Structural core of the executive control network: A high angular resolution diffusion MRI study

Abstract Executive function (EF) is a set of cognitive capabilities considered essential for successful daily living, and is negatively affected by ageing and neurodegenerative conditions. Underpinning EF performance are functional nodes in the executive control network (ECN), while the structural connectivity underlying this network is not well understood. In this paper, we evaluated the structural white matter tracts that interconnect the ECN and investigated their relationship to the EF performance. Using high‐angular resolution diffusion MRI data, we performed tractography analysis of structural connectivity in a cognitively normal cohort (n = 140), specifically targeting the connectivity between ECN nodes. Our data revealed the presence of a strongly‐connected “structural core” of the ECN comprising three components: interhemispheric frontal connections, a fronto‐parietal subnetwork and fronto‐striatal connections between right dorsolateral prefrontal cortex and right caudate. These pathways were strongly correlated with EF performance (p = .003). Post‐hoc analysis of subregions within the significant ECN connections showed that these effects were driven by a highly specific subset of interconnected cortical regions. The structural core subnetwork of the functional ECN may be an important feature crucial to a better future understanding of human cognition and behaviour.


| INTRODUCTION
Executive function (EF) outlines a central set of cognitive capabilities that are essential to successful daily living. EF refers to a collection of higher order processes that guide thoughts and behaviours towards achieving a specific goal (Niendam et al., 2012). It includes processes such as working memory, inhibition of prepotent responses, attentional control, planning and flexibility of switching between different goals. Deficits in EF are observed across many brain diseases such as schizophrenia, major depressive disorder and Alzheimer's disease (Guarino et al., 2018;Snyder, 2013;Vohringer et al., 2013), and EF performance has been demonstrated to be an important factor influencing outcomes across many medical conditions including chronic cardiovascular and metabolic conditions (Broadley, White, & Andrew, 2017;Eggermont et al., 2012;C. Vincent & Hall, 2015). EF contributes to the coordination of activities across a wide range of cortical and subcortical brain structures that would make them vulnerable to reduced communication efficiency. Investigations into EF using functional MRI (fMRI) experiments have identified the activation patterns of an executive control network (ECN) subserving EF tasks (Niendam et al., 2012). However, the structural circuitry defining and moderating this complex function has not been described. It is therefore critical to develop a better understanding of the neural substrates of EF, including the ECN.
EF is primarily associated with the frontal lobe (Stuss & Alexander, 2000). Our previous work demonstrated a common network in various psychiatric conditions using voxel based morphometry analyses (Goodkind et al., 2015), highlighting an anterior insula/dorsal anterior cingulate-based network which may relate to EF deficits. Following the identification of the ECN in task-based fMRI studies (Nee et al., 2013;Niendam et al., 2012;Osaka et al., 2004;Taylor et al., 2004), it was also isolated as an intrinsic connectivity network in resting-state fMRI studies (Beckmann, DeLuca, Devlin, & Smith, 2005;Damoiseaux et al., 2006;J. L. Vincent, Kahn, Snyder, Raichle, & Buckner, 2008), in which the correlations between BOLD signals arising from regions across the brain reveal patterns of dissociable networks (Seeley et al., 2007).
The patterns associated with EF are broadly distributed. The most commonly described components of the ECN are the prefrontal cortex, frontopolar cortex, anterior cingulate cortex and posterior parietal cortex (Damoiseaux et al., 2006). In addition to the frontal-parietal network, it is now recognised that the cuneus, supplementary motor area, motor-related nodes, cingulo-opercular nodes are also involved (Reineberg & Banich, 2016). Other studies have also identified EFrelated activities in the cerebellum and subcortical nuclei (Habas et al., 2009;Monchi, Petrides, Petre, Worsley, & Dagher, 2001). Shirer and colleagues (Shirer, Ryali, Rykhlevskaia, Menon, & Greicius, 2012) delineated the left and right ECN, consisting mainly of networks between dorsolateral prefrontal (dlPFC) and parietal cortices. The left ECN includes nodes in the left middle and superior frontal gyri, inferior frontal and orbitofrontal gyri, superior and inferior parietal, angular gyri, precuneus, inferior and middle temporal gyri, left thalamus and right crus. In the right ECN, there are nodes located in the right middle and superior frontal gyri, right inferior parietal, supramarginal, and angular gyri, left crus, and right caudate.
Very few diffusion MRI (dMRI) studies have specifically examined the structural WM connections within the ECN. It is often assumed that the functional connectivity within the ECN is likely to reflect underlying structural connectivity. Support for this finding comes from a small number of studies that have demonstrated strong correlation between fMRI and structural networks identified by dMRI (Damoiseaux & Greicius, 2009;Hagmann et al., 2008;Honey et al., 2009;Skudlarski et al., 2008). We demonstrated that poorer EF was associated with decreased WM integrity in the prefrontal cortex, parietal lobe and thalamic projections (Grieve, Williams, Paul, Clark, & Gordon, 2007). A more recent study (Fjell, Sneve, Grydeland, Storsve, & Walhovd, 2017) confirmed that the WM integrity overall, as well as individual tracts, correlates with EF. Specifically, WM integrity of the inferior longitudinal fasciculus (ILF) and that of the temporal part of the superior longitudinal fasciculus (SLF) were found to be most significantly correlated with the performance in the Stroop test, a common measure of EF. However, there have also been findings of strong functional connectivity existing between areas with apparently low or no structural connectivity (Cunningham, Tomasi, & Volkow, 2017).
In this study, we aimed to uncover and evaluate the WM tracts that underlie the structural connectivity between these functionally-defined ECN nodes. This has been facilitated by the recent development of multi-shell and high angular resolution dMRI sequences that allow improved description of white matter anatomy (Callaghan et al., 2018). Using a large, well-described cohort, we applied high-angular resolution dMRI to catalogue the WM structural network of the ECN, and to test explicitly how this relates to EF performance. The demographic and psychometric characteristics of the cohort (N = 140) are summarised in Table 1. The cohort was representative of a normal population, with a WebNeuro's standard 10 and normalised EF scores centred at the population mean . The average age was 41.2 years (SD 15.3), from 18 to 79 years, with a spread concentrated across 3 decades (IQR: 28-53 years), and a slight predominance of females.

