Cognitive reserve, neurocognitive performance, and high-order resting-state networks in cognitively unimpaired aging

Neurobiology of Aging 117 (2022) 151–164 Contents lists available at ScienceDirect Neurobiology of Aging journal homepage: www.elsevier.com/locate/neuaging.org Cognitive reserve, neurocognitive performance, and high-order resting-state networks in cognitively unimpaired aging Benxamín Varela-López a , Álvaro Javier Cruz-Gómez b , Cristina Lojo-Seoane c , Fernando Díaz a , A.X. Pereiro c , Montserrat Zurrón a , Mónica Lindín a , Santiago Galdo-Álvarez a , ∗ a Department of Clinical Psychology and Psychobiology, Universidade de Santiago de Compostela, Galicia, Spain b Psychophysiology and Neuroimaging Group, Institute of Biomedical Research Cadiz (INiBICA), Andalusia, Cadiz, Spain c Department of Developmental and Educational Psychology, Universidade de Santiago de Compostela, Spain a r t i c l e i n f o Article history: Received 24 December 2021 Revised 6 May 2022 Accepted 7 May 2022 Available online 28 May 2022 Keywords: Cognitive reserve DAN FPCN Healthy aging Resting-state functional magnetic resonance a b s t r a c t Cognitive Reserve (CR) is considered a protective factor during the aging process. However, although CR is a multifactorial construct, it has been operationalized in a unitary way (years of formal education or IQ). In the present study, a validated measure to categorize CR holistically (Cognitive Reserve Index Questionnaire) was used to evaluate the resting-state functional connectivity in 77 cognitively unimpaired participants aged 50 years and over with high and low CR, and matched brain global atrophy levels. The connectivity of networks linked to attentional (Dorsal Attention Network -DAN-) and executive (Frontal- Parietal Control Network -FPCN-) processes were evaluated by the combination of Independent Component Analysis and seed-based approaches, since these networks have been proposed as candidates to underlie the protective effect of CR in the aging context. Participants with high CR showed an increase of the connectivity in the FPCN and a decrease in the DAN with respect to the low CR group, correlating with neuropsychological scores and supporting that high CR is related to a better neurocognitive preservation during aging. ©2022 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license ( http://creativecommons.org/licenses/by-nc-nd/4.0/ ) 1. Introduction Normal aging is a complex physiological process which involves changes in brain architecture and function ( Mak et al., 2017 ), with several consequences regarding cognition. The cognitive functions most affected by aging are those that require rapid processing or transformation of the information to make a decision, reflected in a decline in speed processing, working memory and executive functioning capacities; however, accumulated knowledge and acquired skills are well maintained into advanced age ( Park et al., 2002 ; Salthouse, 2019 ). The effects of aging on cognition depend on the interactions between a broad range of adverse biological and neurophysiological factors and several protective and compensatory processes ( Reuter- ∗Corresponding author at: Department of Clinical Psychology and Psychobiology, Universidade de Santiago de Compostela, Santiago de Compostela, Galicia, Spain. Tel: +34 881813701, Fax: +34 981521581. E-mail address: santiag[email protected] (S. Galdo-Álvarez). Lorenz and Park, 2014 ). Indeed, no linear relationships between clinical manifestations and the pathological severity of the aging process have been detected ( Bennett et al., 2006 ; Bennett et al., 2012 ; Buchman and Bennett, 2012 ). It has been suggested that cognitive reserve (CR) protects against the adverse effects of aging on cognition, increasing tolerance or compensatory capacity throughout aging ( Barulli and Stern, 2013 ). It has been suggested that CR, which can be described as preexisting cognitive processes dependent on genetic and/or environmental factors, enables more efficient performance in cases of pathology or brain impairment ( Cabeza et al., 2018 ; Jones et al., 2011 ; Stern, 2009 ) and mitigates the effects of neural decline caused by ageing or age-related diseases by means of increased neural capacity and/or increased neural efficiency ( Cabeza et al., 2018 ). Research has shown that CR provides some protection against cognitive impairment in normal aging ( Giogkaraki et al., 2013 ) and in older adults with subjective cognitive complaints ( Lojo- Seoane et al., 2014 , 2018 ). In addition, CR has been also associated 0197-4580/$ –see front matter ©2022 The Author(s). Published by Elsevier Inc. This is an open access article under the CC BY-NC-ND license ( http://creativecommons.org/licenses/by-nc-nd/4.0/ ) https://doi.org/10.1016/j.neurobiolaging.2022.05.012 152 B. Varela-López, Á.J. Cruz-Gómez, C. Lojo-Seoane et al. / Neurobiology of Aging 117 (2022) 151–164 with a lower risk of development of dementia ( Bastin et al., 2012 ; Franzmeier et al., 2017c ) and has been shown to be a protective mediator in the conversion from mild cognitive impairment to dementia ( Facal et al., 2019 ). Considering the dynamic nature of cognitive reserve, that is, it can be modified throughout the lifespan ( Fritsch et al., 2007 ), and as it