White matter tracts and executive functions: a review of causal and correlation evidence

Abstract

Executive functions are high-level cognitive processes involving abilities such as working memory/updating, set-shifting and inhibition. These complex cognitive functions are enabled by interactions among widely distributed cognitive networks, supported by white matter tracts. Executive impairment is frequent in neurological conditions affecting white matter; however, whether specific tracts are crucial for normal executive functions is unclear. We review causal and correlation evidence from studies that used direct electrical stimulation during awake surgery for gliomas, voxel-based and tract-based lesion-symptom mapping, and diffusion tensor imaging to explore associations between the integrity of white matter tracts and executive functions in healthy and impaired adults. The corpus callosum was consistently associated with all executive processes, notably its anterior segments. Both causal and correlation evidence showed prominent support of the superior longitudinal fasciculus to executive functions, notably to working memory. More specifically, strong evidence suggested that the second branch of the superior longitudinal fasciculus is crucial for all executive functions, especially for flexibility. Global results showed left lateralization for verbal tasks and right lateralization for executive tasks with visual demands. The frontal aslant tract potentially supports executive functions, however, additional evidence is needed to clarify whether its involvement in executive tasks goes beyond the control of language. Converging evidence indicates that a right-lateralized network of tracts connecting cortical and subcortical grey matter regions supports the performance of tasks assessing response inhibition, some suggesting a role for the right anterior thalamic radiation. Finally, correlation evidence suggests a role for the cingulum bundle in executive functions, especially in tasks assessing inhibition. We discuss these findings in light of current knowledge about the functional role of these tracts, descriptions of the brain networks supporting executive functions and clinical implications for individuals with brain tumours.

superior longitudinal fasciculus, cognitive control, diffusion tensor imaging, awake surgery, voxel-based lesion-symptom mapping, tract-based lesion-symptom mapping

Introduction

Complex and flexible cognitive processes are underpinned by widely distributed neural systems in the human brain,^1-3 connected by white matter tracts that support structural connectivity, thus enabling information transfer among brain areas and network integration. The central role of white matter in human behaviour and learning has been demonstrated in studies showing clear associations between its abnormalities and a wide range of cognitive deficits, from basic information processing speed to high-level processes, such as executive functions.^4-6

Interest in white matter function has been increasing as conceptions of brain functioning have evolved from localizationist to network models. In localizationist models, brain functions are associated with and attributed to the activity of specialized brain areas,⁷ whereas current conceptions describe brain functions emerging from flexible, dynamic and parallel interactions among distant and hierarchically organized networks of cortical and subcortical zones.² Descriptions of brain architecture and networks^8-10 have provided an account of brain network organization, properties and function not only at the cortical level but also in terms of subcortical connectivity.^11,12 More than just connectors, myelinated fibres reinforce the speed of conduction and the connectivity of the white matter tracts allows the rapid synchronization of information among cortical regions.^2,11,13 In other words, white matter tracts support synaptic function underlying information processing and connect remote grey matter zones into communicating neural assembles. In this vein, current neuroanatomical models explaining brain dysfunction combine the roles of cortical brain zones and axonal connections in a ‘hodotopical’ approach.¹⁴ According to this theoretical framework, brain dysfunction may result from ‘topological’ mechanisms (i.e. cortical lesions) or ‘hodological’ mechanisms (i.e. lesions in axonal pathways underlying connections among brain zones).

In light of these descriptions, disconnection mechanisms have been hypothesized to explain cognitive dysfunction associated with brain pathologies affecting white matter, such as traumatic brain injury,^15-17 multiple sclerosis^18,19 and cerebrovascular disease.²⁰ Also, disconnection mechanisms and dysfunction of neural networks are thought to be involved in the marked executive and functional impairment associated with neuropsychiatric conditions such as major depression, bipolar disorder and schizophrenia.^21,22 White matter has been much studied in normal ageing processes²³: white matter microstructure damage and subsequent disconnection mechanisms are a central hypothesis explaining the decline or change in cognitive functioning.^24-27 Changes in brain architecture related to glioma growth and subsequent post-lesional neuroplasticity clearly revealed the key role of white matter, particularly regarding the dynamics underlying the flexible functioning and recovery potential of the CNS.^28,29 Diffuse low-grade gliomas develop slowly by infiltrating white matter. This induces neurosynaptic reorganization and recruitment of adjacent or even contralateral zones for functional compensation of damaged areas, leading to significant changes in neural network architecture.³⁰ These changes provide opportunities for optimal tumour resection because resection boundaries can be defined according to individual perioperative responses triggered during direct electrical stimulation (DES),³¹ as long as subcortical connectivity is identified and preserved.^29,32

Impairment of executive functions (Box 1) is indeed a common and perhaps a core deficit in brain conditions with predominant white matter damage.^33-35 Such deficits may severely compromise independence and quality of life because of their impact on professional, social and emotional functioning.³⁶

Box 1

Executive functions: conceptual framework and assessment

Executive functions, or cognitive control functions, are high-level cognitive processes serving goal-directed behaviours and flexible adjustment to the environment, as opposed to habits and automatic information processing. They also underpin behavioural processes such as initiation and emotional control, enabling individuals to regulate their thoughts and actions.^37-39 The broad concept of executive functions encompasses a wide set of cognitive abilities, such as attentional control,^40-42 that is critical for monitoring, maintaining task goals and selecting relevant information in conditions of interference⁴³; working memory, cognitive flexibility, inhibitory control.^40,44 These processes engage simultaneously during complex mental functions, such as abstraction, planning, reasoning and decision-making, in a context-sensitive manner.⁴⁴

The neuropsychological assessment of executive functions involve complex and multifactorial tasks that tap executive and low-level non-executive processes.⁴⁴ According to an influential conceptual framework that used latent variable analysis, some of the most frequent tests share a common variance reflecting a common underlying executive ability (the unity of executive functions, termed Common EF factor), but are still separable in three core executive processes: updating (working memory), shifting between tasks or mental sets, and inhibition of prepotent responses.⁴⁴ The development of the unity/diversity model, further exploring the interrelations and specificities of the three core executive processes, corroborates the specificity of working memory/updating and shifting abilities. However, an inhibition-specific factor could not be extracted after accounting for the Common EF factor, despite the fact that all the tasks employed to assess the three executive constructs load on the Common EF factor.⁴⁵ According to the authors, the Common EF reflects the ability to maintain, manage and leverage goals to influence ongoing processes, a general requirement of all executive tasks that may be central for response inhibition.⁴⁰ This mechanism would explain the absence of an additional inhibition-specific factor and aligns with research suggesting that brain activations underlying top-down inhibition may be reflecting context-monitoring, top-down activation of correct responses. Thus, at the neural level inhibition would emerge as an outcome of the interplay between goal-maintenance and suppression of irrelevant responses.^46,47

There are several conceptualizations and different levels of complexity to describe executive functions.^48-51 The unity/diversity framework⁴⁴ contributed to addressing the problem of task impurity, which helps to guide the choice of constructs and tasks used in research. The three core executive domains have been commonly explored³⁹ in studies investigating the neural correlates underlying this complex and multifaceted cognitive construct.⁵⁰

Working memory

Working memory relates to the ability to transiently process, maintain and use information input from diverse sensorial stimuli to serve ongoing tasks and future goal-directed behaviour. The working memory system is characterized by limited capacity and implies an interaction among different levels of cognitive processing, sensorimotor mechanisms and activation of long-term memory representations.^52,53 Baddeley’s theory conceptualizes a multicomponent model of working memory, a system with distinct but interconnected components allowing for temporary storage and manipulation of visuospatial (visuospatial sketchpad) and verbal information (phonological loop) and a system performing attentional control of action (central executive). The central executive would manage attentional resources, deploying them between visual and verbal stimuli modalities, updating representations and task shifting. A subsystem of the central executive, the episodic buffer, holds multimodal representations and links working memory to perception and long-term memory.^52,54,55

Working memory and updating abilities are assessed with tasks such as digit-spans backwards, Wechsler Adult Intelligence Scale’s letter-number sequencing⁵⁶ and n-back tasks.⁵⁷