| Image acquisition
Diffusion MRI and structural T1-weighted sagittal 3D SPGR MRI data were acquired at Macquarie Medical Imaging at Macquarie University Hospital (Sydney, Australia) as previously described (Grieve et al., 2013). Acquisition was performed using a 3-Tesla GE Discovery

| Neuropsychological assessment
Neurocognitive testing was performed for all subjects within a week of the MRI scan, using the "WebNeuro" standardised computer-based battery of cognitive tests. The WebNeuro battery consists of a series of 12 tests and takes 30-40 min to complete (Silverstein et al., 2007).
Participants were presented with instructions on screen prior to each of the tests with a researcher present to aid where necessary. The tests have been validated against pen-and-paper tests  and have sound test-retest reliability .
This battery reports on four overall cognitive markers: thinking, emotion, feeling and self-regulation each encompassing several variables (14 in total) obtained from one or multiple tasks. Thinking consisted of response speed, impulsivity, attention and concentration, information processing efficiency, memory and executive function; feeling consisted of depressed mood, anxiety and stress; emotion consisted of emotion identification and emotion bias; and self-regulation included negativity bias, emotional resilience and social skills. Participants' height, weight, blood pressure, and pulse were recorded on the day at the testing site.
Specifically, the EF performance of the subjects was assessed by the Maze test, Switch Attention, Verbal Interference, and Go-No-Go tasks in the Webneuro battery (Silverstein et al., 2007). In our analysis, we derived the EF composite score from the raw subscores of these EF-related tasks, using factor analysis accounting for the maximum variance (varimax) amongst raw subscore data. For the Maze test, we included the number of trials completed, completion time, path learning time, overrun errors, and total errors; for the Switch Attention, we included the completion time (digits and letters), average connection time (digits and letters), and number of errors; for the Verbal Interference, we included the errors and reaction time for congruent and non-congruent stimuli, and for Go-NoGo we included the reaction time, the variability of reaction time and false miss errors on Go, the false alarm errors on NoGo, as well as total errors.

| Image processing
The structural T1-weighted MRI images were segmented into white matter (WM), grey matter (GM) and cerebrospinal fluid (CSF) and the total intracranial volume (TIV) was estimated using FreeSurfer software version 6 (Fischl, 2012). The subcortical GM structures were segmented using a model-based method implemented by FSL's FIRST tool (Patenaude, Smith, Kennedy, & Jenkinson, 2011). The T1-weighted images were linearly registered to b = 0 images of the dMRI data (Jenkinson, Bannister, Brady, & Smith, 2002), and the anatomical labels were also transformed to the dMRI space.

