brain variate linked to visual EM, we conducted a 641
multiple regression analysis across the ten non-overlapping test samples from the HCP. As 642
outcome, we used the standardized value associated with the predicted brain variate score (as 643
derived from the discovery CCAs), from which we regressed out the observed values associated 644
with potential neural confounds (i.e., number of windows spent in each of the four task modes, 645
task mode switches beyond the semantic-to-perceptual and perceptual-to-semantic processing 646
sequences). As predictors, we used the standardized observed values associated with the brain 647
variables of interest (i.e., global similarity indices for perceptual, semantic, math and motor 648
processing, as well as the number of semantic-to-perceptual and perceptual-to-semantic 649
processing sequences). No outliers (i.e., values of 3.29 standard deviations above/below the 650
Episodic Memory and Spontaneous Neural Dynamics 31
31
sample means) were detected among any of these variables. To identify the variables that make a 651
reliable contribution to the aforementioned residual brain variate, we used the bootci function in 652
Matlab (with default settings and 100,000 bootstrap samples) to obtain 95% confidence intervals 653
(CI) for each predictor variable. Results of this analysis revealed that only the global perceptual 654
processing-rest similarity index (95% CI = [-.20; .0004]) and expression of the semantic-to-655
perceptual processing sequence (95% CI = [-.04; .17]) did not make reliable contributions to the 656
residual brain variate. 657
Hence, based on the results from the multiple regression analysis conducted in the HCP 658
sample (see above), expression of the semantic-to-perceptual processing sequence and similarity 659
between rest and perceptual processing were not included in the computation of the brain variate 660
(i.e., their respective weights were set to zero) in the SAM sample. Instead, these two variables 661
were covaried out from the brain variate because their unreliable contribution to the brain variate 662
was regarded as a potential source of noise, a concern that was rendered salient by the smaller 663
SAM sample (relative to the HCP sample). As in the HCP data, other neural confounds regressed 664
out from the HCP brain variate as well included the number of windows spent in each task mode 665
and number of switches beyond the perceptual-to-semantic processing switch. The resulting 666
residual brain variable was introduced in the correlational analysis described below. 667
Tables 2 and 3 contain summary statistics for the similarity indices between the stable 668
core of the brain’s intrinsic architecture and each of the four task modes, as well as summary 669
statistics for the number of windows spent in each task mode and the number of all possible task 670
mode switches across the full SAM sample. Figure 6 contains a histogram showing the 671
distribution of our core task switch variable, i.e., perceptual-to-semantic processing. 672
Results 673
Episodic Memory and Spontaneous Neural Dynamics 32
32
Results of a partial correlation analysis, based on 100,000 permutation samples, in which 674
we controlled for scores on the remaining three SAM subscales, as well as age, sex, education, 675
handedness, coil type and head motion, revealed the predicted positive association between 676
episodic SAM scores and expression of the brain pattern linked to visual EM in the HCP sample, 677
r of .26, p = .032 (see Figure 7). As expected, correlational analyses, based on 100,000 678
permutation samples, showed no similar associations between the neural organization patterns 679
linked to superior visual EM in the HCP sample and scores on the remaining SAM subscales (all 680
rs < .05, all ps > .53). Using an on-line calculator for comparing correlation coefficients drawn 681
from the same sample (https://www.psychometrica.de/correlation.html#dependent, Lenhard & 682
Lenhard, 2014), we confirmed that the brain pattern linked to visual EM in the HCP sample was 683
significantly more strongly correlated with the SAM Episodic scores than with the SAM 684
Semantic (z = 2.37, p =.009) or Spatial (z = 2..01, p =.022) scores. However, the visual EM-685
linked brain pattern appeared to be similarly linked to SAM Future and SAM Episodic (past) 686
scores (z = 1.346, p =.089), a finding that is compatible with the interpretation that the neural 687
profile herein identified may be broadly relevant to both prospective and retrospective episodic 688
thought. 689
Discussion 690
To date, most neuroscientific investigations on dispositional variations in E-AM have 691
focused on variations in the brain's structural architecture (Freton et al., 2014; Hebscher, Levine, 692
& Gilboa, 2018; Hodgetts et al. 2017; Palombo et al., 2018; but see Sheldon et al., 2016). The 693
present study draws on the reportedly key role of visualization in fostering E-AM (e.g., 694
D’Argembeau & Van, 2006; Daselaar et al., 2008; Greenberg et al., 2005; Vannucci et al., 2016) 695
to provide novel evidence that spontaneous neural dynamics linked to memory formation 696
Episodic Memory and Spontaneous Neural Dynamics 33
33
processes constitute a common mechanism underlying individual differences in visual EM and 697
subjective E-AM, but not other forms of subjective memory ability (spatial or semantic memory 698
skills). Specifically, we show suggestive evidence that superior performance-based memory for 699
unique spatiotemporal contexts and self-reported E-AM are linked to greater similarity in static 700
functional brain organization, potentially indicating greater efficiency in switching, between rest 701
and a range of goal-directed mental states, as well as a reduced predisposition towards 702
semanticizing perceptual information. 703
A core motivation of the present research was to investigate whether a competitive 704
relationship between perceptual and conceptual processes shapes individual differences in 705
