To what extent does white matter map to cognition in bipolar disorder? A systematic review of the evidence

Cognitive impairment is a prominent feature of bipolar disorder (BD), however the neural substrates under-pinning it remain unclear. Several studies have explored white matter as a correlate of cognitive functioning in BD cohorts, but mixed results and varied methodologies from one to another make inferences about this relationship difficult to draw. Here we sought to systematically synthesise the findings of these studies to more clearly explicate the nature and extent of relationships between white matter and cognition in BD and determine best practice methodologies and areas for future research in this area. Using PRISMA guidelines, we identified and systematically reviewed 37 relevant studies, all of which were cross-sectional by design. There was substantial methodological heterogeneity and variability in the clinical presentations of BD cohorts encapsulated within the studies we reviewed, which complicated our synthesis of the findings. Nonetheless, there was some evidence that cognition is related to both white matter macrostructure and microstructure in people with BD. In particular, multiple microstructural studies consistently reported that higher fractional anisotropy, both globally and in the corpus callosum, associated with better complex attention skills and executive functioning. However, several reports did not identify any associations at all, and in general, associations between WM and cognition tended to only be evident in studies utilising larger samples and post-hoc selection of WM regions of interest. Further research with increased statistical power and standardised methods are required moving forward.


Introduction
Bipolar disorder (BD) is a complex affective disorder characterised by a fluctuating illness course involving periods of depression, mania, and euthymia.BD is recognised by the World Health Organisation as a leading cause of disability and mortality (Global Burden of Disease Consortia, 2022), likely due in part, to trait-like cognitive impairments observed in 40-60% of individuals with the disorder (Burdick et al., 2014;Karantonis et al., 2020;Van Rheenen et al., 2017).Although these cognitive impairments are well recognised predictors of day to day functioning, they are not currently addressed by first line BD treatments (Van Rheenen et al., 2020b).To develop new treatments targeted at cognition in BD, a better understanding of the neurobiological mechanisms associated with these impairments is needed.
In the general population, investigations of brain -behaviour relationships have linked cognition to white matter (WM) (Filley and Fields, 2016;Fjell et al., 2011;Kloppenborg et al., 2014).This is not surprising given that cognitive outputs generally reflect the integration of spatially distributed brain regions, and WM encompasses the inter and intra hemispheric tracts of the brain that connect cortical and subcortical grey matter regions to enable information transfer within neural networks (Lebel and Deoni, 2018;Madden et al., 2009).WM tracts therefore have functional relevance to cognitive behaviour.
WM macrostructure refers to the overarching structural properties of these tracts; their lengths, areas and volumes throughout the brain network (Yeh et al., 2019).The connectivity, geometry and magnitude of these properties can be measured using structural magnetic resonance imaging (MRI), whereby changes to these properties indicate WM macrostructural pathologies (Schilling et al., 2022).Such pathologies include hyperintense WM lesions and changes to the overall volume or thickness of the WM.This likely represents biological processes such as demyelination, reduced glial density, cortical thinning, cerebral atrophy or hypoxia (Tomimoto, 2015).
In contrast, WM microstructure refers to the individual fibre pathways, and is quantified using diffusion weighted MRI, and its primary measure fractional anisotropy (FA), which reflects the overall magnitude of water diffusion in one direction along an axon.Constituent measures of FA include i) axial diffusivity (AD) -describing the tendency for diffusion along the principal direction of the axon fibre; ii) radial diffusivity (RD) -describing the tendency for diffusion perpendicular to axonal walls; and iii) mean diffusivity (MD) -describing the average of the principal water diffusivity in the three diffusion directions (Assaf and Pasternak, 2008).A decrease in FA suggests compromised WM tract integrity broadly, while decreased AD and increased RD and MD may more specifically index axonal injury, myelin abnormalities and altered cellular white matter energy metabolism, respectively (Mori et al., 2009).
Both macro and microstructural abnormalities of WM have been reported in BD (Favre et al., 2019;Nortje et al., 2013), with the relationship between WM and cognition in the disorder having been of recent interest.Given the neuroplastic nature of WM as indicated by the observation of activity dependent myelination (Filley, 2021), reduced WM integrity in BD may underpin the cognitive impairments seen in the disorder and could therefore represent a neurobiological treatment target as it relates to restoring dysfunctional cognition.The growing body of work on WM-cognition relationships in BD has explored various combinations of WM regions, pathologies, and cognitive processes using diverse methodological approaches in different sized BD cohorts with different accompanying mood symptoms, in different stages of illness and using different types of medications.The findings of these studies have been mixed, and thus difficult to interpret independently, hampering progression toward understanding brain-behaviour interactions in BD.To date, there has been no systematic aggregation and synthesis of work on this topic, and thus a complete and unbiased overview of the results of the various research studies has not been made.Likewise, systematic limitations across these studies have not been identified, nor is it clear which research questions still need to be addressed in future research, because the available evidence does not provide clear answers.Therefore, the aim of this systematic review is to appraise and summarise the literature pertaining to the presence and extent of relationships between cognition and WM and BD.In doing so, we aim to 1) provide a comprehensive overview of WM-cognition relationships in BD, 2) compare these relationships to those reported in accompanying healthy comparison (HC) samples and 3) determine best practice methodologies and areas for future research in this area.