Shifting

Our constantly changing environment requires flexible reorientation of our attention from one stimulus or mental representation to another, to adapt our behaviour to the demands of new situations or to solve new problems based on acquired experience. Flexibility is a main component of executive functions and relates to the ability of shifting between tasks or mental sets.^58,59 It encompasses several cognitive processes, such as detecting salient stimuli and focusing on it, retaining relevant information and rules representations, and inhibiting previous responses according to the appropriate strategy.⁶⁰ In other words, the successful completion of tests assessing task or set shifting abilities requires interactions among attentional processes, working memory and to a greater extent, inhibition of salient but unsuitable responses.⁵⁸

The Trail-Making Test B (TMT-B)⁶¹ and the Wisconsin Card Sorting Test⁶² are some of the tasks frequently used for assessing set-shifting abilities in both clinical and research settings. For research purposes, different paradigms allowing the estimation of switch costs between tasks conditions have been developed to assess cognitive flexibility.⁵⁹

Inhibition

Inhibition is a multifaceted concept⁶³ with multiple taxonomies⁶⁴ that encompasses behavioural (inhibition of impulses, unwanted memories and contextually inappropriate behaviours)^39,65-67 and cognitive processes (i.e. inhibition of automatic, prepotent responses).^44,64 It relates to the ability to deliberately suppress dominant, automatic responses and resist interference from environmental cues³⁹ to focus on information that is relevant to the accomplishment of ongoing tasks.

The completion of tasks assessing response inhibition involve other cognitive mechanisms such as working memory, resistance to proactive and retroactive interference,⁶⁴ and especially attentional control.³⁹ Inhibitory control of attention (i.e. interference control at the level of perception) enables stimuli selection and suppression of involuntary attention driven by the properties of the stimuli themselves.^39,43 Such selective control of attention is essential for the maintenance and manipulation of information in working memory³⁹ and flexible responses,^58,64 fostering goal-directed behaviour.

Several tasks and paradigms allow for testing inhibition and inhibition-related abilities.^64,68 In clinical practice, the ability to inhibit a prepotent response can be assessed using tasks such as the Stroop task,⁶⁹ Tower of Hanoi and Tower of London^44,70 and Go-No Go tasks.⁷¹

Neural networks supporting executive functions

Various techniques, such as functional MRI (fMRI), including resting state and activation fMRI, electrophysiology and lesion studies, have provided evidence leading to descriptions of large-scale brain networks involving frontal, parietal and insular regions.^72-75 These regions form distinct neural systems with specific roles in directing the cognitive process toward the achievement of goal-directed behaviours.⁷⁶

According to a review article that introduced a new taxonomy for the large-scale functional networks in the human brain,⁷⁷ three functional networks would be prominently involved in executive functions: the dorsal fronto-parietal (D-FPN), lateral fronto-parietal (L-FPN) and midcingulo-insular (M-CIN) networks. The D-FPN or dorsal attention network, supports visuospatial attention by preparing and exerting top-down selection for stimuli and responses.⁷⁵ It includes the superior parietal lobule extending into the intraparietal sulcus, the middle temporal complex and the putative frontal eye fields as core regions and would involve the ventral premotor cortex.^74,75 The L-FPN supports goal-directed control of information flow.^78,79 It comprises the lateral prefrontal cortex (PFC), along the middle inferior parietal lobule (including rostral and dorsolateral PFC) and the anterior inferior parietal lobule into the intraparietal sulcus as core regions, involving as well the midcingulate gyrus.⁸⁰ It has been described as a domain general system for a wide variety of demanding cognitive tasks.^81,82 Meta-analyses showed convergence of activations for the three executive domains in this system,^83,84 that has been associated with common executive ability^44,45 in studies exploring the unity and diversity of the neural substrates of executive functions.^85,86 Also, the M-CIN, a right lateralized system, is thought to direct attention to spatial locations of salient stimuli and maintain task sets in demanding cognitive tasks. Core regions are the bilateral anterior insulae and anterior midcingulate cortex. Other involved regions are the inferior parietal cortex,⁸⁰ right temporal parietal junction and lateral PFC,^75,79 as well as subcortical structures including the basal ventromedial nucleus of the thalamus.^72,77,87

Regarding executive constructs, meta-analyses using activation likelihood estimation revealed several overlapping and also specific patterns of activations within these large-scale networks during performance of executive tasks. Working memory/updating tasks consistently activate a bilateral network of frontal and parietal regions regardless of the type of stimuli, including bilateral inferior frontal gyri (IFG), bilateral presupplementary motor areas (pre-SMA), bilateral intraparietal sulci and bilateral anterior insulae, described as the ‘core’ working memory network.⁸⁸ Task-switching paradigms also activate widely distributed networks depending on task features but the medial PFC, lateral PFC, especially the inferior frontal junction,^58,60 ventrolateral PFC, posterior parietal cortices, insula and anterior cingulate cortices are described as domain-general regions for flexibility.^58,60,89-91 Finally, meta-analysis of domain-general regions for tasks assessing inhibition identified a right-lateralized network of cortical regions including the IFG, insula, median cingulate and paracingulate gyri and superior parietal lobule⁶⁸ that are consistently active across studies using different tasks. Other previously described core nodes, such as the pre-SMA and subcortical structures,^{27,63,67,92-94} were engaged in resolving interference (i.e. the Stroop task) and in response inhibition requiring action withholding (i.e. Go-No Go tasks).⁶⁸

In light of these findings, there is growing interest in investigating white matter structures underlying interactions among these networks in the healthy and diseased brain. Recent years has seen an increase in knowledge of the functional role of white matter tracts in motor, language, visuospatial processing and other cognitive processes, according to results from lesion studies and the neuro-oncology field.^28,95-97 The crucial role of the connectivity underlying the FPN for executive functions has been demonstrated in diffusion MRI and voxel-based lesion-symptom mapping studies.^26,98,99 However, studies exploring relations between white matter integrity and cognition usually report different groups of white matter bundles associated with performance on one or more cognitive tasks,¹⁰⁰ providing heterogeneous results. We still do not know whether specific tracts can prominently support executive functions (i.e. tracts consistently associated with executive functions across tasks, populations and according to different levels of evidence).¹⁰⁰

The objective of this review was to identify the white matter tracts consistently associated with performance on tasks assessing executive functions, according to results from studies using DES, lesion mapping and diffusion MRI techniques (Box 2). We focused on the three constructs defined in the unity/diversity of executive functions model, commonly assessed in clinical practice^36,44 and research,^83,84 namely working memory/updating, set-shifting and inhibition of prepotent responses.^40,44

Box 2

Techniques for the identification of white matter tracts and assessment of white matter damage

Causal evidence: direct electrical stimulation and voxel-based lesion symptom mapping studies

Intraoperative brain mapping and lesion studies provide causal evidence of relationships between brain structure and function. Intraoperative brain mapping, performed during awake surgery using direct electrical stimulation (DES), has greatly contributed to understanding brain processing, especially the role of white matter tracts in cognition and behaviour.¹⁰¹ Generally used during resection of brain tumours such as gliomas, DES to exposed cortical and subcortical brain regions while the patient is performing different cognitive tasks allows for identifying brain areas whose stimulation induces transitory disruption of performance, providing real-time anatomo-functional correlations.³¹ Along with lesion-induced reorganization of functional areas (brain plasticity), intraoperative brain mapping allows for maximal tumour resection while sparing cognitive functions even for lesions located near or within highly eloquent regions.¹⁰² Beyond its undeniable surgical interest, DES remains the only technique that allows for direct in vivo mapping of human white matter tracts.^31,101,103 However, the physiology of DES at the cellular and cortical levels is not completely understood^101,104 and some limitations regarding the types of responses elicited by the stimulation must be considered. For example, DES may induce different behavioural responses at the same cortical region, or false negatives, which may occur particularly when the task is not rigorously chosen.^104,105 These limits are intrinsic to the complexity of the brain’s functioning and network architecture, because complex responses are rarely supported by one single brain region.³¹ Perhaps one of the most challenging aspects of intraoperative brain mapping is that complex cognitive tasks routinely administered in standard neuropsychological assessments are not feasible during awake surgery, which may limit the mapping of higher-order cognitive processes.¹⁰¹ Accurate results depend on the most rigorous control of the technical setting (i.e. appropriate tasks, intensity and duration of stimulation, neuroimaging protocols) and the closest observation of the patient’s overall state, notably their fatigue, which will indeed guide the entire procedure.^104,105