has been considered to protect against the deleterious effects of normal aging and the development of dementia ( Bastin et al., 2012 ; Franzmeier, et al., 2017d ), promoting activities targeting the different proxies included in the Cognitive Reserve Index questionnaire (CRIq) (i.e., education, working activity and leisure time) should be included in potential therapeutic approaches in the context of normal aging and dementia ( Anthony & Lin, 2018 ; Bastin et al., 2012 ; Franzmeier et al., 2017d ). In addition to the traditional study of normal and pathological aging through the functional magnetic resonance imaging (fMRI) task-activation technique ( Hafkemeijer et al., 2012 ), there is a growing interest from the scientific community regarding the study of the effect of aging on the resting state networks ( Hafkemeijer et al., 2012 ; Mak et al., 2017 ). Resting state functional resonance imaging (rsfMRI) measures fluctuations in the BOLD signal and can thus map and evaluate the functional brain connectivity ( Lee et al., 2013 ). This technique has been suggested as a promising tool for the study of normal cognition ( Hoffman and Morcom, 2018 ; Mak et al., 2017 ) and pathological aging ( Dennis and Thompson, 2014 ; Galvin et al., 2011 ; Hojjati et al., 2017 ). Two resting-state networks associated with higher cognitive functions, that is the dorsal attention network (DAN) and the fronto-parietal control network (FPCN), have been found to undergo decline early on in the aging process ( Kim, 2019 ; Murman, 2015 ). The DAN is composed by brain regions of the superior parietal lobule, medial intraparietal sulcus, motion-sensitive middle temporal area (MT + ), inferior temporal cortex, frontal eyes field and the inferior frontal junction ( Kim, 2015 ; Yeo et al., 2011 ; Vossel et al., 2014 ). This network plays an important role in maintaining priority spatial maps for guiding covert and overt attention, saccade planning, observing action and also in visual working memory and thus in memory encoding ( Dukelow et al., 2001 ; Griffis et al., 2015 ; Huk et al., 2002 ; Kim, 2019 ; Vossel et al., 2014 ). The age-related alteration of DAN functional connectivity may be related to the age-related decline in attention ( Betzel et al., 2014 ; Ferreira and Busatto, 2013 ; Murman, 2015 ; Tomasi and Volkow, 2012 ; Varangis et al., 2019 ). Moreover, studies of brain metabolism ( Alexander et al., 1997 ; Bastin et al., 2012 ) and the BOLD signal ( Bastin et al., 2012 ; Franzmeier, et al., 2017b , c ) have shown that the level of CR modulates the resting-state functional activity of the DAN in both normal and pathological aging. The FPCN has been considered a hub for high order functions and cognitive control; by interacting with other structures of the whole brain, it enables rapid, flexible performance of novel tasks ( Marek and Dosenbach, 2018 ). This network is made up of regions that are preferentially involved in maintenance of cognitive control (the anterior prefrontal, insular and cingulate cortices) and regions mainly related to feedback in the context of cognitive control (the dorsolateral prefrontal cortex, the inferior parietal lobe, the caudate nucleus and the lateral cerebellum) ( Chiu et al., 2017 ; Dosenbach et al., 2006 ; Koziol et al., 2016 ; Marek and Dosenbach, 2018 ; Vincent et al., 2008 ). The FPCN also includes areas involved in language production and comprehension, that is the Broca ´ s and Wernicke ´ s regions and their right hemisphere equivalents ( Laird et al., 2011 ; Zhu et al., 2014 ). The aging-related modification of FPCN activity has been attributed to cognitive decline and to compensatory mechanisms in response to cognitive decline ( Jockwitz et al., 2017 ; Malagurski et al., 2020 ; Nashiro et al., 2017 ; Oschmann et al., 2020 ). The FPCN has also been suggested to at least partly explain the protective effect of CR through the lifespan and during pathological aging ( Cole et al., 2012 , 2015 ; Franzmeier et al., 2017b , c ; Jiang et al., 2020 ), as well as in sustaining compensatory mechanisms in early Alzheimer ´ s disease ( Elman et al., 2014 ; Oh et al., 2015 ). Some studies have evaluated samples meeting diagnostic criteria of pathological states, including mild cognitive impairment ( Bosch et al., 2010 ; Bozzali et al., 2015 ; Franzmeier, et al., 2017 a , b , c , d ) and/or Alzheimer’s dementia ( Alexander et al., 1997 ; Bozzali et al., 2015 ; Scarmeas et al., 2003 ; Weiler et al., 2018 ). However, despite promising results, it is difficult to understand the effects of CR on a brain undergoing extensive neurological changes. Most studies determine CR from years of education ( Weiler et al., 2018 ), scores on cognitive/intelligence tests ( Franzmeier et al., 2017d ), or a combination of both ( Bastin et al., 2012 ; Franzmeier et al., 2017 b , c ). However, it has been recognised that a global assessment of CR should include other factors, such as occupation and leisure and social activities ( Bozzali et al., 2015 ). Thus, although some studies have categorized CR on the basis of various indicators that act independently ( Bennet et al., 2006 ) or that are included in latent variables ( Lojo-Seoane et al., 2014 ; Siedlecki et al., 2009 ) or factors ( Mitchell et al., 2012 ), a standard cognitive reserve index