subjective E-AM and visual EM. Evidence for our proposal was mixed. With respect to the 706
stable core of the brain’s functional architecture, we found some support for our hypothesis that 707
individuals with superior subjective E-AM, as well as those with superior visual EM would show 708
organization patterns suggestive of less efficient semantic processing. Specifically, there was 709
evidence of a reliable unique relationship between superior visual EM and reduced similarity in 710
functional brain organization between rest and the semantic processing mode (see Figure 5-b), 711
which was replicated with respect to subjective E-AM. In our opinion, these unique effects best 712
capture our hypotheses regarding semantic processing because the raw relationship between the 713
semantic processing variable and the identified brain variate (see Figure 5-a) is “contaminated” 714
by variance linked to the other brain variables, particularly, variance related to similarity in 715
functional organization between rest and a global task mode, common across all four scrutinized 716
tasks. 717
Of note, there was no reliable unique association between efficiency in perceptual 718
processing and visual EM (see Figure 5-b). The lack of a preferential association with visual, 719
Episodic Memory and Spontaneous Neural Dynamics 34
34
rather than verbal, EM is consistent with the interpretation that efficient perceptual processing is 720
core to EM, irrespective of domain and its susceptibility to conceptual contagion (i.e., it 721
contributes to both visual and verbal EM), a finding reported before (Petrican & Levine, 2018). 722
Clearer support was garnered for our hypothesis relevant to the brain’s dynamic 723
functional architecture (i.e., number of transitions from perceptual to conceptual mental states). 724
As predicted, we found that superior visual EM and subjective E-AM were both linked to less 725
frequent spontaneous transitions between mental states dominated by perceptual processing and 726
those that mainly reflect meaning extraction attempts. We argued that such spontaneous neural 727
dynamics could be interpreted as a predisposition towards mapping perception information onto 728
the relevant conceptual structures. The present results are thus compatible with our proposal that 729
the aforementioned predisposition would impair memory for unique spatiotemporal contexts, as 730
attempts to find a matching conceptual template may distort the mnemonic representation. They 731
are also in line with our hypothesis that a predisposition to semanticize perceptual 732
representations may hinder subjective perceptions of being able to revisit specific past events, 733
potentially because it infringes upon the capacity to enter a state of autonoetic consciousness at 734
retrieval (Tulving, 2002). 735
Beyond our hypotheses, we found that greater efficiency in switching (i.e., reduced 736
functional brain reorganization) from rest to task (see Figure 5-a) was a hallmark of both 737
superior visual EM and subjective E-AM. With respect to specific task modes, it appears that 738
efficiency in switching to a cognitive mode linked to temporally extended manipulation of 739
mental images (i.e., the math task mode) makes the strongest unique contribution (see Figure 5-740
b). Although unpredicted, this effect may reflect the role of serial mental operations in 741
supporting the encoding and reinstatement of spatio-temporal contexts, which are relevant to the 742
Episodic Memory and Spontaneous Neural Dynamics 35
35
presently used visual EM task (Barch et al., 2013) and are generally assumed to also support the 743
subjective sense of being able to revisit the past (Levine et al., 2002; Tulving, 2002; Wheeler & 744
Buckner, 2004). 745
Interestingly, this neural task mode evidenced the greatest segregation (i.e., the highest 746
number of communities, see Math in Figure 3). This pattern likely speaks to its greater 747
processing efficiency and resilience in the face of environmental stressors (Kashtan & Alon, 748
2005; Kashtan et al., 2007; Braun et al., 2015; Betzel et al., 2016; Sporns & Betzel, 2016). Its 749
associated community structure was compatible with neural organization patterns previously 750
linked to successful episodic learning, such as greater connectivity between the DMN and visual 751
systems, a pattern that is likely relevant to the creation of mental representations based on 752
perceptual information (see community 4 in Figure 3; cf. Sheldon et al., 2016). Complementing 753
the aforementioned unique community features of the math task mode, there are also 754
organizational characteristics, such as community 3 (see Figure 3), which show significant 755
commonalities across all the scrutinized task modes and may explain their shared contribution to 756
visual EM and subjective E-AM. Community 3, which brings together ROIs from the DMN, 757
VAN and SAL, is likely instrumental in the creation and manipulation of mental representations 758
based on environmentally driven attentional and control processes, dynamics that are key to 759
externally cued instances of mental time travel. 760
Our present findings regarding the neural dynamics correlates of E-AM complement 761
earlier research in an overlapping sample on the stable functional connectivity patterns that 762
distinguish high episodic from semantic SAM scorers (Sheldon et al., 2016). The respective 763
study documented that higher episodic SAM scorers demonstrate stronger intrinsic coupling 764
between medial temporal (MTL) regions and posterior regions implicated in visual perceptual 765
Episodic Memory and Spontaneous Neural Dynamics 36
36
processing. In contrast, higher semantic SAM scorers evidenced stronger functional connectivity 766
between the MTL and frontal regions implicated in categorization. These findings suggest that 767
higher episodic SAM scores may reflect a predisposition towards using visual imagery when 768