| Statistical analysis
In our analysis, we performed a univariate analysis of correlation between EF and connectivity followed by a network-based analysis (NBS) (Zalesky, Fornito, & Bullmore, 2010) Hanggi, 2015). We calculated the effect size in terms of ΔR 2 , namely the difference in R 2 comparing models including and excluding the connectivity measurements as an independent variable.
The statistical significance of the relationship between ECN connectivity and EF performance in terms of p-value was also evaluated.
We excluded from the analysis any connections with less than 1% of the total SIFT2 weight in the ECN network, in order to include only connections with a substantial strength, thus reducing the effect of multiple comparisons.
NBS was used to evaluate the statistical significance of the correlation between EF performance and WM connectivity within the structural network and ECN. In NBS analysis, the connectome variables, namely the connection strength and tract-average FA were used as dependent variable in GLM, with the EF score as the independent variable, and subject's age, sex, years of education, and TIV as covariables.
The statistical significances of topological clusters were determined via permutation testing and corrected for the family-wise error. In NBS, as in the univariate analysis, the connections with strength less than 1% of the total connectivity in ECN were also excluded.

| Exploratory investigation of ECN anatomy
Based on structural connections between ECN nodes, we further investigated connectivity between the anatomical substructures of ECN nodes parcellated by Brainnetome atlas (Fan et al., 2016). For each component in the structural network identified by NBS, we computed the connectivities between the Brainnetome subdivisions, and evaluated their correlation with the EFs by a GLM.

| RESULTS
A linear regression analysis on the EF composite score, which we calculated using factor analysis is shown in Table 2. The results showed strong correlations with age and years of education. The EF performance deteriorates significantly with age, and subjects with more years of education showed higher EF performance.
3.1 | Structural connectivity networks Figure 1 shows the average structural connectivity matrix across the ECN. Table 3  Additional strong connections were seen involving the left anterior dlPFC, bilateral cerebellum, and right parietal lobe, which together accounted for an additional 31.6% of total connections (see connections ranked 5-10 in Table 3).

| Univariate relationship between ECN connection strengths and EF performance
A univariate correlation between performance and connection strength controlling for subject's age, sex, and years of education was performed in order to evaluate the relationships between each component of the ECN. The analysis was only performed for connections with a total connection strength >1% of total inter-node connections and is presented in Table 3, as well as in Figure 2. An exploratory

| Network based statistics
The results of NBS analysis of the relationship between ECN connectivity (SIFT2 weights) and EF performance are shown in  The NBS analysis of the FA-based structural connectivity identified at threshold t = 1.5 a subnetwork in the right hemisphere correlating with EF. It consists of fronto-parietal connection between the parietal lobe and dlPFC, and a fronto-striatal connection between right posterior dlPFC and right caudate (FWE corrected p = .019). The correlation coefficient between the connection strength of this F I G U R E 2 The structural subnetworks in ECN correlated with EF performance. Top row: the major white matter tracts in ECN listed in Table 3. Middle row: the ECN subnetwork with significant positive correlation between EF and structural connectivity measured by connection strength in NBS analysis with t-statistic threshold t = 1.5. From left to right: the ECN nodes and the connections that form the subnetworks correlating with EF, axial and sagittal views of sample tractograms of the ECN subnetworks identified in

| Exploratory analysis of ECN anatomy
The results of our exploratory analysis of ECN substructures were shown in Table 5, in which the major connections between anatomical Brainnetome defined nodes were evaluated for connection strength only. The results confirmed that the univariate correlations with EF of connection strength between anatomical Brainnetome nodes were consistent with those between ECN nodes (Table 3). No single univariate correlation of region-to-region connection strength and EF approached that of the whole sub-network (β = .284). The subregions from the Brainnetome did reveal some considerable regional heterogeneity, however, with the strongest functional correlations typically being more than 100% stronger than the weakest within each node.
The strongest univariate correlation with EF performance in the L Parietal-L posterior dlPFC node was between the inferior frontal junction and caudal area 40 (β = .18, ΔR 2 = 0.026, versus β = .09, ΔR 2 = 0.006 for the weakest connection within these regions). Similar heterogeneity was seen in the L posterior dlPFC-R posterior dlPFC node (β = .17, ΔR 2 = 0.028 for the L dorsolateral area 8-R ventrolateral area 8, versus β = .10, ΔR 2 = 0.010 for the weakest connection within this), and to a lesser degree in the other nodes.