Lesion mapping provides a powerful tool for identifying brain regions that are critical for a given cognitive task. In lesion studies using voxel-based lesion-symptom mapping, inferences are based on statistical analyses of the relation between tissue damage and brain function on a voxel-by-voxel basis, taking into account individual variability of spatial location of lesions, thus without constraints related to spatial resolution.¹⁰⁶ However, several aspects need to be considered to avoid biased interpretations. Deficits may occur not only because a given brain region is damaged but also from disconnection mechanisms, which implies that structurally healthy regions may have their function impaired. Moreover, brain plasticity may induce functional reorganization of brain regions in response to damage.¹⁰⁷ For the analysis of white matter tracts, the output statistical map, which reflects the degree of statistical association between lesioned voxels and behavioural performance, is typically overlaid by a white matter atlas, which allows for making inferences on the contribution of tracts to a given symptom/function.¹⁰⁸ Results must be interpreted with caution because voxel-based symptom mapping approaches are ill conceived to provide biologically plausible anatomo-functional correlations with white matter pathways. Other neuropsychological evidence is derived from tract-based lesion-symptom analyses or disconnection-symptom mapping that more directly correlate measures of disconnection severity to behavioural performance, at the level of the tract (e.g. lesion load, probability of disconnection) or the level of the voxel (e.g. whole-brain disconnection maps).^108-113

Correlation evidence: neuroimaging techniques

MRI techniques can provide rich and complementary information to probe white matter microstructural integrity. Diffusion-weighted sequences are particularly useful in providing quantitative measurements characterizing the diffusion properties of water molecules, which can be linked to alterations of the microstructure. Additionally, these sequences help to recover the geometry of the major white matter bundles using tractography algorithms.¹¹⁴

Diffusion-weighted acquisitions consist of several images characterizing the diffusion properties in several directions (i.e. diffusion gradient sampling scheme) and according to potentially various diffusion weighting (i.e. b-values). Diffusion tensor imaging (DTI) is one of the most used techniques, relying on the assumption that diffusion properties can be modelled by a six-parameters tensor (i.e. a symmetric 3 × 3 matrix), from which several DTI metrics can be derived. Fractional anisotropy (FA) is a measure ranging from 0 to 1 that quantifies diffusion anisotropy¹¹⁵; regions with high FA values correspond to more organized and aligned neural pathways. The mean diffusivity (MD) reflects the average magnitude of water diffusion regardless of direction. Pathological processes that disrupt the integrity of white matter tracts are often associated with increased MD and decreased FA.¹¹⁶ Axial diffusivity (i.e. diffusivity of water molecules along the principal axis of white matter fibres) and radial diffusivity (i.e. diffusivity of water molecules perpendicular to the principal axis of white matter fibres) are complementary measures that help to more finely characterize the microstructural properties of brain tissue, especially for investigating the integrity of the myelin sheath or the axonal membrane.¹¹⁴ DTI techniques present some limitations regarding the identification of ‘kissing’ or ‘crossing’ fibres.¹¹⁷ More involved diffusion model can be considered such as multi-compartment model, which can be used to disentangle intracellular from extra-cellular diffusion properties, diffusion kurtosis imaging (DKI), which expands on the standard DTI model by incorporating information about non-Gaussian diffusion, or methods based on the estimation of orientation distribution function,^117-119 allowing for the reconstruction of multiple fibre orientations.¹¹⁴ The use of such diffusion models requires diffusion weighted sequences with a high number of diffusion gradient directions, typically ranging from 30 to 100, and with several b-values, resulting in a quite long acquisition duration, which limits their use in research settings.¹²⁰

Tractography algorithms are computational methods that aim at reconstructing the geometry of white matter tracts in the brain from diffusion weighted MRI acquisition. There are several tractography algorithms available, including deterministic and probabilistic methods. Deterministic methods, generate tractography by following the direction of maximum diffusion along the estimated fibre tract. These algorithms generate a single fibre trajectory for each seed point. Probabilistic tractography algorithms generate a probability distribution for the possible fibre pathways, rather than a single trajectory.¹¹⁴

Literature search and methods

We searched PubMed for studies published from 2009 to 2023 that explored associations between executive functions and integrity of white matter tracts in healthy and cognitively impaired human adults. The searches were conducted using general and specific terms related to executive processes and techniques used to explore white matter. Executive tasks commonly used in clinical practice and research were included in the search terms, to retrieve studies that explored general cognition or used composite scores. We also included attentional processes to the search terms, to retrieve studies that explored attention using executive tests. The following terms were used: ‘anti-saccade’, ‘attentional control’, ‘attentional set shifting’, ‘cognitive control’, ‘executive functions’, ‘flexibility’, ‘inhibition’, ‘inhibitory control’, ‘interference control’, ‘selective attention’, ‘set shifting’, ‘sustained attention’, ‘task switching’, ‘updating’, ‘working memory’, ‘digit spans’, ‘n-back’, ‘Go No Go’, ‘Stop signal task’, ‘Stroop task’, ‘Trail Making Test’, ‘Tower of London’, ‘Wisconsin Card Sorting Test’, combined with ‘awake surgery’, ‘diffusion tensor imaging’, ‘tract-based lesion-symptom mapping’, ‘voxel-based lesion-symptom mapping’ and ‘white matter’. We selected studies and reviewed references of included studies that could be relevant (Fig. 1). The procedures used in each study to avoid bias and validate their methods were rated according to the PRISMA guidelines,¹²¹ using a checklist adapted from previous work¹²² (Supplementary material). Briefly, we used the following basic criteria to rate the studies: sufficient description of population sources and report of confounders, sufficient detail of the protocol for reproducibility, type of task and task parameters used for studying associations with white matter parameters. Studies included in the main results provided sufficient anatomical description/delineation of white matter tracts.

Figure 1

Flow diagram¹²¹: selection of studies of correlations of white matter tracts with executive functions.

Results

In the following sections, we summarize findings according to different levels of evidence, from causal to correlational. For each executive process, we present convergent findings for different white matter tracts, from lesion analyses and DES during awake surgery, followed by results from diffusion MRI studies of healthy and clinical populations. For neuroimaging evidence, we focused on studies using whole brain voxel wise analyses; studies using anatomical regions of interest (ROI) were selected according to methods and rationale.

Working memory

Causal evidence

Results from lesion studies and awake surgery have provided strong evidence for the role of the superior longitudinal fasciculus (SLF), connecting the superior parietal to superior frontal lobes longitudinally along the dorsal premotor and dorsolateral prefrontal regions¹²³ in working memory. Kinoshita et al.¹²⁴ used voxel-based lesion-symptom mapping, DES and a visuospatial two-back task paradigm to assess visuospatial working memory retrospectively and at least 6 months postoperatively in patients who underwent craniotomy under general anaesthesia for right prefrontal glioma. Compared with healthy controls and two patients who had undergone awake surgery, the performance of patients operated under general anaesthesia at follow-up assessment was significantly lower and correlated with lesions in the region overlapping the SLF-I (dorsal) and II (middle) parts.¹²⁵ These results agree with several findings pointing to a major role for the SLF in supporting interactions among a right-lateralized system underlying visuospatial attention^75,126-128 within the D-FPN.⁷⁷ According to descriptions of functional differentiation of its segments,¹²⁸ the SLF-I would be involved in the voluntary orienting of spatial attention (i.e. the top-down, dorsal attention network)⁷⁵ and the SLF-II would modulate communication between the dorsal and ventral attention network, a system that directs attention to spatial location of salient stimuli.^75,80

In a study of 29 patients undergoing awake surgery for removal of neoplastic lesions in the left hemisphere, those performing the digit spans task while receiving DES in the left SLF-III (ventral part) made order errors rather than item errors, but no language deficits were elicited.¹²⁹ The authors propose that the SLF-III would underlie the transferal of order information from the phonological store of the inferior parietal lobule to Broca’s area, in other words, supporting the working memory phonological loop.⁵⁵