accessing the past, while greater semantic SAM scores may indicate a proficiency in organizing 769
information. Extending these findings, the present study provides evidence that the dynamic 770
neural patterns that typify reinstatement of unique spatiotemporal contexts are linked to self-771
reported E-AM, but not the other SAM subscales. Moreover, broadly consistent with the results 772
of Sheldon et al. that MTL-related functional connectivity patterns suggestive of greater 773
proficiency in categorization are associated with weaker episodic, relative to semantic, memory 774
skills, we show a link between lower self-reported E-AM and a predisposition towards reducing 775
discrete perceptually rich experiences to semanticized representations. 776
Our present findings lend support to the construct of trait mnemonics, whereby stable, 777
lifelong patterns of encoding information predispose towards engagement in specific mental 778
activities (Palombo, Sheldon, & Levine, 2018). Whereas high E-AM promotes rich visual re-779
experiencing of past events that are segregated in consciousness, lower E-AM may be associated 780
with more stable abstract and non-visual representations that generalize across experiences. 781
Accordingly, people with Highly Superior Autobiographical Memory (HSAM) have obsessive 782
tendencies that reflect an extreme focus on specific details (LePort, Stark, McGaugh, & Stark, 783
2016), whereas people with Severely Deficient Autobiographical Memory (SDAM) show intact 784
learning and daily functioning in spite of their impaired recollection (Palombo et al, 2015; see 785
also Greenberg & Knowlton, 2014). Beyond these extremes, such biases may yield paradoxical 786
effects, such that those with higher visual EM are more susceptible to visual interference 787
(Sheldon, Amaral, & Levine, 2016), whereas those with lower E-AM may become resilient to 788
Episodic Memory and Spontaneous Neural Dynamics 37
37
the effects of neurodegenerative disorders affecting EM through the development of cognitive 789
reserve (Fan, Romero, & Levine, 2019; Stern, 2003; Stern, Scarmeas, & Habeck, 2004). 790
Our present research focused on dispositional variations in the subjective sense that one 791
can revisit one’s past. Individual differences in self-rated E-AM abilities have been shown to be 792
meaningfully related to other cognitive-affective traits, as well as structural and functional brain 793
characteristics (e.g., Palombo et al., 2013, 2016; Sheldon et al., 2016). Such subjective E-AM 794
evaluations, as those assessed by Episodic SAM, are likely to tap distinct aspects of one’s 795
mnemonic experience compared to performance-based measures of E-AM, which assess the 796
quantity of past event fragments that one can recover. For one, differences between subjective 797
and objective E-AM measures may arise due to the fact that the holistic evaluation underlying 798
the former need not equal the sum of the parts indexed by the latter. Second, performance on 799
objective E-AM measures, which index the amount of retrieved event details, can be 800
contaminated by non-episodic memory processes. For example, one can recover details 801
pertaining to a specific event not through mental time travel, but through the repeated use of 802
external aids, such as photographs, diaries, conversations with close others (Cermak & 803
O’Connor, 1983; Rabin et al., 2013). Such alternate routes for retrieving autobiographical details 804
have been used to explain previously demonstrated dissociations between subjective and 805
objective E-AM performance (i.e., recovery of a relatively high amount of episodic details 806
without the accompanying subjective sense of having mentally travelled back to the respective 807
event, Levine et al., 2009). Third, most current measures of objective E-AM focus on the 808
retrieval of a relatively small number of past episodes, which is why performance on these tasks 809
is not necessarily indicative of the capacity to recollect majority of previously experienced 810
Episodic Memory and Spontaneous Neural Dynamics 38
38
events. In our opinion, the SAM-Episodic scale is a useful alternative measure for assessing such 811
stable individual differences in E-AM, albeit from a subjective standpoint. 812
Our study demonstrates that some of the brain mechanisms that distinguish EM for visual 813
scene sequences from EM for information with significant links to the semantic knowledge base 814
also feed one’s subjective sense of being able to revisit the past. Our findings thus imply that a 815
superior capacity to engage in goal-directed behavior, particularly, to manipulate one’s online 816
mental contents, and a reduced tendency to semanticize perceptually rich mental representations 817
are associated with both visual EM and subjective E-AM skills. Further research is required to 818
determine how this overlap in neurocognitive component processes may support the link 819
between subjective E-AM and visual EM. Evidence for such a link is garnered from recent 820
findings that individuals with higher subjectively related E-AM show a tighter coupling between 821
oculomotor behavior and objectively assessed E-AM (Armson et al., 2019). 822
Our present research has several limitations. First, future studies are needed to 823
examine whether spontaneous expression of the neural task modes and sequences, herein 824
investigated, is meaningfully linked to neural connectivity patterns observed during encoding 825
and retrieval of autobiographical memories, as well as with our proposed unfolding of mental 826
processes (e.g., mapping of novel perceptual information onto pre-existing knowledge 827
structures). Second, our SAM sample was primarily composed of younger women, which is why 828
our present results need to be replicated in samples with a balanced gender composition and 829
better lifespan coverage. Third, we used a self-report measure of E-AM because it captures best 830
the experiential aspects of E-AM (i.e., the subjective sense of being able to revisit specific past 831