| DISCUSSION
Our study reveals the presence of a "structural core" in the ECN, providing convergent evidence linking components of the functionallydefined ECN with structural network strength. This network was defined by both connection strength and integrity (as measured by FA), and involved the bilateral dlPFC, fronto-parietal network, and the right caudate. In a cognitively normal cohort, we that found that the connection strength of this network significantly correlated with the overall performance of EF-related tasks. We suggest that this "structural core" network may represent the static architecture from which the dynamic functional connectivity underlying EF emerges (Park & Friston, 2013).
We found that the structural connection between bilateral prefrontal cortices formed a key part of the network that is highly correlated with EF performance.
The left and right hemispheres of the brain are connected by the WM of the corpus callosum, which is involved in the shifting and inhibition aspects of EF (Bettcher et al., 2016). The genu of the corpus callosum interconnecting the frontal lobes and the splenium-parietal connections in the right hemisphere have both been found to mediate the ageing effect on task switching performance in EF (Madden, Bennett, & Song, 2009 performance with age may be more strongly predicted by changes in brain structure than by functional connectivity (Fjell et al., 2017). Fjell and colleagues demonstrated WM volume changes and brain connectivity together explained nearly half of the decline in EF, whereas functional connectivity alone explained nearly none. Our results extend these observations by revealing the specific anatomical detail of the circuits that may be most important to this process of "disconnection" with age. In our tractography-based analysis, we normalised the total number of connections across the cohort, thus controlling for the effect of decline of global connectivity with age. The normalised total structural connectivity within ECN did not show significant decline over age (β = −.05, p = .60). Thus, the structural connectivity in our analysis measured network disconnections independent of the global effect from ageing.
In our analysis of structural connectivity in relation to EF, we used consensus data derived from functional MRI to guide our investigation. This prior data enabled us to narrow the scope of tractography based analysis of structural network to a few nodes of ECN that are consistently shown to be involved in EF-related activities. An exploratory analysis performed for the whole brain, without specifically targeting these nodes, did not reveal any further relationships between the strength of connectivity and EF performance, but did reveal strong correlations with general fluid cognition. While not definitive, this supports the view that, for the EF component, the relationship between structural features and function is quite specific to a subnetwork of connections (Madden et al., 2017). This may, however, only be true in the absence of severe dysfunction: previous work has shown correlation between global WM in dMRI and EF performance in a cohort with cognitive decline, but not amongst the normal subjects (Ohlhauser, Parker, Smart, & Gawryluk, 2019).
We measured the structural connectomes on a high angular resolution dMRI dataset with 140-gradient directions with higher accuracy compared to the conventional 64-direction DTI protocol. The structural connectome of normalised connection strength is shown to have higher reliability (Prckovska et al., 2016) than DTI-based measurements such as TBSS (Madhyastha et al., 2014), and using high angular resolution dMRI has shown better longitudinal consistency (Prckovska et al., 2016). We previously showed that these high angular resolution datasets provide an improved delineation of key white matter pathways (Callaghan et al., 2018) including the most detailed model to date of the hippocampal connectome (Maller et al., 2019).
In our analysis, we measured the EF by a single composite score aggregating subjects' performance on various EF-related tasks, namely Maze test, switch of attention, verbal interference, and Go-NoGo.
This may mask the different components of EF measured by these tasks, including abilities to updating working memory, switching between mental sets, and inhibition of prepotent response (Miyake et al., 2000). There are connections between specific subsets of EF related to distinct circuitries (Tekin & Cummings, 2002) which are not investigated in this work.
To conclude, we investigated the structural WM network underlying the functionally defined ECN using tractography-based analysis on high angular resolution dMRI data. We identified a WM network comprising the fronto-parietal SLF tracts between dlPFC and the parietal lobe, the corpus callosum between bilateral dlPFC, and the fronto-striatal connection between right dlPFC and caudate. This formed a structural network at the core of the functional ECN, and the structural connectivity of this network significantly correlated with EF performance. Alongside previous studies, there is convergent evidence for this structural core subnetwork of the functional ECN that may be crucial to our future understanding of higher cognitive function.

ACKNOWLEDGMENTS
SMG acknowledges the support of the Parker-Hughes Bequest, the Frecker Bequest and the Heart Research Institute.

CONFLICT OF INTERESTS
SB acknowledges speaking honoraria from Novartis, F. Hoffmann La Roche and the Alzheimer's Association not related to the work presented here.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.