Verbal working memory deficits have been associated with lesions in a crossroad of white matter pathways comprising the SLF (primarily branch 1), the frontal aslant tract (FAT) and fronto-striatal tracts, in the left hemisphere.¹³⁰ Connecting the inferior frontal gyrus with the anterior cingulate cortex and medial regions of the superior frontal gyrus (i.e. pre-SMA and SMA), the FAT has only recently been described,¹³¹ although previously identified in models of inhibitory control.¹³² Evidence from DES and clinical studies points to a prominent role of the left FAT in language and in speech initiation.^133-135 In the right hemisphere, its support of executive functions has been hypothesized because it underlies a right-lateralized system involved in cognitive control including the inferior frontal gyrus, the pre-SMA and subcortical structures^132,134,136 (i.e. regions integrating the L-FPN).⁷⁷ One case report of awake surgery described the patient’s inability to cite digits backwards during DES of the right FAT, whereas digit spans forward were normally performed.¹³⁷ Associations with working memory would be possible given the central role of the pre-SMA and SMA in verbal and non-verbal working memory tasks.^88,138,139

Concerning the fronto-striatal tract (also called the subcallosal fasciculus), connecting the SMA to the caudate nucleus,¹⁰² causal evidence showed its support of motor inhibition¹³⁵ and, in the left hemisphere, speech initiation.^102,135,140 Considering these findings and the role of the caudate in subcortical-cortical modulation of goal-directed cognition and action,^136,141,142 the fronto-striatal tracts could have at least an indirect role in the flexible manipulation of verbal information in working memory.

Correlation evidence

In studies involving whole brain analyses of healthy individuals across ages^143-149 and in clinical populations,^150,151 the integrity of the different segments of the corpus callosum has been associated with working memory, assessed with different tasks. This was confirmed in a study including a large group of participants and controlling for multiple factors modulating performance, including processing speed.¹⁵² The corpus callosum genu, connecting symmetrical dorsolateral prefrontal regions,¹²³ has been most consistently associated with working memory performance, in studies exploring age-related decline in working memory performance.^{143-145,147,148}

The integrity of the SLF has been strongly correlated with working memory capacity, including in large samples of healthy individuals.^147,153 In analyses exploring brain networks underlying working memory, higher fractional anisotropy (FA) in bilateral SLF modulated both performance at a high load of working memory capacity and brain–blood oxygenation responses within the FPN.¹⁵⁴ It has also supported better reconfiguration of networks from resting state to performance conditions, contributing to better n-back performance.¹⁵⁵ Other studies exploring the effects of working memory training on white matter plasticity using visual and verbal tasks have shown gains in performance in healthy young and older adults associated with decreased mean diffusivity (MD) (i.e. increased myelination) in the right SLF,^156,157 right SLF-I and left SLF-II.¹⁵⁶ Similar associations have been reported in clinical populations,^{150,151,158,159} some based on diffusion kurtosis imaging (DKI) metrics.¹⁵⁸ In other analyses, activations of cortical networks involved in working memory supported such findings.¹⁵¹ Moreover, DTI metrics in the arcuate fasciculus (AF) have been correlated with verbal working memory performance, in a large sample of healthy ageing individuals¹⁴⁷ and in individuals with traumatic brain injury.¹⁵⁰ The arcuate fasciculus is a lateral bundle lying close to the SLF-III, connecting the perisylvian cortex of the frontal, parietal and temporal lobes.¹⁶⁰

The number of diffusion tensor imaging (DTI) studies presenting results for the cingulum bundle, connecting the cingulate cortex to numerous brain regions through short and long fibres,¹⁶¹ is limited. However, working memory performance has been correlated with the integrity of this tract in large samples of healthy ageing individuals.^145,162 One study showed the involvement of the bilateral anterior and posterior parts of the cingulum.¹⁴⁵ In smaller populations, analyses of training-induced plastic changes within fronto-parietal regions showed a significant effect on both working memory capacity and integrity of parahippocampal segments of the left cingulum, suggesting its support of working memory.¹⁵⁶

Despite a hypothesized role of the FAT in executive functions,¹³⁴ studies using whole brain analyses did not report correlations. In ROI analyses restricted to language-associated pathways, fractional anisotropy in the bilateral FAT has been correlated with performance in working memory in healthy ageing individuals.¹⁵³ Also, in a study exploring the functional differentiation of the FAT in a larger sample, fibre density of its anterior right segments has been associated with n-back performance,^163,164 with left segments correlated with language.¹⁶⁴

Less consistently, other associative and projection tracts have been correlated with performance on n-back tasks in clinical populations, such as the fornix,¹⁵⁰ connecting the medial temporal lobe to the mammillary bodies and hypothalamus and belonging to the limbic system¹⁶⁰; the inferior longitudinal fasciculus (ILF),¹⁵¹ a ventral associative bundle connecting the occipital and temporal lobes; and the inferior fronto-occipital fasciculus (IFOF),¹⁵¹ connecting the ventral occipital lobe and the orbitofrontal cortex.¹⁶⁰

Results from studies of large groups of healthy individuals suggest the involvement of fibres supporting cortico-subcortical networks in working memory performance. Some showed significant correlations with DTI metrics in the internal capsule (IC),^143,152 mean diffusivity¹⁴⁷ and DKI¹⁴⁹ indexes in the left superior and posterior corona radiata (CR). The internal capsule and corona radiata contain ascending fibres from the thalamus to the cerebral cortex and descending fibres from the fronto-parietal cortex to the subcortical nuclei.¹⁶⁰ Other studies showed significant correlations with mean diffusivity and fractional anisotropy in the external capsule (EC),^145,165 transporting fibres from prefrontal, temporal, SMA and pre-occipital regions to the caudate and putamen. Along with low integrity of the external capsule, low performers of n-back tasks had low connectivity within the FPN and between the FPN and the striatum.¹⁶⁵ Moreover, in whole brain analyses using sensitive models to detect structural changes in white matter microstructure, the integrity of the posterior thalamic radiation (TR), carrying fibres between the thalamus and the PFC, was significantly correlated with working memory performance in individuals with psychiatric disorders.¹⁶⁶

Summary

Taken together, evidence supports the prominent role of the corpus callosum and the SLF in working memory. According to causal evidence, the left SLF-I would be associated with verbal working memory¹³⁰; more specifically, the left SLF-III would support the working memory phonological loop.¹²⁹ In the right hemisphere, damage to the SLF-I and II induced impairment in visuospatial working memory.¹²⁴ Diffusion MRI studies have shown converging results in different populations. Although most DTI studies do not provide details on the segments of the SLF, causal and correlation evidence indicate left lateralization for verbal and right lateralization for visual working memory.^124,130,157

According to correlation evidence in large populations, the bilateral cingulum bundle, underlying connections among salience nodes that integrate the M-CIN,^77,87 would support working memory performance.^145,156,162

Causal evidence has shown that the disruption of the FAT would affect verbal working memory performance, with results for the left¹³⁰ and the right hemisphere.¹³⁷ Diffusion MRI evidence supporting these findings is scarce, based on ROI analyses and provided results for both hemispheres.^153,164 Likewise, causal evidence indicates a potential support of the fronto-striatal tract to a network involved in verbal working memory in the left hemisphere, along with fronto-parietal connections and the FAT,¹³⁰ however, to our knowledge such an association has not been reported in DTI studies.

Results from diffusion MRI studies included in this review suggest a potential support of projection tracts carrying cortico-subcortical connections to working memory, notably the external capsule and internal capsule, along with the corpus callosum and fronto-parietal connections^{143,145,152,165} (Fig. 2 and Supplementary Table 1).

Figure 2

Evidence of white matter tracts involvement in executive functions. Causal and correlation evidence are represented for each executive process and white matter tracts. The colours of the circles represent the total number of participants included in each study; the width of the circles represent the number of included studies that reported significant associations (Supplementary Table 1). The white matter tracts were generated from the Human Connectome Project tractography atlas.¹⁶⁷ AF = arcuate fasciculus; CC = corpus callosum; Cing = cingulum bundle; CST = cortico-striatal tract; FAT = frontal aslant tract; SLF = superior longitudinal fasciculus; TR = thalamic radiation.