episodes). As we argued in the Introduction, we propose that this subjective sense-of-self-in-the-832
past is an emerging property of the state of awareness that typifies retrieval of purely episodic 833
Episodic Memory and Spontaneous Neural Dynamics 39
39
details (i.e., autonoetic consciousness, Tulving, 2002). Our present results suggest that although 834
this mnemonic trait is based on self-report, its supporting brain network architecture is associated 835
with objective performance on EM tasks in a separate sample. Future studies combining 836
performance-based measures of episodic recall with subjective ratings of trait mnemonics in the 837
same sample would be pivotal in shedding further light on the neural mechanisms herein 838
documented. 839
Finally, to test our main hypotheses, we combined sliding window with graph theoretical 840
analyses of resting state data. The former have been the topic of some controversy. For example, 841
Laumann et al. (2017) provided evidence suggesting that most of the variability associated with 842
resting state connectivity can be accounted for by sampling variability, head motion and 843
sleepiness. Subsequently, it has been pointed out though that Laumann et al.’s findings are 844
amenable to alternative interpretations, specifically, some that do not exclude the possibility of 845
meaningful fluctuations in resting state connectivity patterns (for an in-depth discussion, see 846
Lurie et al., 2019). Others have also underscored the fact that Laumann et al.’s results are based 847
on relatively long sliding windows (100 s), which tend to be suboptimal for detecting individual 848
differences in dynamic reorganization patterns and, thus, cannot really speak to the validity of 849
resting state dynamics assessed with shorter time windows (Abrol et al., 2017). 850
That being said, we do agree that resting state connectivity, particularly when based on 851
shorter time windows, is vulnerable to the influence of confounding factors, including 852
physiological noise and rigid motion. This is why we implemented strict preprocessing 853
procedures for minimizing the impact of such factors (i.e., through CompCor, regression of the 854
24 motion parameters and their derivatives, use of the summary motion metric in the brain-855
behavior analyses). Of note, almost one hour of continuously acquired data went into the main 856
Episodic Memory and Spontaneous Neural Dynamics 40
40
resting state analyses in the HCP sample. The fact that our neural indices based on the sliding 857
window analyses showed reliability values as good as those previously reported for similar graph 858
metrics derived from stable task and resting state connectivity patterns (Braun et al., 2012; Cao 859
et al., 2014), as well as the conceptual replication of the brain-behavior relationships across two 860
independent samples give us confidence in our presently reported results. Nonetheless, future 861
studies using alternate measures of dynamic resting state connectivity would be instrumental in 862
shedding further light on the neural mechanisms herein documented. 863
In summary, we have provided evidence that individual differences in self-rated E-AM 864
draw on some of the brain mechanisms also implicated in memory for visual scene sequences. 865
These findings support the relationship between subjective mental time travel and visual imagery 866
(D’Argembeau & Van, 2006; Daselaar et al., 2008; Greenberg et al., 2005), specifically, raising 867
the possibility of an association between subjective E-AM and the ability to access temporally 868
ordered mental records of previous experiences. Complementarily, our results imply that 869
perceived difficulties in accessing the past may be traced back to a cognitive style that prioritizes 870
schematic, gist-based information over rich perceptual representations.871
Episodic Memory and Spontaneous Neural Dynamics 41
41
References 872
Abrol, A., Damaraju, E., Miller, R. L., Stephen, J. M., Claus, E. D., Mayer, A. R., & Calhoun, V. 873
D. (2017). Replicability of time-varying connectivity patterns in large resting state fMRI 874
samples. NeuroImage, 163, 160-176. 875
Andrews-Hanna, J.R., Smallwood, J.S., & Spreng, R.N. (2014). The default network and self-876
generated thought: Component processes, dynamic control, and clinical relevance. Annals 877
of New York Academy of Sciences - Year in Cognitive Neuroscience Special Issue, 1316, 878
29-52. 879
Armson, M.J., Diamond, N.B., Levesque, L., Ryan, J.D., & Levine, B. (2019). The relationship 880
between eye movements and autobiographical recollection is mediated by individual 881
differences in autobiographical capacity. https://psyarxiv.com/pztv4/ 882
Barch, D.M., Burgess, G.C., Harms, M.P., Petersen, S.E., Schlaggar, B.L., Corbetta, M., Glasser, 883
M.F., Curtiss, S., Dixit, S., Feldt, C., Nolan, D., Bryant, E., Hartley, T., Footer, O., Bjork, 884
J.M., Poldrack. R., Smith, S., Johansen-Berg, H., Snyder, A.Z., & Van Essen DC; for the 885
WU-Minn HCP Consortium (2013). Function in the human connectome: Task-fMRI and 886
individual differences in behavior. NeuroImage, 80, 169-189. 887
Bassett, D.S., Wymbs, N.F., Rombach, M.P., Porter, M.A., Mucha, P.J., & Grafton, S.T. 888
(2013). Task-based core-periphery organisation of human brain dynamics. PLoS 889
Computational Biology, 9, e1003171. 890
Bassett, D. S., Wymbs, N. F., Porter, M. A., Mucha, P. J., Carlson, J. M., & Grafton, S. T. 891
(2011). Dynamic reconfiguration of human brain networks during learning. Proceedings 892
of the National Academy of Sciences of the United States of America, 108, 7641-7646. 893
Betzel, R. F., & Bassett, D. S. (2017). Multi-scale brain networks. NeuroImage, 160, 73-83. 894
Episodic Memory and Spontaneous Neural Dynamics 42
42
Betzel, R. F., Byrge, L., He, Y., Goñi, J., Zuo, X., & Sporns, O. (2014). Changes in structural 895
and functional connectivity among resting-state networks across the human 896
lifespan. NeuroImage, 102, 345-357. 897