is required. Therefore, the aims of the present study were as follows: (1) to determine the effect of CR, evaluated holistically, by comparing resting state DAN and FPCN connectivity in cognitively unimpaired older adults (participants over 50 years old) divided into 2 groups (high CR and low CR); and (2) to determine whether the effect of CR on the activity of the 2 networks is related to the cognitive performance. To fulfil the study aims, the CRIq ( Nucci et al., 2012 ) was administrated to participants. This questionnaire provides a standard CR measure in which each piece of information has a weighted value. In line with the Scaffolding Theory of Aging ( Park and Reuter- Lorenz, 2009 ), which postulates that the anterior brain regions have the greatest capacity to functionally reorganize and maintain good levels of cognitive performance during aging, and also considering previous findings on CR ( Cole et al., 2012 , Cole et al., 2015 ; Elman et al., 2014 ; Jiang et al., 2020 ; Oh et al., 2015 ), we expected to observe better neuropsychological performance in the high CR group than in the low CR group. Associated with this better performance, it should be possible to observe an increase in the connectivity of anterior brain regions within the 2 networks explored, reflecting an increase in neural capacity during the aging process ( Cabeza et al., 2018 ) and/or a decrease in the connectivity of posterior regions as a result of improved network efficiency ( Bastin et al., 2012 ; Cabeza et al., 2018 ). 2. Material and methods 2.1. Sample A total of 77 participants (66 women and 11 men) who had undergone head MRI scanning were selected from a sample of 214 cognitively unimpaired older adults included in the Compostela Aging Study (CompAS), an ongoing longitudinal project to detect cognitive impairment in people aged 50 years and over attending primary care centres in Galicia (north-west Spain) ( Juncos- Rabadán et al. 2012 ). None of the participants had prior diagnosis of dementia, psychiatric or neurological disorders, severe illness, deafness or blindness; they were not receiving chemotherapy and they did not have B. Varela-López, Á.J. Cruz-Gómez, C. Lojo-Seoane et al. / Neurobiology of Aging 117 (2022) 151–164 153 Table 1 Demographic and neuropsychological variable scores (means and standard deviation in parentheses) obtained by the two groups CR + (High cognitive reserve) and CR- (Low cognitive reserve), and group comparison (t values, df: degrees of freedom) Variable Group CR + (n = 38) Group CR-(n = 39) Group comparisont-values, df Age (years) 64.13 (7.67) 65.65 (7.37) 0.52, 75 Gender 30 women/8 men 36 women/3 men 2.806, 1 BV/CSF Index 42.15 (23.18) 39.06 (23.38) -0.581, 75 Years of formal education 17.55 (4.08) 7.38 (2.87) -12.60 a , 66.24 MMSE 29.05 (1.13) 27.53 (1.84) -4.34 a , 63.46 CAMCOG-R Total 97.55 (4.10) 89.33 (5.10) -7.80 a , 72.43 CAMCOG-R Attention-Calculation 8.36 (0.85) 6.92 (1.70) -4.71 a , 56.13 CAMCOG-R Memory 23.44 (2.08) 21.28 (2.13) -4.49 a , 75 CAMCOG-R Executive functions 23.73 (2.61) 18.66 (3.75) -6.89 a , 68.01 CVLT Immediate Free Recall 56.86 (8.50) 49.25 (8.00) -4.04 a , 75 CVLT Short-Delayed Free Recall 12.73 (2.45) 10.43 (2.26) -4.27 a , 75 CVLT Long-Delayed Free Recall 13.10 (2.29) 11.30 (2.24) -3.46 a , 75 WAIS-III Vocabulary 53.68 (7.65) 37.23 (12.00) -7.14 a , 64.72 BDAE 55.55 (4.84) 46.76 (6.10) -6.97 a , 75 CRIq Total score 128.31 (9.67) 86.89 (2.24) -28.02 a , 61.89 Key: BDAE, Boston diagnostic aphasia examination; BV/CSF, Global brain atrophy index; CAMCOG-R, Cambridge cognitive examination-revised; CRIq, cognitive reserve index questionnaire; CVLT, California verbal learning test; MMSE, mini-mental state examination; WAIS-III = Wechsler adult intelligence scale. a p < 0.001 alcohol or other substance use disorder. All participants gave their written informed consent prior to participation in the study. The research was approved by the Galician Ethics Committee for Clinical Research and was performed in accordance with the ethical standards established in the 1964 Declaration of Helsinki and revised in Seoul 2008. Each participant underwent extensive evaluation, including review of their medical history and neuropsychological assessment. The following assessments/tools were used during study: (1) Spanish version of the CRIq ( http://www.cognitivereserveindex.org ). The CRIq is a semi-structured interview of 24 item that quantifies CR through information regarding the individual’s entire adult life. The questionnaire combines information about education (such as years of formal schooling and/or courses completed), working activity (number of years of dedication in different work activities according to their cognitive complexity) and leisure time (e.g., reading newspapers and books, going to the cinema or theatre, gardening, social activities, trips) for holistic categorization of the CR construct. (2) General cognitive functioning was evaluated by the Spanish version of the Mini-Mental State Examination MMSE ( Lobo et al., 1999 , with norms for age and education) and by the Spanish version of the Cambridge Cognitive Examination (CAMCOG-R, total score) ( Roth et al., 1999 ; with norms for age and education, Pereiro et al., 2015 ). (3) Memory was assessed using the corresponding