Set-shifting

Causal evidence

Causal evidence indicated the involvement of the left SLF in performance on the Trail Making Test B (TMT-B). According to lesion analyses of cholinergic lateral pathways in 106 patients after acute and chronic stroke, deficits were associated with damage to the left SLF.¹⁶⁸ Moreover, according to a case report, postoperative damage to the SLF induced significant deterioration in TMT-B shifting cost after surgical resection of a glioma located in the right hemisphere. Tractography analysis revealed damage to the SLF-III, the arcuate fasciculus and to a lesser extent the middle longitudinal fasciculus and callosal fibres (tapetum); the SLF-I and II seemed relatively spared.¹⁶⁹ According to the authors, deterioration in set-shifting abilities resulted from disruption of a cognitive control network that spreads bilaterally in prefrontal but also cingular, parietal and temporal areas.⁸² Indeed, the flexible switching between numbers and letters necessary for completing the TMT-B also relies on working memory and visuospatial attention^170,171 and, in previous studies, both the SLF-III and arcuate fasciculus have been associated with visuospatial processing.^113,126

Results from the study of Cochereau et al.¹⁴⁰ are slightly different. In this large cohort, long-lasting deficits on the TMT-B switch cost were prominently associated with post-surgical damage to the left SLF-II.

Correlation evidence

Several studies have shown contributions of the corpus callosum to set-shifting abilities, in clinical populations^172,173 and in healthy individuals across ages.^{143,162,174,175} Other studies indicated associations between better DTI metrics in the SLF and better performance on different tasks assessing set-shifting, based on switch costs^172,176,177 and TMT-B completion time.^178,179 More precisely, in large samples of healthy adults, axial diffusivity in the left SLF-II strongly predicted shifting-specific ability.¹⁷⁵ Such associations were not observed for grey matter¹⁷⁵ and remained strong after accounting for global white matter.¹⁷⁷

The anterior and posterior segments of the cingulum were strongly correlated with attention and flexible responses in large groups of healthy ageing adults,^162,180 with no associations found with diffusivity or atrophy in cortical structures.¹⁸⁰ Also, the cingulum, along with the corpus callosum and the SLF, was associated with cognitive flexibility in persons with traumatic brain injury, suggesting its involvement in cognitive control networks.¹⁷²

In two studies of healthy adults that used the TMT-B completion time¹⁷⁸ and switch cost¹⁷⁷ as assessment parameters, the integrity of the left and right IFOF strongly predicted performance, probably reflecting its implication in visually guided behaviour.¹⁸¹

Regarding cortico-subcortical connections, results may vary depending on assessment methods and variables controlled. Analyses of clinical and healthy populations have shown associations between low performance on set-shifting paradigms and low integrity of cortico-subcortical loops with the pre-SMA and superior frontal gyrus, including clusters in the bilateral superior corona radiata, including the internal capsule and posterior thalamic radiation.^182,183 In studies using the TMT-B completion time, better performance has been associated with better DTI parameters in the bilateral superior corona radiata¹⁷⁹; right¹⁸⁴ and bilateral internal capsule.¹⁸⁵ However, in analyses of more than 300 healthy ageing adults, correlations between the integrity of the left anterior thalamic radiation and TMT-B completion time lost significance after controlling for processing speed by means of different tasks.¹⁸⁶

Summary

Causal and correlation evidence have shown consistent associations between integrity of the SLF and set-shifting performance. Severe decline of set-shifting abilities was prominently associated with damage to the SLF-III and arcuate fasciculus in the right hemisphere, but also the middle longitudinal fasciculus and callosal fibres reflecting network disruption.¹⁶⁹ In large populations, causal evidence indicate that the left SLF,¹⁶⁸ and more precisely the left SLF-II,¹⁴⁰ supported better performance. Convergent evidence from diffusion MRI studies in large populations agree with these findings, showing results for the left SLF¹⁷⁷ and left SLF-II.¹⁷⁵ Correlation studies exploring the corpus callosum showed frequent associations, especially with the genu.^{143,172,173,187}

To our knowledge, there is no causal evidence indicating associations between lesions in the cingulum bundle and decline in set-shifting abilities; however, according to correlation evidence it could be involved in networks supporting flexibility.^162,172

Regarding cortico-thalamic and cortico-striatal connections, DTI studies indicate the support of the bilateral superior corona radiata to flexible responses, within cortico-subcortical loops underlying cognitive control^182,183 (Fig. 2 and Supplementary Table 1).

Inhibition

Causal evidence

Puglisi et al.¹⁸⁸ assessed the feasibility of the Stroop task during awake surgery and showed that DES of specific white matter regions beneath the IFG and middle frontal gyrus, anterior to the insula and over the putamen induced task disruption, but patients made no errors in control tasks. In these patients, administration of the intraoperative Stroop task prevented executive deficits 3 months after resection.¹⁸⁸ In other analyses, the authors compared pre- and post-surgical cognitive outcomes of 29 patients undergoing resection of a right frontal glioma, by using DES, lesion-symptom mapping and diffusion tractography. DES of the anterior thalamic radiation and fronto-striatal tracts produced errors during the Stroop test, but results from lesion analyses showed that inhibition was preserved in seven of eight patients with the inferior fronto-striatal tract spared.¹⁸⁹ However, overall differences before and after the surgery only approached significance and showed slower performance rather than errors. Moreover, all patients had lesions in the right hemisphere and were assessed only 1 month after surgery, which may be a short delay estimating long-lasting impairments.¹⁹⁰ In larger groups of patients, with lesions in the right and left hemisphere and at least 3 months after tumour resection, the Stroop interference index (i.e. time to complete the interference minus the colour-naming condition) was predominantly associated with integrity of the SLF-II, to a lesser extent the SLF-III and weakly with fronto-striatal tracts, all left-lateralized.¹⁴⁰

In the case report of Rutten et al.¹³⁷ DES to the right FAT induced errors in the Stroop task in a patient undergoing awake surgery for resection of a right frontal glioma; motor function was not disrupted.

Correlation evidence

Analyses of white matter associations with measures of inhibition (Stroop interference index, Tower of London) have shown widespread correlations in healthy and clinical populations. Several findings include the corpus callosum.^162,191-194 Regarding the fronto-parietal connectivity, studies using the Stroop and Go-No Go tasks have shown significant correlations between better performance and better DTI parameters (fractional anisotropy, axial diffusivity) in both the left and right SLF.^185,195-198

Investigations of limbic-cortical networks and inhibition in small groups of people with psychiatric disorders suggested the involvement of the cingulum, showing DTI metrics of fractional anisotropy and mean diffusivity associated with performance on the Stroop and Tower of Hanoi.^199,200 Convergent findings have been reported in other populations,^162,201,202 with its left posterior parts prominently associated with verbal rules (i.e. Stroop task) and its bilateral anterior parts sensitive to rule violations in the Tower of London task.²⁰¹

Other tracts such as the fornix^185,203 and the uncinate fasciculus,^196,202 bilaterally, have been less consistently associated with performance on tasks assessing inhibition.