Betzel, R. F., Fukushima, M., He, Y., Zuo, X., & Sporns, O. (2016). Dynamic fluctuations 898
coincide with periods of high and low modularity in resting-state functional brain 899
networks. NeuroImage, 127, 287-297. 900
Binder, J.R., Gross, W.L., Allendorfer, J.B., Bonilha, L., Chapin, J., Edwards, J.C., Grabowski, 901
T.J., Langfitt, J.T., Loring, D.W., Lowe, M.J., Koenig, K., Morgan, P.S., Ojemann, J.G., 902
Rorden, C., Szaflarski, J.P., Tivarus, M.E., Weaver, K.E., 2011. Mapping anterior 903
temporal lobe language areas with fMRI: a multicenter normative study. NeuroImage, 54, 904
1465–1475. 905
Bolt, T., Nomi, J. S., Rubinov, M., & Uddin, L. Q. (2017). Correspondence between evoked and 906
intrinsic functional brain network configurations. Human Brain Mapping, 38, 1992-907
2007. 908
Braun, U., Schäfer, A., Walter, H., Erk, S., Romanczuk-Seiferth, N., Haddad, L., . . . Bassett, D. 909
S. (2015). Dynamic reconfiguration of frontal brain networks during executive cognition 910
in humans. PNAS Proceedings of the National Academy of Sciences of the United States 911
of America, 112, 11678-11683. 912
Braun, U., Plichta, M. M., Esslinger, C., Sauer, C., Haddad, L., Grimm, O., . . . Meyer-913
Lindenberg, A. (2012). Test–retest reliability of resting-state connectivity network 914
characteristics using fMRI and graph theoretical measures. NeuroImage, 59, 1404-1412. 915
Episodic Memory and Spontaneous Neural Dynamics 43
43
Buckner, R. L., Krienen, F. M., Castellanos, A., Diaz, J. C., & Yeo, B. T. T. (2011). The 916
organization of the human cerebellum estimated by intrinsic functional 917
connectivity. Journal of Neurophysiology, 106, 2322-2345. 918
Cao, H., Plichta, M. M., Schäfer, A., Haddad, L., Grimm, O., Schneider, M., . . . Tost, H. (2014). 919
Test–retest reliability of fMRI-based graph theoretical properties during working 920
memory, emotion processing, and resting state. NeuroImage, 84, 888-900. 921
Cermak, L. S., & O'Connor, M. (1983). The anterograde and retrograde retrieval ability of a 922
patient with amnesia due to encephalitis. Neuropsychologia, 21, 213-234. 923
Chen, T., Cai, W., Ryali, S., Supekar, K., & Menon, V. (2016). Distinct global brain dynamics 924
and spatiotemporal organization of the salience network. PLoS Biology, 14, 21. 925
Cicchetti, D. V., & Sparrow, S. A. (1981). Developing criteria for establishing interrater 926
reliability of specific items: Applications to assessment of adaptive behavior. American 927
Journal of Mental Deficiency, 86, 127-137. 928
Corbetta, M. & Shulman, G.L. (2002). Control of goal-directed and stimulus-driven attention in 929
the brain. Nature Reviews Neuroscience, 3, 201–215. 930
D'Argembeau, A., & Van, d. L. (2006). Individual differences in the phenomenology of mental 931
time travel: The effect of vivid visual imagery and emotion regulation 932
strategies. Consciousness and Cognition: An International Journal, 15, 342-350. 933
Daselaar, S. M., Rice, H. J., Greenberg, D. L., Cabeza, R., LaBar, K. S., & Rubin, D. C. (2008). 934
The spatiotemporal dynamics of autobiographical memory: Neural correlates of recall, 935
emotional intensity, and reliving. Cerebral Cortex, 18, 217-229. 936
Episodic Memory and Spontaneous Neural Dynamics 44
44
Fan, C., Romero, K., & Levine B. (2019). Older adults with lower autobiographical memory 937
abilities report less age-related decline in everyday cognitive function. 938
https://psyarxiv.com/zqs78/ 939
Farooq, U., Sibille, J., Liu, K., & Dragoi, G. (2019) Strengthened Temporal Coordination within 940
Pre-existing Sequential Cell Assemblies Supports Trajectory Replay, Neuron, 103, 719-941
733. 942
Fivush, R. (2011). The development of autobiographical memory. Annual Review of Psychology, 943
62, 559-582. 944
Fox, M. D., Snyder, A. Z., Vincent, J. L., Corbetta, M., Van Essen, D. C., & Raichle, M. E. 945
(2005). The human brain is intrinsically organized into dynamic, anticorrelated functional 946
networks. PNAS Proceedings of the National Academy of Sciences of the United States of 947
America, 102, 9673-9678. 948
Freton, M., Lemogne, C., Bergouignan, L., Delaveau, P., Lehéricy, S., & Fossati, P. (2014). The 949
eye of the self: Precuneus volume and visual perspective during autobiographical 950
memory retrieval. Brain Structure & Function, 219, 959-968. 951
Gold, B. T., Kim, C., Johnson, N. F., Kryscio, R. J., & Smith, C. D. (2013). Lifelong 952
bilingualism maintains neural efficiency for cognitive control in aging. The Journal of 953
Neuroscience, 33, 387-396. 954
Greenberg, D. L., Eacott, M. J., Brechin, D., & Rubin, D. C. (2005). Visual memory loss and 955
autobiographical amnesia: A case study. Neuropsychologia, 43, 1493-1502. 956
Greenberg, D.L., & Knowlton, B.J. (2014). The role of visual imagery in autobiographical 957
memory. Memory & Cognition, 42, 922-934. 958
Episodic Memory and Spontaneous Neural Dynamics 45
45
Griffiths, B., Parish, G., Roux, F., Michelmann, S., Staresina, B., Wimber, M., ... Hanslmayr, S. 959
(2019). Directional coupling of slow and fast hippocampal gamma with neocortical 960
alpha/beta oscillations in human episodic memory. Proceedings of the National Academy 961
of Sciences of the United States of America. https://doi.org/10.1101/305698 962
Gur, R.C., Ragland, J.D., Moberg, P.J., Bilker, W.B., Kohler, C., Siegel, S.J., Gur, R.E. (2001a). 963
Computerized neurocognitive scanning: II. The profile of schizophrenia. 964
Neuropsychopharmacology, 25, 777–788. 965
Gur, R.C., Ragland, J.D., Moberg, P.J., Turner, T.H., Bilker, W.B., Kohler, C., Siegel, S.J., & 966
Gur, R.E. (2001b). Computerized neurocognitive scanning: I. Methodology and 967
validationin healthy people. Neuropsychopharmacology, 25, 766–776. 968
Gur, R.C., Richard, J., Hughett, P., Calkins, M.E., Macy, L., Bilker, W.B., Brensinger, C., & 969
Gur, R.E. (2010). A cognitive neuroscience-based computerized battery for efficient 970
measurement of individual differences: standardization and initial construct validation. 971
Journal of Neuroscience Methods, 187, 254–262. 972