subscale of CAMCOG-R and the Spanish version of the California Verbal Learning Test (CVLT) ( Benedet and Alejandre, 1998 ; Delis et al., 1987 ) with scores in Immediate free recall, Short-delayed free recall and Long-delayed free recall. (4) Attention and Executive functions were evaluated using the corresponding subscales of the CAMCOG-R. E) Language was assessed using the Vocabulary subscale of the Wechsler intelligence scale (WAIS-III, Wechsler, 2002 ) and the abbreviated form of the Spanish version of the Boston Diagnostic Aphasia Examination (BDAE) – Spanish version ( Goodglass and Kaplan, 1996 ). All participants were classified as cognitively unimpaired adults according to norms for age and years of education. Participants were classified as having a high level of CR or low level of CR according to the percentile distribution of the total CRIq scores from the 214 unimpaired participants (source sample): high cognitive reserve (CR + ), when the scoring was equal to or greater than 75th percentile, and low cognitive reserve (CR-), when the scoring was equal to or less than 25th percentile. Furthermore, following the recommendations of the Reserve and Resilience initiative ( https://reserveandresilience.com ), both groups were matched on a measure of age-related brain change that has been linked to cognitive decline, that is the Brain Volume/Cerebrospinal Fluid Index (BV/CSF index; Orellana et al., 2016 ), a validated proxy for the global measurement of brain atrophy, was obtained (see Data preprocessing section). This index has been strongly associated with age and also distinguishes between groups of participants with different clinical status, as cognitively unimpaired or with Mild Cognitive Impairment or Alzheimer Dementia ( Orellana et al., 2016 ). Demographic and neuropsychological profiles of both groups are shown in Table 1 . 2.2. MRI acquisition parameters Magnetic resonance imaging (MRI) was performed with a Philips 3T Achieva scanner (Philips Medical System). A sagittal T1-weighted 3D Magnetization Prepared Rapid Acquisition Gradient Echo (MPRAGE) sequence (repetition time/echo time = 7.45 ms/3.40 ms, flip angle = 8 º; 180 slices, voxel size = 1 ×1 ×1 mm, field of view = 240 ×240 mm 2 , matrix size = 240 ×240 mm) and functional magnetic resonance images were acquired with a gradient echo-planar imaging (EPI) sequence sensitive to blood oxygen level-dependent (BOLD) contrast (repetition time/echo time = 20 0 0 ms / 30 ms, flip angle = 87 º; 37 interleaved slices, voxel size = 3 ×3 ×3.5 mm, field of view = 240 ×240 mm 2 , matrix size = 80 ×80 mm) during open-eyes resting state (while a fixation cross was presented, following previous recommendations; Patriat et al., 2013 ). Four dummy scans were automatically discarded before image acquisition to prevent signals arising from progressive saturation. Head movements of the participants were minimized by using a vacuum cushion during the MRI data acquisition. 2.3. Data pre-processing and functional connectivity analysis Both structural and functional images were pre-processed and analysed using the CONN 19c ( https://www.nitrc.org/projects/conn ; Whitfield-Gabrieli and Nieto-Castanon, 2012 ) and SPM12 (Wellcome Department of Imaging Neuroscience, London, UK; http: //www.fil.ion.ucl.ac.uk/spm/ ) toolboxes implemented in Matlab R2019a. In addition, the high-resolution sagittal 3-dimensional T1-weighted MPRAGE sequences (1 ×1 ×1 mm 3 ) were processed with the FreeSurfer v6.0.0 program ( https://surfer.nmr.mgh. harvard.edu/ ), to obtain the BV/CSF index (where the total brain 154 B. Varela-López, Á.J. Cruz-Gómez, C. Lojo-Seoane et al. / Neurobiology of Aging 117 (2022) 151–164 Fig. 1. The 3 independent components derived from the whole sample. DAN (top), left FPCN (middle) and right FPCN (bottom). Color bars correspond to t-scores. “(For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)” volume is put in relation to the total CSF volume). Considering that this index is significantly associated with total intracranial volume (TIV), volume measurements were adjusted using the estimated total intracranial volume (eTIV) using the formula, adjusted_volume = volume_observed –b ∗(eTIV – mean_eTIV) where mean_eTIV is the average eTIV of all participants and b is the coefficient of regression between the volume observed and the eTIV ( Voevodskaya, 2014 ). We followed the default pre-processing pipeline of the CONN toolbox, which included functional realignment and unwarping, functional centring of the image to (0, 0, 0) coordinates, slicetiming correction, functional outlier detection, simultaneous functional direct segmentation and normalization to MNI space, structural centring to (0, 0, 0) coordinates, simultaneous structural segmentation and normalization to MNI space and spatial smoothing with a kernel of 8 mm FWHM. Potential confounding effects of the estimated BOLD signal (noise components from white matter and cerebrospinal areas, estimated subject-motion parameters, scrubbing and rest effects) were estimated and removed separately for each voxel, participant and session, using Ordinary Least Squares regression to project each BOLD signal time series to the sub-space orthogonal to all potential confounding effects as part of the de-noising step ( Behzadi et al., 2007 ; Friston et al., 1996 ; Power et al., 2014 ; Whitfield-Gabrieli and Nieto-Castanon, 2012 ). CONN’s default de-noising pipeline implements an anatomical component-based noise correction procedure (aCompCor), which can extract white matter and cerebral spinal fluid noise components ( Behzadi et al., 2007 ). This procedure was chosen rather than global signal regression because aCompCor has demonstrated higher levels of specificity and sensitivity for positive correlations while addressing negative correlations ( Chai et al., 2012 ). This procedure is combined with quantification of participant motion and identification of outlier scans in the Artifact Rejection Toolbox ( Shirer et al., 2015 ; Whitfield-Gabrieli and Nieto-Castanon, 2012 ). The Artifact Rejection Toolbox was set to the 97th percentile setting, with the mean global-signal deviation threshold at z = ±5 and the participant-motion threshold at 0.9 mm. Motion and outlier information were included as covariates in the first-level analyses. Linear regression of the potential confounds and temporal band-pass filter of 0.008–0.09 Hz were applied to the data to exclude signal frequencies outside of the range of expected BOLD signals, reduce the impact of participant motion, extract white matter and cerebral spinal fluid noise components, account for com- B. Varela-López, Á.J. Cruz-Gómez, C. Lojo-Seoane et al. / Neurobiology of Aging 117 (2022) 151–164 155 Fig. 2. Significant differences in brain connectivity obtained through ICA (left) and effect size based on mean Fisher’s Z scores for the significant clusters (right). Top panel: decrease in connectivity of the inferior temporal gyrus within the DAN (CR + < CR-). Bottom panel: increased connectivity in 2 clusters mainly located in the ventrolateral prefrontal cortex within the left FPCN (CR + > CR-). Color scales correspond to t-scores, warm colors represent positive t values, cool colors indicate negative t values. “(For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)” mon rest effects and control for within-participant realignment and scrubbing covariates. No significant differences were observed between groups in terms of the movement parameters measured by CONN or between groups in relation to the number of scans after outlier regression. Two modalities of resting state functional analysis were applied with the aim of studying the right and left dorsal attention (DAN) and the frontoparietal (FPCN) networks. First, Independent Component Analysis (ICA) satisfactorily identified the DAN and FPCN (right and left) networks. Since the ICA method only addresses within-network connectivity, seed-based analysis (SBA) was conducted to better understand the nature of the observed differences at network-level connectivity (i.e., whether the differences are constrained to within-network or without-network regions) ( Smitha et al., 2017 ; Yang et al., 2020 ). A seed to voxel analysis was performed, with the regions that showed maximum significant differences in connectivity through group-ICA analysis (post-hoc analysis) used as seeds (for a similar approach see Soman et al., 2020 ). In addition, bivariate correlations (Pearson’s r for variables adjusted to normal distribution, Spearman’s rho for variables not adjusted to normal distribution) between neuropsychological and functional connectivity (ICA and seed-to-voxel analysis) measures were determined. The automated anatomical atlas 3 (AAL3) was used for anatomical reference ( Rolls et al., 2020 ). Both ICA & SBA 1-sample and 2-sample t-tests were assessed at p < 0.05, with FWE cluster-level corrected for multiple comparisons in a combination with a threshold of p < 0.001 at the uncorrected voxel level. 2.4. Independent component analysis Independent component analysis was performed using the Group ICA of the fMRI Toolbox ( http://mialab.mrn.org/software/ , GroupICATv4.0b), the CONN 19c, ICN_Atlas ( Kozák et al., 2017 ) and SPM12 toolboxes implemented in Matlab R2019a. First, GroupICATv4.0b was used to estimate the number of independent components in the sample, as the CONN toolbox does not provide this function. We used the minimum description length (MDL) criterion to reduce the dimension of the data and estimate the number of components, which increases the efficiency of ICA ( Akhbari and Fatemizadeh, 2009 ). The MDL criterion reported a value of 34 Independent Components for the sample (n = 77). The first level ICA analysis was then carried out using the CONN toolbox for 34 components (estimation by the MDL criterion). We completed the following steps, implemented in CONN and derived from the Calhoun’s group-level ICA approach ( Calhoun et al. 2001 ): variance normalization pre-conditioning, subject-level dimensionality reduction (64, CONN default value), subject/condition concatenation of BOLD signal data along the temporal dimension, group-level dimensionality reduction (to target the 34 components estimated by the MDL criterion), fastICA1 ( Hyvarinen, 1999 ) for estimation of independent spatial components, and GICA3 backprojection for the individual subject-level spatial map estimation (see Calhoun et al. 2001 , CONN’s official website https:// web.conn-toolbox.org/home and/or Whitfield-Gabrieli and Nieto- Castanon, (2012) for detailed information). 