Results from studies of healthy and clinical populations using different tasks agree with descriptions of a right-lateralized fronto-subcortical network supporting inhibition^132,136 with convergent results for the right anterior thalamic radiation.^{192,198,203,204} In individuals with psychiatric conditions, executive impairment has been associated with abnormal activation in cortico-subcortical networks,^205-207 and low integrity of the right anterior internal capsule.²⁰⁵ In line with these findings, comprehensive analyses of neural correlates underlying cognitive control deficits have shown reduced fractional anisotropy in the right anterior limb of the internal capsule associated with impaired Stroop interference index, along with abnormal activation of a right lateralized fronto-thalamo-cerebellar network.²⁰⁸ Similar associations have been reported in other populations, concerning the bilateral anterior corona radiata, bilateral anterior limb of the internal capsule¹⁹¹ and left external capsule.²⁰⁹

Summary

Results for inhibition point to a role for the corpus callosum, fronto-parietal and cortico-subcortical connections. Convergent findings from causal and correlation evidence using verbal tasks showed involvement of the left SLF.^140,195,196 More precisely, causal evidence in large populations showed results for the left SLF-II and III.¹⁴⁰

To our knowledge, there is no causal evidence showing the impact of altered integrity of the cingulum bundle in inhibition, but this has been supported by convergent findings from diffusion MRI analyses, using different tasks tapping inhibition.^199-202 Although other studies are needed to detail whether the cingulum bundle supports global or specific executive process, such associations would be in line with the role of the cingulate cortex in error detection^210,211 within the M-CIN network, engaged in salience detection and in maintaining task sets in demanding cognitive tasks.⁷⁷

Causal evidence suggest the involvement of the right anterior thalamic radiation in inhibition and showed changes in performance associated with damage to the right inferior fronto-striatal tract.¹⁸⁹ However, these results have not been confirmed in analyses of long-term executive impairment following tumour resection in large populations.¹⁴⁰ The diffusion MRI studies included in this review did not report associations between inhibition and integrity of the fronto-striatal tracts but several showed results for the right anterior thalamic radiation.^{192,198,203,204}

In line with the putative role of the right FAT in executive processes,¹³⁴ DES to the right FAT disrupted Stroop performance,¹³⁷ but this result has not been confirmed in post-lesional analyses of larger populations.¹⁴⁰ Diffusion MRI studies included in this review did not investigate such associations (Fig. 2 and Supplementary Table 1).

Discussion

This review aimed to identify white matter tracts consistently associated with performance on tasks assessing cognitive control abilities according to evidence from cognitive assessments during intraoperative brain mapping, studies using lesion-mapping and diffusion MRI. As expected, the corpus callosum was involved in all executive tasks, in different populations. Broadly and according to how executive functions are most frequently assessed,⁴⁴ results point to several white matter tracts, in particular those underlying fronto-parietal interactions,⁹⁸ within a network in which the SLF would be the main connection.

Causal and correlation evidence showed bilateral contributions of the SLF to global executive performance.^{140,175,177,198} More precisely, the SLF was consistently involved in working memory,^{124,129,130,147,150,151,153-159} assessed with verbal and visual tasks; the left SLF-II supported flexible responses^140,175; causal and correlation evidence showed left lateralization for tasks assessing set-shifting and inhibition.^140,168,177 Moreover, convergent findings from diffusion MRI studies have shown significant contributions of the cingulum bundle to cognitive control.^{162,172,180,199-202} To our knowledge, there is no causal evidence confirming findings for the cingulum. This can be partly explained by the scarcity of studies using executive tasks while performing brain mapping with DES and assessing associations between postoperative lesions and long-term executive abilities.^212-214

Because subcortical structures modulate cortical responses,^136,142,215 the involvement of cortico-subcortical connections in cognitive control functions is expected. Several analyses indicate a role for cortico-thalamic and cortico-striatal white matter, especially for the right anterior thalamic radiation in tasks assessing inhibition.^{189,192,198,203,204} Other studies, including causal evidence, showed the involvement of the FAT and fronto-striatal tracts^{130,137,140,164,189} in performance on executive tasks. However, only one study¹⁴⁰ exploring associations with the FAT and fronto-striatal tracts used whole-brain analyses and comprehensive executive assessment, which allows for comparing possible associations. Whether these connections are crucial for executive functions needs further research.

In the next subsections, we discuss in more detail the results from this review for each white matter tract and their specificity regarding other findings about their functional roles.

The corpus callosum

The corpus callosum is the largest commissural tract in the brain connecting left and right hemispheres. It ensures interhemispheric transfer and integration of motor, sensory and cognitive information between homologous regions. It would also regulate homologous regions via an excitatory and inhibitory process, for optimal information processing. Its central importance in cognition has been largely demonstrated (i.e. in individuals with traumatic brain injury, the high vulnerability of the corpus callosum to long-term axonal injury was found to be associated with diverse cognitive deficits and with a significant impact on clinical outcomes).^216-218 Different subdivisions of the corpus callosum have been described according to topographical organization, including the genu, connecting symmetrical dorsolateral prefrontal regions; the body, containing fibres connecting the premotor and precentral frontal cortex; and the splenium, connecting occipital-temporal regions.¹²³ Studies of patients with partial callosotomy showed functional differences of these subcomponents, its posterior parts supporting visual, auditory and somatosensory information and anterior parts supporting cognitive processes such as attention and executive functions.²¹⁹ As expected, studies included in this review have shown consistent support of the corpus callosum to all executive processes, notably the corpus callosum genu in working memory. Anterior corpus callosum parts, such as the genu, are thought to be involved in executive functions given the numerous projections within regions, such as the left and right dorsolateral PFC, IFG and superior frontal gyrus, playing a critical role in such abilities.²²⁰ The non-specific but essential role of the corpus callosum has been further confirmed in DTI studies showing independent contributions of the corpus callosum to executive performance and global cognition.^162,221-223

Consistent involvement of the superior longitudinal fasciculus in executive functions

Causal evidence from studies using DES and post-lesion cognitive assessment included in this review suggests that the SLF is a key white matter structure subserving executive processes.^{124,129,130,140,168,169} The SLF is a major association tract connecting lateral frontal, parietal and temporal cortical regions. Descriptions identify three^224,225 and others four segments²²⁶ and functional differentiations. The SLF-I (dorsal part) connects parts of the superior parietal lobule with the superior frontal lobe, extending from the medial and dorsal parietal cortex to the dorsal part of the premotor and prefrontal cortices. The SLF-II (middle part) connects regions around the intraparietal sulcus to regions in the superior and middle frontal gyrus including the frontal eye fields, extending from the caudal part of the inferior parietal lobule to the dorsal premotor and prefrontal cortices. The SLF-III (ventral part) connects the inferior parietal lobule to the IFG, extending from the rostral inferior parietal lobule to the ventral part of the premotor and prefrontal cortices. The fourth subcomponent, the SLF-temporoparietal segment (SLF-tp), connects temporal and parietal cortices, running posteriorly from the inferior parietal lobe to the posterior temporal lobe.^224,226 A number of studies in the past decade have indicated the support of SLF to a wide array of cognitive processes, from visual attention to language and complex cognitive control processes.^{95,110,113,126-128,227}

In the right hemisphere, three branches of the SLF underlie a visuospatial network with distinct functional roles.¹²⁸ The SLF-I is thought to support goal-directed spatial attention via a dorsal-mesial pathway,¹¹⁰ whereas the SLF-III would support automatic capture of salient stimuli via a ventral pathway.¹²⁸ The SLF-II would integrate information from the SLF-I and III, therefore mediating interaction between dorsal and ventral attention systems.^128,224 Damage to this segment in the right hemisphere induced left spatial neglect.^113,125 Beyond visuospatial attention, the right SLF-II and III would also be involved in motor aspects of attention (i.e. visuomotor speed and movement trajectory control) and bilaterally with cognitive aspects of motor function.^226,228 In comprehensive analyses combining tractography and a meta-analytic approach to cluster fronto-parietal functions, the SLF-I was the main tract associated with spatial/motor clusters (e.g. saccades, voluntary oriented attention) and the SLF-III, with non-spatial/motor clusters (e.g. verbal working memory, decision-making); the SLF-II was associated with both.²²⁷ In line with these descriptions, results from causal evidence included in this review have shown the support of the right SLF-I and II to visuospatial working memory¹²⁴ and the left SLF-I to verbal working memory.¹³⁰ Also, DES of the left SLF-III produced selective disruption of digit spans performance, eliciting order rather than item errors. A similar disruption occurred when the supramarginal gyrus was stimulated, leading to the hypothesis of its role in the phonological loop by transferring order information from the phonological store to Broca’s area.¹²⁹ The role of the SLF and its different branches in language processes has been demonstrated.⁹⁵ The SLF-III is thought to support phonological and articulatory processes according to causal evidence^95,229 and results of Papagno et al.¹²⁹ suggest a specific functional role of the SLF-III in verbal working memory tasks. DTI studies have shown similar associations between the integrity of the left SLF and performance on different verbal working memory tasks.^147,153,159 The SLF is indeed a major connection underlying areas of activation during working memory tasks regardless of their specific aspects and features, forming a ‘core’ working memory network, including clusters in frontal, parietal and insular cortices.⁸⁸