Gurguryan, L., & Sheldon, S. (2019). Retrieval orientation alters neural activity during 973
autobiographical memory recollection. NeuroImage, 199, 534-544. 974
Hair, J.F., Tatham, R.L., Anderson, R.E . & Black, W, (1998) Multivariate data analysis. (Fifth 975
Ed.) Prentice-Hall: London. 976
Hair Jr., J.F., Black, W.C., Babin, B.J. & Anderson, R.E. (2009) Multivariate Data Analysis. 977
(Seventh Ed.) Prentice Hall: Upper Saddle River, 761. 978
Hebscher, M., Levine, B., & Gilboa, A. (2018). The precuneus and hippocampus contribute to 979
individual differences in the unfolding of spatial representations during episodic 980
autobiographical memory. Neuropsychologia, 110, 123-133. 981
Episodic Memory and Spontaneous Neural Dynamics 46
46
Heinzel, S., Lorenz, R. C., Brockhaus, W., Wüstenberg, T., Kathmann, N., Heinz, A., & Rapp, 982
M. A. (2014). Working memory load-dependent brain response predicts behavioral 983
training gains in older adults. The Journal of Neuroscience, 34, 1224-1233. 984
Hodgetts, C. J., Postans, M., Warne, N., Varnava, A., Lawrence, A. D., & Graham, K. S. (2017). 985
Distinct contributions of the fornix and inferior longitudinal fasciculus to episodic and 986
semantic autobiographical memory. Cortex: A Journal Devoted to the Study of the 987
Nervous System and Behavior, 94, 1-14. 988
Kashtan, N. & Alon, U. (2005) Spontaneous evolution of modularity and network motifs. 989
Proceedings of the National Academy of Sciences, 102, 13773–13778. 990
Kashtan, N., Noor, E., & Alon, U. (2007) Varying environments can speed up evolution. 991
Proceedings of the National Academy of Sciences, 104, 13711–13716. 992
Laumann, T. O., Snyder, A. Z., Mitra, A., Gordon, E. M., Gratton, C., Adeyemo, B., . . . 993
Petersen, S. E. (2017). On the stability of BOLD fMRI correlations. Cerebral Cortex, 27, 994
4719-4732. 995
Lebel, C., Gee, M., Camicioli, R., Wieler, M., Martin, W., & Beaulieu, C. (2012). Diffusion 996
tensor imaging of white matter tract evolution over the lifespan. NeuroImage, 60, 340-997
352. 998
Lenhard, W. & Lenhard, A. (2014). Hypothesis Tests for Comparing Correlations. available: 999
https://www.psychometrica.de/correlation.html. Bibergau (Germany): Psychometrica. 1000
DOI: 10.13140/RG.2.1.2954.1367 1001
Leonardi, N., & Ville, V. D. (2015). On spurious and real fluctuations of dynamic functional 1002
connectivity during rest. NeuroImage, 104, 430-436. 1003
Episodic Memory and Spontaneous Neural Dynamics 47
47
LePort, A. K., Stark, S. M., McGaugh, J. L., & Stark, C. E. (2016). Highly Superior 1004
Autobiographical Memory: Quality and Quantity of Retention Over Time. Frontiers in 1005
psychology, 6, 2017. 1006
Levine, B., Svoboda, E., Turner, G. R., Mandic, M., & Mackey, A. (2009). Behavioral and 1007
functional neuroanatomical correlates of anterograde autobiographical memory in 1008
isolated retrograde amnesic patient M. L. Neuropsychologia, 47, 2188-2196. 1009
Levine, B., Svoboda, E., Hay, J. F., Winocur, G., & Moscovitch, M. (2002). Aging and 1010
autobiographical memory: Dissociating episodic from semantic retrieval. Psychology and 1011
Aging, 17, 677-689. 1012
Lurie, D. J., Kessler, D., Bassett, D. S., Betzel, R. F., Breakspear, M., Keilholz, S., Kucyi, 1013
A…Calhoun, V. D. (2019). Questions and controversies in the study of time-varying 1014
functional connectivity in resting fMRI. Network Neuroscience. Advance publication. 1015
https://doi.org/10.1162/netn_a_00116 1016
Margulies, D.S., Ghosh, S.S., Goulas, A., Falkiewicz, M., Huntenburg, J.M., Langs, G., … & 1017
Smallwood, J. (2016). Situating the default-mode network along a principal gradient of 1018
macroscale cortical organization. Proceedings of the National Academy of Sciences, 1019
113,12574-12579. 1020
Moscovitch, M., Cabeza, R., Winocur, G., & Nadel, L. (2016). Episodic memory and beyond: 1021
The hippocampus and neocortex in transformation. Annual Review of Psychology, 67, 1022
105-134. 1023
Neubauer, A. C., & Fink, A. (2009). Intelligence and neural efficiency. Neuroscience and 1024
Biobehavioral Reviews, 33, 1004-1023. 1025
Episodic Memory and Spontaneous Neural Dynamics 48
48
Palombo, D.J,. Alain, C., Soderlund, H., Khuu, W., & Levine, B. (2015). Severely deficient 1026
autobiographical memory (SDAM) in healthy adults: A new mnemonic 1027
syndrome. Neuropsychologia, 72, 105-118. 1028
Palombo, D. J., Bacopulos, A., Amaral, R. S. C., Olsen, R. K., Todd, R. M., Anderson, A. K., & 1029
Levine, B. (2018). Episodic autobiographical memory is associated with variation in the 1030
size of hippocampal subregions. Hippocampus, 28, 69-75. 1031
Palombo, D. J., Sheldon, S., & Levine, B. (2018). Individual differences in autobiographical 1032
memory. Trends in Cognitive Sciences, 22, 583-597. 1033
Palombo, D. J., Williams, L. J., Abdi, H., & Levine, B. (2013). The survey of autobiographical 1034
memory (SAM): A novel measure of trait mnemonics in everyday life. Cortex: A Journal 1035
Devoted to the Study of the Nervous System and Behavior, 49(6), 1526-1540. 1036
Petrican, R. & Grady, C.L. (2017). Contextual and developmental differences in the neural 1037
architecture of cognitive control. Journal of Neuroscience, 37, 7711-7726. 1038
Petrican, R., & Levine, B. T. (2018). Similarity in functional brain architecture between rest and 1039
specific task modes: A model of genetic and environmental contributions to episodic 1040
memory. NeuroImage, 179, 489-504. 1041
Power, J.D., Cohen, A.L., Nelson, S.M., Wig, G.S., Barnes, K.A., ...& Petersen, S. E. (2011) 1042
Functional network organization of the human brain. Neuron, 72, 665–678. 1043
Power, J.D., Barnes, K.A., Snyder, A.Z., Schlaggar, B.L., & Petersen, S.E. (2012). Spurious but 1044
systematic correlations in functional connectivity MRI networks arise from subject 1045
motion. NeuroImage, 59, 2142-2154. 1046
Power, J. D., Schlaggar, B. L., & Petersen, S. E. (2015). Recent progress and outstanding issues 1047
in motion correction in resting state fMRI. NeuroImage, 105, 536-551. 1048