156 B. Varela-López, Á.J. Cruz-Gómez, C. Lojo-Seoane et al. / Neurobiology of Aging 117 (2022) 151–164 Three components of interest were identified by the combination of visual and mathematical criteria examined with the spatial involvement measure implemented in the ICN_Atlas toolbox ( Kozák et al., 2017 ) and using the networks (including the main brain networks) described by Laird et al. (2011) as templates. The second level analysis included the individual maps of the components of interest (DAN, left FPCN and right FPCN). 2.5. Seed-based connectivity analysis We also performed a weighted SBA to compare the functional connectivity of the regions showing differences in maximum connectivity in the group-ICA analysis, with the intention of shedding some light on the origin of the differences observed at the network level (ICA). We seeded the regions that showed the greatest significant differences using 10 mm spheres created through CONN toolbox. For the FPCN results, we created a seed around the left pars opercularis ([-46, 8, 14] cluster-derived coordinates) and another one around the left pars orbitalis ([-4 4, 4 4, -6] cluster-derived coordinates). For the DAN results, one seed was created in the left inferior temporal gyrus ([-54, -50, -18] cluster-derived coordinates). Weighted Seed based connectivity maps are used to characterize task-specific or condition-specific functional connectivity strength ( Whitfield-Gabrieli and Nieto-Castanon, 2012 ). We applied a GLM hemodynamic response function (HRF) weighted to the first level by using the standard bivariate correlation measure for functional connectivity analyses. Individual seed-to-voxel functional connectivity maps were created for each participant and seed (pars orbitalis, pars opercularis and inferior temporal gyrus). The mean BOLD time series was calculated across all voxels within each ROI, and Fisher Z transformation was applied to correlation values between each pair of sources. The second-level analysis was formed by the individual seed-to-voxel maps evoked for the seeds during the resting condition. As the SBA method can assess both intra-network and internetwork connectivity, the atlas of 7 major networks constructed by Yeo et al. (2011) was used as reference for comparison of the SBA results. 3. Results 3.1. Independent component analysis A total of 34 components were computed (according to the MDL criterion) using ICA. The respective spatial maps of the FPCN and DAN networks derived from the 1 sample analysis are shown in Fig. 1 . Significant differences were observed in the DAN (see Table 2 and Fig. 2 A). Specifically, relative to the CR- group, the CR + group exhibited lower connectivity, for a cluster mainly located in the left inferior temporal gyrus (ITG). ITG connectivity values within the DAN were negatively correlated with years of formal education, the vocabulary subscale of the WAIS-III, immediate, short term and long-term free memory recall scores in a verbal learning test (CVLT), total score of the CAMCOG-R, and with its subscales of executive function, and memory (see Table 5 ). Regarding the FPCN, greater functional connectivity was observed in the CR + group than in the CR- group, for 2 clusters located in the left ventrolateral prefrontal cortex (see Table 2 and Fig. 2 B). The largest cluster was composed by the posterior part of the left ventrolateral prefrontal cortex (pars opercularis, BA 6/44), the posterior left pars triangularis (IFGtriang), the left insular cortex and the left rolandic operculum. This cluster was positively correlated with years of formal education (see Table 5 ). The second cluster, located in the mid-ventrolateral prefrontal cortex, mainly involved the left pars orbitalis and triangularis in the IFG, extending to a small portion of the left middle frontal gyrus (MFG). The second cluster was also positively correlated with years of formal education (see Table 5 ). 3.2. Seed-based post-hoc analysis A post-hoc seed-based analysis was carried out to determine the origin of the significant differences indicated at the network level. With this aim, significant clusters were seeded with spheres located around regions that showed the maximum significant differences in the ICAs, with the intention of performing a seed to voxel analysis. One-sample t-tests results for each seed can be observed in Fig. 3 . Compared to the network atlas reported by Yeo et al. (2011) , the left ITG seed evoked a network consistent with the DAN, whereas the seed in the pars orbitalis elicited a network consistent with the FPCN. On the other hand, the seed in the pars opercularis reflected a more diffuse network, including regions of the dorsal attention network (DAN), fronto-parietal control network (FPCN), ventral attention network (VAN), somatomotor network (SMN), visual network (VN), as well as a negative connectivity pattern in regions belonging to the default mode network (DMN). For the left ITG sphere (see Table 3 and Figs. 3 A and 4 A), relative to the CR- group, the CR + group showed lower functional connectivity between the left ITG and two clusters: one comprising the right inferior temporal/inferior occipital gyri (ITG/IOG), showing overlaps of 148 voxels with the visual network and of 59 voxels with the DAN defined by the altas constructed by Yeo et