According to results from causal and correlation evidence in large populations, the left SLF-II would play a prominent role in flexible responses.^140,175 Strong evidence indicates a role for the left SLF-II and III in performance of tasks assessing flexibility and inhibition, according to analyses of resection cavities and disconnection probability in a large cohort of patients with brain tumour, showing long-lasting impairment on the Stroop task and TMT-B switch cost. The left SLF-II was strongly correlated with both measures and with global executive performance.¹⁴⁰ Causal evidence also showed that damage to the SLF-II, SLF-III and the arcuate fasciculus disrupted performance on the TMT-B, which assesses attentional flexibility with significant visuospatial attentional demands.^140,169 In the study of Mandonnet et al., these tracts were located in the right hemisphere,¹⁶⁹ thus reflecting recruitment of visuospatial attention networks. Indeed, the role of the fronto-parietal white matter in maintaining executive function has been recently demonstrated in a large study exploring the impact of cardiovascular risk factors on cognition.⁹⁸

Results from this review showed involvement of the SLF in all executive tasks, regardless of the parameters used to interpret performance across studies. Given that attentional processes are essential for executive functions, the SLF is likely crucial because it supports the integration of information within low top-down, visual, goal-oriented attention^75,128,228 and cognitive control networks (i.e. the D-FPN and L-FPN networks)^75,77 that is needed for goal-oriented behaviour.

Contributions of the cingulum bundle to executive functions

Despite no causal evidence for the cingulum to our knowledge, it has been associated with executive performance in several DTI studies of large populations.^145,162,180 Indeed, according to a recent synthesis focusing on other methods, frequent correlations have been reported between executive assessments and integrity of the cingulum.¹⁰⁰ In detailed analyses of executive performance, higher fractional anisotropy in the cingulum bundle independently contributed to better performance on the three executive domains.¹⁶² This complex white matter tract comprises short and long fibres that extend above the corpus callosum and connect the cingulate cortex with different cortical and subcortical regions, some belonging to the limbic system. Some of its fibres connect the anterior and lateral dorsal thalamic nuclei, dorsolateral PFC and insula. Others connect parts of the temporal lobes, such as subicular cortices, parahippocampal cortices and amygdala.¹⁶¹ Different segments of the cingulum have been described to improve the understanding of its potential underlying functions, from two (‘dorsal’ and ‘ventral’) to five subdivisions (subgenual, anterior, midcingulate, retrosplenial, and parahippocampal or temporal).^161,230,231 Via a large fronto-parietal connectivity within the cingulum, all of its components likely support interactions between working memory and attentional resource allocation.¹⁶¹ Results from this review suggest the involvement of its bilateral anterior, posterior and left parahippocampal segments in verbal working memory.^145,156 Several studies have suggested its support of flexible responses and inhibition abilities,^{162,172,180,199,200,202} consistent with the previously described role of the anterior cingulate cortex in error detection.^210,211 Moreover, the anterior/superior cingulum is connected to the central precuneus, which has been recently shown to support set-shifting²³² and, more dorsally, is likely a D-FPN node.⁷⁷ By supporting connections among key network nodes of salience detection,⁸⁷ the cingulum bundle is probably an important connection underlying interactions within the L-FPN and M-CIN networks.⁷⁷ Other studies are needed to increase knowledge about the weight and specificity of cingulum contributions to executive functions and describe possible functional differentiations between its different segments.

The frontal aslant tract and its putative role in cognitive control

According to direct in vivo evidence, the right FAT would be involved in executive functions because DES of this white matter tract in the right hemisphere disrupted the manipulation of verbal information in working memory and inhibition of prepotent responses, whereas the performance of digit spans forward remained normal.¹³⁷ Also, lesion analyses of large populations have shown results for verbal working memory,¹³⁰ moderate correlations between global executive performance but also verbal fluency, and damage to the FAT in the left hemisphere.¹⁴⁰ Regarding DTI studies, there is limited evidence: two studies included in this review have suggested the involvement of the FAT in verbal working memory,^153,164 in one of them bilaterally.¹⁵³ The FAT was first detected in a study exploring brain structures involved in cognitive control. Aron et al.¹³² identified a white matter structure linking the inferior frontal cortex and the pre-SMA to the subthalamic nucleus, underlying the connectivity of a frontal-subcortical network for response control. Given its connectivity with the IFG and cases of SMA syndrome (reduction of spontaneous movement and speech function) after tumour resections,⁹⁶ much of the research on the FAT has been related to language function, showing its support of speech and language initiation, especially in the left hemisphere.^133,135,233 Recently, detailed analyses showed a prominent role of the FAT in motor aspects of the language.²³⁴ However, such associations are not restricted to the left hemisphere: some studies indicate the involvement of the right FAT in stuttering and verbal fluency.^235,236

A role for the right FAT in executive functions has been hypothesized.¹³⁴ The pre-SMA and SMA subserve motor control, stopping behaviours and more specifically, conflict resolution between competing action plans^68,237,238 and have been identified as key nodes for verbal and non-verbal working memory.^88,138,139 These structures and the right IGF would interact during context monitoring^46,239 and the FAT would support these interactions within a broad cortico-basal ganglia-thalamic-cerebellar circuit involved in action control.^134,240 In line with this hypothesis, causal^137,140 and correlation evidence^153,164 showed its association with executive, goal-directed processes during verbal tasks. The FAT underlies connections among key nodes engaged in executive processes^87,88,240; however, the current literature does not allow for inferences about the need or specificity of its role in such processes, i.e. whether the FAT would support broad cognitive control processes or if it would have an indirect role in executive performance via the control of verbal responses.

Cortico-striatal and cortico-thalamic connections

Several studies showed associations between executive functions and the integrity of tracts supporting of cortico-striatal and cortico-thalamic circuits, involved in the L-FPN and M-CIN.⁷⁷ DES results from Puglisi et al.¹⁸⁹ differ from post-surgical assessments, in terms of tracts involved and nature of the post-surgery changes in performance. DES of the right anterior thalamic radiation and right inferior fronto-striatal tracts elicited mainly errors rather than slowing during performance of the Stroop task, but conclusions of lesion analyses in a subgroup of patients describe a subtle slowing of performance only in those with the fronto-striatal tract resected. Other results indicate a role for the fronto-striatal tracts in verbal working memory, along with the FAT and the SLF¹³⁰ but also in global executive function in the left hemisphere.¹⁴⁰ According to a recent study using voxel-based lesion-symptom mapping in 36 patients who underwent glioma resection in the right hemisphere, simultaneous damage of the middle cingulate cortex and the fronto-striatal tract caused selective visual attention deficits, indicating that it could support attention control.²⁴¹ Nonetheless, the specificity of such associations remains unclear. Damage to the fronto-striatal tract may affect cognitive control functions because it connects key nodes supporting these processes; moreover, it has been involved in motor inhibition and language initiation.^102,135,140 Analyses in large populations suggest that, in the left hemisphere, its role is predominantly related to the verbal demands of executive tasks.¹⁴⁰ Concerning the right anterior thalamic radiation, DTI studies have shown converging findings suggesting its involvement in performance on tasks assessing response inhibition.^8-11 This is in line with a voxel-based lesion-symptom mapping showing that lacunar lesions located in the SLF and anterior thalamic radiation were associated with poor global executive function,²⁴² reflecting disruption of large-scale FPN-subcortical interactions. Moreover, the integrity of the right SLF-II and left anterior thalamic radiation supported better common executive function¹⁷⁵ in analyses of the neural substrates underlying the unity/diversity model. The question of whether the anterior thalamic radiations and fronto-striatal tracts are necessary for normal cognitive control needs further investigation, but these convergent findings provide strong evidence of their implication in executive functions.