Search protocol
This review was conducted in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (Page et al., 2021) and its protocol was registered as a preprint prior to commencement (10.31234/osf.io/nspfb).The databases PubMed, Scopus and Web of Science were searched to identify studies published from the 1st of January 1980 to the 1st of December 2022, using search terms based on the three key concepts relevant to this review; BD, WM and cognition (see Supplementary 1 for complete electronic search strategy).Reference lists of retrieved records were also examined for additional works relevant to the search terms.

Study selection
Studies were assessed and selected through a two-stage screening process.At stage 1, the titles, abstracts and keywords were scanned by a single reviewer (GFC), and included in stage 2 if they met the following eligibility criteria: 1) written in English and empirical in format; 2) included an adult sample with formal diagnoses of BD (BD I, BD-I, BD II, or BD Not Otherwise Specified[NOS]); 3) used neuroimaging methodologies to examine WM microstructure, hyperintensities and/or volume in the context of cognition; 4) explored WM-cognition relationships, including general or domain specific measures of cognition.The decision to exclude studies of exclusively paediatric samples (<18 years old) from this review was made post-hoc, to avoid confounding conclusions that may arise due to the developmental and diagnostic challenges related to paediatric BD (Carlson, 2012).At stage 2, two reviewers (GFC and SPC) independently screened full text articles according to these additional criteria: 1) the article was a peer-reviewed original article for which the full text is available; 2) cognition was measured objectively using validated task(s); 3) WM micro or macro structure was measured using imaging methods such as diffusion weighted imaging, T1 or T2 weighted structural imaging, or fluid attenuation inversion recovery (FLAIR) sequencing; 4) the analysis and results included a BD-only analysis; and 5) an independent sample was investigated.Discrepancies in the eligibility and proceeding selection of studies were discussed until a full consensus was reached.

Data extraction
Data extraction was conducted by a single reviewer (GFC) and audited by a second reviewer (SPC), along with the authorship team at three time points during the extraction process.The data points extracted focused on 1) sample characteristics, 2) methodological characteristics and 3) key findings.The sample characteristics included sample size, and means and standard deviations for age, education years and percent right-handedness.Additional sample characteristics extracted for BD samples only included diagnosis and diagnostic measure, symptom severity rating and symptom measure, and means and standard deviations for age of illness onset, illness duration, number of mood episodes, number of hospitalisations and current medication usage.
Methodological characteristics related to assessments/analyses of WM and cognition were extracted.For WM, this included the analytical approach (whole brain or ROI), MRI scanning procedures, processing pipeline, analytic modelling and WM measures derived.For cognition, this included domains(s) explored, neuropsychological task(s) used and premorbid intelligence assessment used.To enable consistency in synthesising results, we recategorized the cognitive assessments from each study according to a hierarchical model of key cognitive domains; complex attention, language, visuospatial function, learning/memory and executive function (American Psychiatric Association, 2013) (see Fig. 1).Global cognition and intelligence are also acknowledged with this framework given the generalised ability they represent.
Data extraction of key findings included relationship(s), or lack thereof, between WM and cognition, including WM region(s), cognitive domains, and directionality; and statistical approaches used, including significance level, effect size, confounders or other factors of interest.
This data extraction template was designed by the authors, based on existing systematic reviews in BD, neuroimaging and cognition published to date.Similar templates have been previously piloted and used by our group (Carruthers et al., 2019;Carruthers et al., 2022;Furlong et al., 2021;Karantonis et al., 2022;Karantonis et al., 2021), giving us confidence in its suitability for this systematic review.