al. (2011) , and another cluster located in the left superior parietal/inferior parietal gyri (SPG/IPG) along the left intraparietal sulcus (IPS), in this case showing an overlap of 168 voxels with the DAN defined in the aforementioned atlas. The connectivity values between the left ITG and right ITG/IOG was negatively correlated with years of formal education (see Table 5 ). The connectivity between the left ITG and the left SPG/IPG was negatively correlated with years of formal education and with the memory CAMCOG-R subscale (see Table 5 ). Regarding the left FPCN network (see Table 4 ), 2 spheres were created to seed the 2 significant clusters observed in the group- ICA (IFGoperc and IFGorb). For the IFGoperc sphere, the CR + group showed greater functional connectivity than the CR- group, with a large cluster covering parietal and occipital regions, mainly with the superior parietal gyrus (SPG), inferior parietal gyrus (IPG), precuneus (PCUN), middle occipital gyrus (MOG), superior occipital gyrus (SOG) and angular gyrus (ANG). This large cluster overlaps with parieto-occipital regions belonging to the DAN (752 voxels) and parietal regions of the FPCN (232 voxels), according to the network atlas of Yeo et al (2011) (see Fig. 3 B and 4 B). The connectivity between these regions was positively correlated with years of formal education (see Table 5 ). The IFGorb sphere, corresponding to the left mid-ventrolateral prefrontal cortex (BA 45/47), showed higher connectivity for the CR + group than for the CR- group, mainly at the posterior part of the pars triangularis (IFGtriang), with an overlap of 230 voxels with the FPCN network ( Yeo et al., 2011 ) (see Fig. 3 C and 4 C). The connectivity was significantly correlated with years of formal education and with scores on the executive function CAMCOG-R subscale, the vocabulary subscale of the WAIS-III and with the global scores on the BDAE (see Table 5 ). 4. Discussion The present study explored the resting state activity of two networks that were found to play an important role in modulating brain activity related to the level of cognitive reserve (CR), B. Varela-López, Á.J. Cruz-Gómez, C. Lojo-Seoane et al. / Neurobiology of Aging 117 (2022) 151–164 157 Table 2 Significant ICA results obtained by the two groups comparison (CR + > CR-) Cluster size L/R Brain regions # Voxels in specificregion (overlap %) MNI Coordinates (x,y,z) t-value Significant results for the DAN extracted from ICA. 324 L ITG 248 (8) -68 -48 -16 -4.59 Significant results for the left FPCN extracted from ICA. 551 L IFGoperc 281 (27) -46 8 14 5.01 L IFGtriang 81 (3) L INS 80 (4) L ROL 75 (8) 342 a L IFGorb 114 (14) -44 44 -6 5.64 L MFG 93 (2) L IFGtriang 84 (3) L OFClat 29 (15) Results are significant at p < 0.05 Family-Wise Error (FWE) cluster-corrected in a combination with a threshold of p < 0.001 at the uncorrected voxel level. Only brain regions with > 1% cluster overlap were presented. Key: Cluster size, numbers of voxels in each cluster; IFGoperc, pars opercularis; IFGorb, pars orbitalis; IFGtriang, pars triangularis; INS, insula; ITG, inferior temporal gyrus; L/R, Left or right hemisphere; MFG, middle frontal gyrus; MNI, Montreal neurological institute coordinates; OFClat, lateral orbital gyrus; ROL, rolandic operculum. a Also significant at peak p-FWE level. Table 3 Significant post-hoc SBA results obtained by the two groups comparison (CR + > CR-) for the ITG seed [-54, -50, -18], extracted from the DAN ICA Cluster size L/R Brain regions # Voxels in specificregion (overlap %) MNI Coordinates(x, y, z) t-value 281 L SPG 143 (7) -20 -62 58 -3.97 L IPG 73 (3) 209 R ITG 159 (4) 46 -64 -12 -4.38 R IOG 32 (3) Results are significant at p < 0.05 Family-Wise Error (FWE) cluster-corrected in a combination with a threshold of p < 0.001 at the uncorrected voxel level. Only brain regions with > 1% cluster overlap were presented. Key: Cluster size, numbers of voxels in each cluster; IOG, inferior occipital gyrus; IPG, inferior parietal gyrus; ITG, inferior temporal gyrus; L/R, Left or right hemisphere; MNI, Montreal neurological institute coordinates; SPG, superior parietal gyrus. Table 4 Significant post-hoc SBA results obtained by the two groups comparison (CR + > CR-) for the FPCN seeds Cluster size L/R Brain regions # Voxels in specificregion (overlap %) MNI Coordinates (x,y,z) t-value Significant results for the left IFGoperc sphere [-46, 8, 14], extracted from the FPCN ICA. 1338 L SPG 391 (19) -12 -74 50 4.80 L IPG 251 (10) L PCUN 239 (7) L MOG 177 (5) L SOG 114 (8) L ANG 18 (2) Significant results for the left IFGorb sphere [-44, 44, -6], extracted from the left FPCN ICA. 280 L IFGtriang 258 (10) -40 32 18 4.51 Results are significant at p < 0.05 Family-Wise Error (FWE) cluster-corrected in a combination with a threshold of p < 0.001 at the uncorrected voxel level. Only brain regions with > 1% cluster overlap were presented. Key: ANG: angular gyrus; cluster size: numbers of voxels in each cluster; IFGtriang: pars triangularis; IPG: inferior parietal gyrus; L/R: Left or right hemisphere; MOG: middle occipital gyrus; MNI: Montreal neurological institute coordinates; PCUN: precuneus; SOG: superior occipital gyrus; SPG: superior parietal gyrus. specifically the dorsal attention network (DAN) (Alexander et al., 