Episodic Memory and Spontaneous Neural Dynamics 49
49
Preti, M. G., Bolton, T. A. W., & Ville, V. D. (2017). The dynamic functional connectome: 1049
State-of-the-art and perspectives. NeuroImage, 160, 41-54. 1050
Rabin, J. S., Carson, N., Gilboa, A., Stuss, D. T., & Rosenbaum, R. S. (2013). Imagining other 1051
people’s experiences in a person with impaired episodic memory: The role of personal 1052
familiarity. Frontiers in Psychology, 3, 10. 1053
Renoult, L., Irish, M., Moscovitch, M., & Rugg, M. D. (2019). From knowing to remembering: 1054
The semantic–episodic distinction. Trends in Cognitive Sciences. 1055
doi:http://dx.doi.org.myaccess.library.utoronto.ca/10.1016/j.tics.2019.09.008 1056
Rubinov, M. & Sporns O. (2010) Complex network measures of brain connectivity: uses and 1057
interpretations. NeuroImage, 52,1059–1069. 1058
Rubinov, M. & Sporns O. (2011) Weight-conserving characterization of complex functional 1059
brain networks. NeuroImage, 56,2068-2079. 1060
Sadaghiani, S. & D’Esposito, M. (2015) Functional characterization of the cingulo-opercular 1061
network in the maintenance of tonic alertness. Cerebral Cortex, 25, 2763–2773. 1062
Satterthwaite, T.D., Wolf, D.H., Ruparel, K., Erus, G., Elliott, M.A., Eickhoff, S.B., Gennatas, 1063
E.D., Jackson, C., Prabhakaran, K.,Smith, A., Hakonarson, H., Verma, R., Davatzikos, 1064
C., Gur, R.E., & Gur, R.C. (2013). Heterogeneous impact of motion on fundamental 1065
patterns of developmental changes in functional connectivity during youth. NeuroImage, 1066
83, 45–57. 1067
Satterthwaite, T. D., Wolf, D. H., Roalf, D. R., Ruparel, K., Erus, G., Vandekar, S., . . . & Gur, 1068
R. C. (2015). Linked sex differences in cognition and functional connectivity in youth. 1069
Cerebral Cortex, 25(9), 2383-2394. 1070
Episodic Memory and Spontaneous Neural Dynamics 50
50
Schultz, D. H., & Cole, M. W. (2016). Higher intelligence is associated with less task-related 1071
brain network reconfiguration. The Journal of Neuroscience, 36, 8551-8561. 1072
Sheldon. S., Amaral, R.S.C., & Levine. B. (2016). Individual differences in visual 1073
imagery determine how event information is remembered. Memory, 5, 1-10. 1074
Sheldon, S., Farb, N., Palombo, D. J., & Levine, B. (2016). Intrinsic medial temporal lobe 1075
connectivity relates to individual differences in episodic autobiographical 1076
remembering. Cortex: A Journal Devoted to the Study of the Nervous System and 1077
Behavior, 74, 206-216. 1078
Sherry, A., & Henson, R. K. (2005). Conducting and interpreting canonical correlation analysis 1079
in personality research: A user-friendly primer. Journal of Personality Assessment, 84, 1080
37-48. 1081
Shrout, P. E., & Fleiss, J. L. (1979). Intraclass correlations: Uses in assessing rater 1082
reliability. Psychological Bulletin, 86, 420-428. 1083
Smith, S. M., M. Jenkinson, M.W. Woolrich, C.F. Beckmann, T.E.J. Behrens, … & P.M. 1084
Matthews (2004). Advances in functional and structural MR image analysis and 1085
implementation as FSL. NeuroImage, 23, 208-219. 1086
Smith, S.M., Beckmann, C.F., Andersson, J., Auerbach, E.J., Bijsterbosch, J., Douaud, G., et al. 1087
(2013). Resting-state fMRI in the Human Connectome Project. NeuroImage, 80, 144–1088
168. 1089
Smith, S. M., Nichols, T. E., Vidaurre, D., Winkler, A. M., Behrens, T. E. J., Glasser, M. F., . . . 1090
Miller, K. L. (2015). A positive-negative mode of population covariation links brain 1091
connectivity, demographics and behavior. Nature Neuroscience, 18, 1565-1567. 1092
Episodic Memory and Spontaneous Neural Dynamics 51
51
Sporns, O. & Betzel, R.F. (2016). Modular brain networks. Annual Review of Psychology, 67, 1093
613-640. 1094
Staresina, B. P., & Wimber, M. (2019). A neural chronometry of memory recall. Trends in 1095
Cognitive Sciences. 1096
doi:http://dx.doi.org.myaccess.library.utoronto.ca/10.1016/j.tics.2019.09.011 1097
Stern, Y. (2003). The concept of cognitive reserve: a catalyst for research. Journal of Clinical 1098
and Experimental Neuropsychology, 25, 589–593. 1099
Stern, Y., Scarmeas, N., & Habeck, C. (2004). Imaging cognitive reserve. International Journal 1100
of Psychology, 39, 18–26. 1101
Telesford, Q. K., Lynall, M., Vettel, J., Miller, M. B., Grafton, S. T., & Bassett, D. S. (2016). 1102
Detection of functional brain network reconfiguration during task-driven cognitive 1103
states. NeuroImage, 142, 198-210. 1104
Toga, A.W., Thompson, P.M., & Sowell, E.R. (2006). Mapping brain maturation. Trends in 1105
Neurosciences, 29, 148–159. 1106
Tsvetanov, K. A., Henson, R. N. A., Tyler, L. K., Razi, A., Geerligs, L., Ham, T. E., & Rowe, J. 1107
B. (2016). Extrinsic and intrinsic brain network connectivity maintains cognition across 1108
the lifespan despite accelerated decay of regional brain activation. The Journal of 1109
Neuroscience, 36, 3115-3126. 1110
Tulving, E. (2002) Episodic memory: From mind to brain. Annual Review of Psychology, 53, 1-1111
25. 1112
Tulving, E. (1985). Memory and consciousness. Canadian Psychology/Psychologie 1113
Canadienne, 26, 1-12. 1114
Episodic Memory and Spontaneous Neural Dynamics 52
52
Vannucci, M., Pelagatti, C., Chiorri, C., & Mazzoni, G. (2016). Visual object imagery and 1115
autobiographical memory: Object imagers are better at remembering their personal 1116
past. Memory, 24, 455-470. 1117
Vatansever, D., Menon, D. K., Manktelow, A. E., Sahakian, B. J., & Stamatakis, E. A. (2015). 1118
Default mode dynamics for global functional integration. The Journal of 1119
Neuroscience, 35, 15254-15262. 1120
Vatansever, D., Bzdok, D., Wang, H., Mollo, G., Sormaz, M., Murphy, C., . . . Jefferies, E. 1121
(2017). Varieties of semantic cognition revealed through simultaneous decomposition of 1122
intrinsic brain connectivity and behaviour. NeuroImage, 158, 1-11. 1123
Vatansever, D., Menon, D. K., & Stamatakis, E. A. (2017). Default mode contributions to 1124
automated information processing. PNAS Proceedings of the National Academy of 1125
Sciences of the United States of America, 114, 12821-12826. 1126