Study quality evaluation
All studies included in this review underwent a quality assessment based on an adapted Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) statement (von Elm et al., 2007; see supplementary 2).Each study was evaluated by two reviewers (GFC, SPC) in terms of the clarity of the study design, rationale and hypotheses, methodological standards described, reporting of all results, and acknowledgement of limitations in the work.An amendment to this checklist was made regarding study sample size.A recent systematic review of 1038 structural and functional neuroimaging studies determined that clinical structural MRI studies of high quality had a median sample size of 50 participants (Szucs and Ioannidis, 2020).Thus, we sought to appraise relevant studies in line with this best practice point of reference in this systematic review.Each eligible study received an overall quality assessment score that was calculated by rating all applicable criteria of the adapted STROBE statement as 0 (does not meet criteria), 1 (does meet criteria) or not applicable, with the final score for each study corresponding to its percentage of the total number of applicable criteria (Supplementary Table 4).If a substantial discrepancy in study quality score arose throughout this process, the study in question was presented to a third individual (SLR) for discussion, until a consensus in study quality score was achieved across all reviewing parties.

Results
In the following section we provide a general overview of the characteristics of studies included in this review.We then report the key results, organised by associations between cognition and WM matter macrostructure, followed by associations between cognition and WM microstructure.

Study selection
The study selection process is described in Fig. 2. A total number of 3926 records were initially identified and 2689 records remained after duplicate removal.Articles were then screened based on title, abstract, and keywords, and excluded if they did not meet at least one of the eligibility criteria.105 records remained eligible for full text screening, with 37 subsequently meeting full inclusion/exclusion criteria for this narrative review.

Quality assessment findings
There was substantial variability in the methods and sample characteristics of each of the studies included in this review.However, most studies included were of moderate-high quality (Supplementary Table 4), using appropriate neuroimaging and cognition modalities.However, two-thirds of the included studies featured small sample sizes.Only 13 studies (35%) included samples of ≥50 BD participants; the median sample size observed in highly cited clinical neuroimaging studies, as reported in the comprehensive evaluation by Szucs and Ioannidis (2020).Of these, only 3 had samples >100 (Abramovic et al., 2018;Francis et al., 2016;Macoveanu et al., 2021).Hence, existing studies of WM-cognition relationships in BD appear to have been largely underpowered and exploratory.Some studies also neglected to report critical characteristics that could contribute to alterations in both WM and cognition in BD, including: diagnostic subtype (Yamada et al., 2015), mood state (Ajilore et al., 2015;Alves et al., 2018;Bearden et al., 2011;Bruno et al., 2006;Rej et al., 2014;Shahab et al., 2019) and medication status (Bauer et al., 2015;Bearden et al., 2011;Dev et al., 2017;El Nagar et al., 2022;Rej et al., 2014).

Sample characteristics
As presented in Table 1, participants were predominantly middleaged with a near even sex distribution.The vast majority of the BD sample had BD-I, with only nine studies including participants with BD-II (Bauer et al., 2015;Bruno et al., 2006;Gildengers et al., 2015;Haller et al., 2011;Liu et al., 2010;Macoveanu et al., 2021;Masuda et al., 2020;Rej et al., 2014;Shahab et al., 2019).Thus, unless diagnostic subtype was incorporated into the analysis of WM-cognition relationships as a variable of interest, BD subtype is not specified in subsequent descriptions/discussion of findings.Two studies examined participants with psychotic BD (Francis et al., 2016;Lewandowski et al., 2015), and one included participants with BD-NOS (Bauer et al., 2015).The majority of the BD participants within the included studies were euthymic at the time of assessment, although five studies examined participants in a depressive episode (Dupont et al., 1995;Magioncalda et al., 2016;Martino et al., 2016;Melloni et al., 2019;Poletti et al., 2015) and three examined those in a manic/hypomanic state (Magioncalda et al., 2016;Martino et al., 2016;Ramirez-Bermudez et al., 2021).A HC group was included in all but three studies (Bruno et al., 2006;Melloni et al., 2019;Poletti et al., 2015).Several studies also examined participants with other psychiatric diagnoses (Dupont et al., 1995;El Nagar et al., 2022;Francis et al., 2016;Knochel et al., 2016;Knochel et al., 2014;Masuda et al., 2020;Shahab et al., 2019), but the results of these examinations are not reported given they are not within the scope of this review.