1997; Bastin et al., 2012 ; Franzmeier, et al., 2017b ) and frontoparietal control network (FPCN) ( Cole et al., 2012 , Cole et al., 2015 ; Franzmeier et al., 2017c ; Franzmeier et al., 2017d ; Jiang et al., 2020 ). Both groups were matched in a validated index of global brain atrophy, the BV/CSF index ( Orellana et al., 2016 ), indicating that the effects are related to the CR proxy rather than to other mechanisms of preservation or deterioration in aging. Independent components analysis and seed to voxel post-hoc analysis (seeding the significant clusters resulting from the ICA analyses) were carried out. The results revealed that CR + participants showed reduced connectivity in the DAN, but increased connectivity in the FPCN, with both effects correlated with better performance in neuropsychological tests. The study findings therefore suggest that CR allows greater recruitment of anterior regions (FPCN results) in detriment to posterior regions (DAN results), as proposed in the Scaffolding Theory of Aging ( Park et al., 2009 ; Reuter-Lorenz and Park, 2014 ). In fact, the results support the prediction of the theory that life experiences impact the function of the brain, in this case promoting efficient connectivity of neural networks, and therefore affect the cognitive performance by enhancing neural resources enrichment and developing compensatory mechanisms (scaffolding) to cope with life-span related changes ( Reuter-Lorenz and Park, 2014 ). Regarding the DAN, network-level results showed multiple significant (all negative) correlations between the connectivity in the DAN with overall scores of CAMCOG (global, memory, and executive function scores), CVLT memory subscales, the vocabulary subscale of the WAIS, and years of formal education. In addition, activation of the left ITG within the DAN was negatively 158 B. Varela-López, Á.J. Cruz-Gómez, C. Lojo-Seoane et al. / Neurobiology of Aging 117 (2022) 151–164 Table 5 Significant correlations between ICA/SBA results, years of formal education and neuropsychological variables ICA/SBAresults DAN FPCN ITG seed IFGoperc seed IFGorb seed MNI Coordinate (x,y,z) MNI Coordinate (x,y,z) MNI Coordinate (x,y,z) MNI Coordinate (x,y,z) MNI Coordinate (x,y,z) MNI Coordinate (x,y,z) MNI Coordinate (x,y,z) -68 -48 -16 -46 8 14 -44 44 -6 -20 -62 58 46 -64 -12 -12 -74 50 -40 32 18 Years of formal education rho = - 0.434 b rho = 0.473 b rho = 0.428 b rho = - 0.431 b rho = - 0.388 a rho = 0.430 b rho = 0.434 b MMSE –––––– – CAMCOG-R Total rho = - 0.544 c ––––– – CAMCOG-R Attention- Calculation –––––– – CAMCOG-R Memory r = - 0.402 a ––r = - 0.419 a –– – CAMCOG-R Executive functions rho = - 0.404 a ––––– rho = 0.387 a CVLT Immediate Free Recall r = - 0.433 b ––––– – CVLT Short-Delayed Free Recall rho = - 0.517 c ––––– – CVLT Long-Delayed Free Recall rho = - 0.458 b ––––– – WAIS-III Vocabulary rho = - 0.421 a ––––– rho = 0.380 a BDAE –––––– rho = 0.419 a Key: BDAE, Boston diagnostic aphasia examination; CAMCOG-R, Cambridge cognitive examination-revised; CVLT, California verbal learning test; MMSE, mini-mental state examination; r, Pearson’s (parametric) correlation; rho, Spearman’s non-parametric correlation; WAIS-III = Wechsler adult intelligence scale. All p corrected to Bonferroni: a < 0.05 b < 0.01 c < 0.001 B. Varela-López, Á.J. Cruz-Gómez, C. Lojo-Seoane et al. / Neurobiology of Aging 117 (2022) 151–164 159 Fig. 3. One-sample t-tests results evoked by the regions used in seed-based analyses. (A) ITG seed. (B) Pars opercularis seed. (C) Pars orbitalis seed. “(For interpretation of the references to color in this figure legend, the reader is referred to the Web version of this article.)” correlated with memory outcomes. Considered together, and in line with the reserve hypothesis on neural activity ( Cabeza et al., 2018 ), the present results indicate that CR leads to a greater efficiency/effectiveness within DAN, which is consistent with previously reported findings. Thus, Bastin et al. (2012) reported that high CR was related to lower metabolic activity in the left anterior intraparietal sulcus within the DAN in cognitively unimpaired older adults. Similarly, Karim et al. (2019) observed that educational level was negatively associated with left ITG activity. Furthermore, Strenziok et al. (2014) reported that cognitive training was associated with a reduction in the functional connectivity of the right superior parietal cortex and the left inferior temporal cortex within the DAN. More specifically, reduced connectivity within the DAN was found in the group with high CR relative to the group with low CR of left ITG, an area that seems to play an important role in the top-down processing of visual information in the context of the visual working memory operations ( Ranganath, 2006 ; Ranganath and D’Esposito, 2005 ; Hamamé et al., 2012 ; Postle et al., 20 0 0 ) Left ITG presented a reduced connectivity with the left superior and inferior parietal gyri (SPG/IPG), a key region in the DAN, underlying the focus of attention and acting as working memory storage; Cowan, 2011 ), and with the right inferior temporal and occipital gyri (ITG/IOG), which forms part of the Visual Network. Previous studies have revealed that DAN regions modulate the activity of bilateral regions within the Visual Net-