Other results from DTI studies in large populations suggest that the external capsule, internal capsule and corona radiata are involved in cognitive control processes^{143,145,152,165,179,182,183,191,208,209} along with fronto-parietal connections and the corpus callosum. Some studies showed contributions of these fibres to global executive function^209,243 and in others, changes in activations of cortico-subcortical networks during task performance supported such associations.^165,208 Collectively, causal and correlation evidence agree with research describing thalamic modulation of functional connectivity across cortical regions, and the role of thalamic nuclei in cognitive processes such as attention, behavioural flexibility and more broadly, attentional control.^67,244 As mentioned above, more than being relays of sensorial information, basal ganglia structures have been identified as key nodes in neural models of executive functions,¹³⁶ supporting cognitive control within the L-FPN and the M-CIN networks.⁷⁷

Taken together, results from studies included in this review indicate the essential role of the structural connectivity underlying the D-FPN, L-FPN and M-CIN networks in executive functions. These networks very flexibly support a wide array of cognitive processes, such as attentional control, selection and salience detection.^77,245 Complex responses emerge from the interactions among these systems and widespread hub regions, which in turn mediate the integration of low-level sensorial, motor and cognitive processes generated in other brain regions.^76,77 Among them, the L-FPN, originally described as the multiple-demand system,^81,82 is thought to be paramount for executive responses.⁷⁷ The central role of white matter underlying this network has been corroborated in a recent study showing that lesions to the subcortical connectivity of the multiple-demand system in the left hemisphere were associated with greater impairments in tasks assessing cognitive control.⁹⁹ At the same time, the complex interconnectivity within the PFC reflects some degree of specialization in small areas. Indeed, executive processes involve the integration of different levels of information processing and dynamic network reconfiguration in a context-sensitive manner,²⁴⁵ implying that parallel to these domain-general structures, the information processing of other task features and sensorial demands rely on several other white matter tracts. Accordingly, results from some DTI studies included in this review indicated the involvement of white matter tracts that were previously associated with sensorial, language or semantic processing in executive functions. For example, the uncinate fasciculus, playing a central role in language,^95,246 was among the tracts associated with performance on executive tasks with verbal demands.¹⁹⁶ Also supporting language, the arcuate fasciculus is associated with the phonological loop^95,96,247 and, according to causal evidence, disconnection of its right fronto-parietal segment significantly predicted spatial neglect.¹¹³ In line with these findings, severe decline in set-shifting abilities assessed with the TMT-B, involved damage to the arcuate fasciculus,¹⁶⁹ whereas DTI studies showed its associations with verbal working memory.^147,150 Moreover, the IFOF, one of the major association tracts mediating visuospatial awareness,^{96,125,248,249} language,⁹⁵ including lexical and semantic processing,^96,250 but also non-verbal semantic cognition,²⁵¹ significantly predicted performance on tasks requiring visuospatial processing.^177,178,252 In a study using the TMT-B completion time to measure set-shifting performance, the right IFOF remained the single predictor after controlling for age.¹⁷⁸ These results should be interpreted in light of the diversity of cognitive processes involved in tasks assessing executive functions.⁴⁴ Findings showing the involvement of multiple tracts do not necessarily imply the need for all of them for a given executive process, but show that these structures participate in the integration of information from multiple networks recruited during the performance of a given cognitive task, according to its demands.

There are limitations to this review. Although the cognitive tasks were included in different sections (working memory, shifting, inhibition) according to their common use in clinical practice and research, such presentation of results could hinder the interpretations concerning other cognitive processes required for performing executive tasks. It is important to bear in mind that, although in theory there are separable executive constructs, executive processes interact with each other and tasks tap executive and non-executive processes (i.e. executive tasks are generally used as a proxy for one particular construct but may capture variance associated with several constructs).⁴⁰ For example, the integrity of projection fibres has been correlated with parameters of tasks or performance that also tap working memory capacity,^171,253 such as the TMT-B completion time,^184,185 and the Tower of London.²⁰³ However, the objective of this review was to identify associations with tasks that are commonly used in the literature to assess cognitive control abilities. Considering the multifactorial aspects of executive tasks and the multiple functional roles described for some white matter tracts, overlapping is expected and the results of this review underline this aspect. Moreover, results were presented regarding the parameters used in the studies to interpret performance and other cognitive processes potentially involved. Other limitations relate to technical aspects of each method used for the study of white matter and the availability of causal evidence. At the neural level, performance of executive tasks relies on the activity of specific brain networks and also the interactions among these networks.^254,255 Accordingly, the DTI studies selected in this review have shown different groups of white matter tracts associated with executive tasks. Our objective was to highlight consistent associations, but these are certainly more complex because a single tract can be correlated with multiple cognitive tasks and behaviours.¹⁰⁰ However, highlighting consistent findings may contribute to identifying not only tracts subserving multiple functions, but also those playing a less essential role that can be compensated for in case of a lesion.²⁵⁶ Concerning diffusion weighted MRI studies, the sensitivity of the DTI metrics used as parameters for white matter integrity may be limited and hinder the identification of the spatial limits of white matter structures, especially white matter tracts containing ‘kissing’ or crossing fibres.^101,117 Metrics derived from DKI seem to be more sensitive to changes in fibre diffusivity^117,119; however, DKI is a time-consuming imaging technique mostly restricted to study protocols and not used in routine clinical practice. Only two studies included in this review used DKI to assess white matter tracts associated with specific executive tasks.^149,158 Nonetheless, several well controlled studies including large samples have shown convergent findings (Supplementary Table 1), thus indicating the tracts underlying the connectivity of well described executive networks. Finally, several DTI studies contained no precise information on tract segments, although some have been described, such as the SLF^123,160,224 and the cingulum bundle.^161,230,231 This information is essential, especially for complex and long connections that may induce specific deficits when damaged at precise locations.⁹⁶ Concerning causal evidence, some conclusions remain limited because of the scarce data from lesion analyses and DES results from the neuro-oncology field, regarding executive functions.^212,214

Conclusions: clinical implications and perspectives

Evidence shows that executive functions are supported by white matter tracts connecting widespread brain regions and those underlying the connectivity of fronto-parietal networks may be necessary for these complex cognitive processes. Research on the involvement of white matter tracts in cognition is of paramount interest, especially for the neuro-oncology field. DES during awake surgery has considerably improved the quality of tumour resection while preserving cognitive functions, even in highly eloquent regions. Understanding the role of some specific white matter bundles allows for the optimizing of subcortical mapping and tumour resection while limiting post-surgical sequelae.^96,101 Indeed, compensation mechanisms are possible upon white matter sparing; otherwise, spatial communication and synchronization among potentially compensating neural networks are compromised.^29,32 However, although the intraoperative mapping of white matter tracts involved in executive functions has started only recently and certainly needs to be optimized, assessing its effective benefits for patients in terms of onco-functional balance remains (i.e. the best trade-off between resecting the tumour and sparing functions).²⁵⁷ To the best of our knowledge, only a small number of large-scale neuropsychological studies has comprehensively assessed executive functions in a longitudinal setting (i.e. before and after surgery) in an attempt to characterize levels of recovery.²¹⁴ Because tumours, especially lower-grade glioma, trigger massive compensatory mechanisms,³² these kinds of studies are essential to determine whether executive functions recover after surgery and, if not enough, which kinds of behavioural tasks should used to reduce the likelihood of postoperative impairments in a patient-specific manner.²⁵⁸ These studies are also important to determine the extent to which tracts supporting executive functions are compensable when invaded by the tumour, a key issue for the surgical planning.

In many cases, radiotherapy is an essential treatment for brain tumours but was found to be associated with radiation-induced cognitive decline after white matter damage in a subpopulation of long-term survivors.²⁵⁹ Several DTI studies exploring white matter in patients with brain tumour revealed early microstructural changes after radiotherapy, which progressed over time and were dose-dependent.^260,261 Among these structures, the cingulum bundle was especially sensitive to radiation-induced demyelination in white matter,²⁶² so damage of this tract could participate in the long-term cognitive decline observed in these patients.⁶ Knowledge of the role of white matter tracts can contribute to the development of radiotherapy techniques²⁶³ allowing to the functionally important white matter tracts to be spared, which seems a promising strategy to limit severe cognitive disability in patients with brain tumour.

Acknowledgements

We received logistic and administrative support from the C.R.N.O. (Centre de Recherche en Neuro-Oncologie) and the SiRIC (Site de Recherche Intégrée sur le Cancer) CURAMUS (INCa-DGOS-Inserm_12560), to accomplish this work.

Funding

No funding was received towards this work.

Competing interests

The authors report no competing interests.

Supplementary material

Supplementary material is available at Brain online.