Methodological approaches
All 37 included studies were cross-sectional, and the neuroimaging approaches used within these studies varied considerably.As summarised in Table 2, fourteen explored WM macrostructure, measuring WM hyperintensities (n = 5), volume (n = 8) and thickness (n = 1) across both the whole brain (n = 11) and ROIs (n = 3).Twenty-six investigated WM microstructureusing diffusion weighted imaging to measure the whole brain (n = 12) and tract integrity (n = 12), as well as magnetisation transfer imaging (MTI) to explore myelin and axon abnormalities (n = 2) (summarised in Table 3).Three of these studies employed both macro and microstructural approaches.Basic summaries of the neuroimaging methodologies of the included studies are provided in Supplementary Tables 4 and 5. 1  Regarding cognition, all major cognitive domains were assessed, with Fig. 3 showing the number of studies examining each cognitive domain (by WM imaging approach).Complex attention and executive function were most prolifically studied, being examined in 29 studies.It is also of note, that 11 studies indexed cognition as a composite score of several domains.In these cases, we were unable to recategorize the individual assessments from which these scores were derived according to the hierarchical cognitive domain model, and the findings are presented according to the cognitive domain specified in the original report.Supplementary Table 3 provides details regarding the cognitive assessments administered in each included study, and the original and recategorized domain that it represents.

WM hyperintensities
Mixed findings were evident across the 5 studies exploring relationships between WM hyperintensities and cognition.Dupont et al. (1990) examined 19 mildly depressed BD patients and found those with WM hyperintensities had worse language and executive function than those without.In a larger follow-up study of 36 BD patients, Dupont et al. (1995) also observed worse executive function, language and learning/memory performance in BD patients with a high versus low burden of WM hyperintensities, but the findings did not maintain statistical significance when corrected for multiple comparisons.Similarly,  1 Reporting of between group differences in WM microstructure and cognition are outside the scope of the current review.However, brief summaries of key comparisons are provided in Supplementary Tables 4-6 for convenience.Gildengers et al. (2015) found that whole brain WM hyperintensities were inversely correlated with complex attention, executive function, learning/memory, visuospatial function and overall global cognition, in a sample of 58 euthymic BD patients.An earlier study by this group contrasted this finding, however.That is, Rej et al. (2014) found no relationship between whole brain or prefrontal WM hyperintensities and any of the aforementioned domains, in their pilot sample of 27 euthymic BD patients.This study was, however, the sole WM hyperintensity investigation that observed relationships with cognition in HCs.Indeed, whilst no relationships were observed in BD, higher WM hyperintensity burden was associated with lower global cognition in HCs (n = 12) in addition to prefrontal hyperintensities correlating negatively correlating with learning/memory performance.Finally, Rolstad et al. (2016) found that total WM hyperintensity levels were associated with worse executive function but not complex attention or learning/memory in 75 euthymic BD patients.

WM volume
Limited evidence of relationships between WM volume and cognition were identified in the 6 studies examining these associations.In one analysis of global WM (i.e.averaged across the whole brain), Chakrabarty et al. (2021) clustered 91 euthymic first episode mania BD patients according to their complex attention (processing speed), executive function and learning/memory performance.These authors found reduced total WM volume in globally impaired compared to selectively impaired (predominantly complex attention deficits) and above-average performing patients, as well as HCs.This pattern of results remained even when covarying for psychotropic medication use.Macoveanu et al. (2021) also clustered 153 euthymic BD patients into cognitively impaired (n = 91) and cognitively normal (n = 62) groups based on their complex attention, executive function and learning/memory performance.In line with the findings of Chakrabarty et al. (2021), this cognitively impaired subgroup also presented with significantly lower total cerebral WM volume compared to both the cognitively normal BD subgroup and HCs.
In another whole brain analysis, Bruno et al. (2006) observed that lower intelligence was associated with reduced left superior temporal sub-gyral WM volume in BD-II (n = 11) but not BD-I participants (n = 25), all experiencing a mild depressive episode.Despite these findings, no relationships were observed in the 5 other studies employing either ROI (Francis et al., 2016;Zimmerman et al., 2006) or cognitive subgrouping approaches (Alonso-Lana et al., 2016), or considering the impact of age (Sarnicola et al., 2009) or age of illness onset (Alves et al., 2018) on the relationship between cognition and WM volume in BD.Furthermore, no relationships between WM volume and cognition were observed in any HC cohort involved in these studies.

WM thickness
A single study by Bearden et al. (2011) explored links between WM thickness across the corpus callosum and cognition in a twin study including 21 euthymic BD probands, 19 non-BD co-twins and 34 HC twin subjects.Thinning of the genu and splenium was evident in both BD probands and their co-twins, and positively correlated with a measure of language.Thinning in these regions was also correlated with worse executive function, specifically increased errors and response time on the Stroop task, although the latter did not survive correction for multiple testing.No correlations between the WM and cognitive measures were evident in HCs.