Vinh, N.X, Epps, J., & Bailey, J. (2010). Information theoretic measures for clusterings 1127
comparison: Variants, properties, normalization and correction for chance. Journal of 1128
Machine Learning Research, 11, 2837-2854. 1129
Westphal, A. J., Reggente, N., Ito, K. L., & Rissman, J. (2016). Shared and distinct contributions 1130
of rostrolateral prefrontal cortex to analogical reasoning and episodic memory 1131
retrieval. Human Brain Mapping, 37, 896-912. 1132
Wheatley, T., Milleville, S.C., & Martin, A. (2007). Understanding animate agents: distinct roles 1133
for the social network and mirror system. Psychological Science, 18, 469–474. 1134
Wheeler, M. E., & Buckner, R. L. (2004). Functional-anatomic correlates of remembering and 1135
knowing. NeuroImage, 21, 1337–1349. 1136
Episodic Memory and Spontaneous Neural Dynamics 53
53
Whitfield-Gabrieli, S. & Nieto-Castanon, A. (2012). CONN: a functional connectivity toolbox 1137
for correlated and anticorrelated brain networks. Brain Connectivity, 2, 125-141. 1138
Wig, G.S., Schlaggar, B.L.,Petersen,S.E. (2011). Concepts and principles in the analysis of brain 1139
networks. Annals of the New York Academy of Sciences,1224, 126–146. 1140
Xia, M., Wang, J., & He, Y. (2013) BrainNet Viewer: A Network Visualization Tool for Human 1141
Brain Connectomics. PLoS ONE 8: e68910. 1142
Yeo, B. T. T., Krienen, F. M., Sepulcre, J., Sabuncu, M. R., Lashkari, D., Hollinshead, M., . . . 1143
Buckner, R. L. (2011). The organization of the human cerebral cortex estimated by 1144
intrinsic functional connectivity. Journal of Neurophysiology, 106, 1125-1165. 1145
1146
1147
Episodic Memory and Spontaneous Neural Dynamics 54
54
Figure Captions 1148
Figure 1. Schematic representation of our preprocessing pipeline. 1149
Figure 2. Schematic representation of our analysis pipeline. 1150
Figure 3. The group-based consensus partitions for each task mode under scrutiny. Network 1151
labels are based on Power et al., (2011). The brain networks were visualized with the BrainNet 1152
Viewer (http://www.nitrc.org/projects/bnv/) (Xia et al., 2013). 1153
Figure 4. Histogram showing the distribution of our core task switch variable, i.e., number of 1154
perceptual-to-semantic processing transitions, in the HCP sample. 1155
Figure 5. The median canonical loadings (panel [a]) and the median canonical weights (panel 1156
[b]) of the brain and behavioral variables on their corresponding canonical variate across all 1157
discovery CCAs, as well as the scatter plot describing the linear association between the two 1158
canonical variates across all the “test” folds (panel [c]). In panels a and b, error bars reflect the 1159
smallest and largest value, respectively, corresponding to the loading (panel [a]) or canonical 1160
weight (panel [b]) of each variable on its corresponding variate across all the discovery CCAs. 1161
The scatter plot in panel (c) is based on standardized variables. 1162
Figure 6. Histogram showing the distribution of our core task switch variable, i.e., number of 1163
perceptual-to-semantic processing transitions, in the SAM sample. 1164
Figure 7. Scatter plot describing the association between Episodic SAM scores and the residual 1165
brain variate linked to visual EM in the HCP sample (see main text for further details) 1166
Extended data 1. File containing the scripts used in the community detection analyses 1167
(“community_louvain.m”,” agreement.m”,”consensus.m [tau = 0, 100 reps]”) and the analyses 1168
characterizing similarity in functional brain organization between rest and the various task 1169
modes 1170
Episodic Memory and Spontaneous Neural Dynamics 55
55
(“normalized_mutual_information.m”,”Max_AMI_multiple.m”,”Count_states.m”,”Count_sequ1171
ences.m”) 1172
1173
1174
1
Table 1. Average similarity among the group-based task architectures across the three values of
the spatial resolution parameter, expressed as the adjusted normalized mutual information index
(AMI)
Note. The normalized mutual information index ranges from 0 (no similarity between the two
partitions) to 1 (perfect similarity between the two partitions).
Perceptual
Processing
Semantic
Processing
Math Processing
Motor
Processing
Perceptual
Processing
--
.23
.53
.43
Semantic Processing
.23
--
.32
.23
Math Processing
.53
.32
--
.47
Motor Processing
.43
.23
.47
--
1
Table 2. Average number of task states expressed during rest and average similarity between
individual rest and the group-based task architectures across the three values of the spatial
resolution parameter in the HCP and the SAM samples
Note. M= mean; SD = standard deviation. The HCP data comprises 1548 states, whereas the
SAM data comprises 166 states.
HCP States
(M ± SD)
HCP Global AMI
(M ± SD)
SAM States
(M ± SD)
SAM Global AMI
(M ± SD)
Perceptual Processing
346.80 ± 80.99
.35 ± .09
32.80 ± 12.38
.11 ± .06
Semantic Processing
214.35 ± 65.31
.22 ± .06
24.63 ± 10.97
.07 ± .03
Math Processing
536.17 ± 88.86
.33 ± .08
66.68 ± 13.03
.12 ± .05
Motor Processing
450.68 ± 106.53
.35 ± .08
41.89 ± 15.95
.12 ± .06
1
Table 3. Average number of switches between task states, as observed during rest, across the
three values of the spatial resolution parameter in the HCP and the SAM samples
Note. M= mean; SD = standard deviation. In the HCP data, there is a maximum of 1547 possible
switches, whereas in the SAM data there is a maximum of 165 switches.
HCP (M ± SD)
SAM (M ± SD)
Perceptual_Semantic Processing
18.75 ± 6.16
2.08 ± 1.46
Perceptual Processing_Math
55.31 ± 10.84
6.28 ± 2.34
Perceptual_Motor Processing
41.66 ± 9.52
3.46 ± 1.69
Semantic_Perceptual Processing
18.55 ± 6.00
2.36 ± 1.69
Semantic Processing_Math
40.88 ± 11.35
4.77 ± 2.42
Semantic_Motor Processing
24.82 ± 7.26
2.46 ± 1.49
Math_Perceptual Processing
55.61 ± 11.09
5.99 ± 2.25
Math_Semantic Processing
41.06 ± 11.21
4.93 ± 2.79
Math_Motor Processing
65.47 ± 12.61
6.88 ± 1.98
Motor_Perceptual Processing
41.30 ± 9.27
3.47 ± 1.67
Motor_Semantic Processing
24.50 ± 7.16
2.58 ± 1.64
Motor Processing_Math
66.06 ± 12.69
6.79 ± 2.08