Global WM microstructure
Significant relationships between global WM microstructure and cognition were evident in all 3 studies exploring these variables.Specifically, Gildengers et al. (2015) found that higher global FA correlated with better global cognition as well as performance on the domains of complex attention, executive function, language, learning/memory, and visuospatial function in a sample of 58 euthymic BD participants.Magioncalda et al. (2016) also probed the relationships of complex attention and language with not only global FA, but also global MD, RD, and AD in 61 BD patients in a range of mood states (euthymic n = 20, manic n = 21, depressed n = 20).Better language and complex attention (increased CPT hits and reduced omission errors) was associated with increased global FA and reduced global MD and RD in this study after adjusting for mood state and severity.No correlations between measures of WM and cognition were evident in the HC included in this study or that of Gildengers. However, Abramovic et al. (2018) found correlations of a similar direction in a combined BD-HC analysis (N = 424) in the context of intelligence linking to increased global FA and reduced global MD.Relevantly, no associations were evident in the analysis of 257 euthymic BD patients on their own.3.5.2.Tract specific WM microstructure 3.5.2.1.ROI analyses in tracts defined post-hoc.Eleven studies examined cognition and WM microstructure, in tracts that were initially shown to be abnormal in BD patients versus HCs.Five studies analysed only FA, whilst the remaining six also analysed MD (n = 6), AD (n = 5), and RD (n = 5).Of the work focused only on FA, no evidence of a relationship with executive function was found in either Masuda et al. (2020), Liu et al. (2010) or Alves et al. (2018) studies of n = 30, n = 27 and n = 30 BD patients, respectively.However, Masuda et al. did find significant positive associations between one of two measures of complex attention (i.e., Rapid Visual Processing Test) and FA of the left body of the corpus callosum.Learning/memory were also associated with tract-specific FA in the studies of Liu et al., and Alves et al., albeit       Specifically, Liu et al. (2010) found positive correlations between learning/memory and FA of the right subgenual anterior cingulate cortex and right inferior frontal area in patients with BD-I.In patients with BD-II, these domains positively correlated with FA of the subgenual anterior cingulate cortex bilaterally as well as the left middle temporal region.On the other hand, Alves et al. (2018) found that in patients whose illness began earlier than 19 years of age, FA of the right inferior longitudinal fasciculus was positively correlated with the CVLT Total Score, while FA of the left superior longitudinal fasciculus positively correlated with trial 1 of CVLT Delayed Free Recall.This differed to late onset patients, in whom positive correlations were evident between left inferior longitudinal fasciculus FA and CVLT Delayed Free Recall trials 1 and 2, as well as right inferior longitudinal fasciculus FA and CVLT Discriminability scores.Meanwhile, Melloni et al. (2019) found that the combined average FA of tracts that were individually associated with illness duration were also associated with complex attention, executive function, language, and learning/memory performance in 88 patients with BD.These tracts included the bilateral anterior corona radiata, bilateral corona radiata, fornix and genu of the corpus callosum.Secondary mediation analyses further identified that the combined FA of these tracts mediated the negative association of illness duration and complex attention as well as learning/memory.
Finally, whilst Shahab et al. (2019) observed no BD specific relationships between FA and cognition, they did observe that FA of the anterior limb of the internal capsule, body of the corpus callosum, corona radiata, fornix and superior longitudinal fasciculus was positively associated with a combined principal component of complex attention, executive function and learning/memory when both BD and HC participants were analysed together (N = 143).Aside from this study, there was no other evidence that cognitive performance correlated with FA in the HC groups of any of the aforementioned studies.
With regards to the studies examining AD, RD and/or MD in addition to FA, Poletti et al. (2015) found extensive evidence of relationships between these microstructural measures and both complex attention and executive function, using whole-brain tract-wise analyses on data from 78 BD patients experiencing a depressive episode.Specifically, two measures of complex attention (i.e., Symbol Coding and Token Motor Task) were positively associated with FA across most WM tracts denoted by the standard Montreal Neuroimaging Institute (MNI) space.Symbol Coding FA signal peaks were in the anterior thalamic radiation and forceps major, and symbol coding was also negatively associated with MD in the left inferior longitudinal fasciculus, as well as RD and AD in the right posterior thalamic radiation.Token Motor Task FA signal peaks were seen in the forceps minor, left posterior corona radiata and left superior longitudinal fasciculus, while negative associations with MD were evident in the splenium of the corpus callosum, and RD in the body of the corpus callosum.Executive function, as measured by the Wisconsin Card Sorting Test was negatively associated with MD in the body of the corpus callosum and left inferior longitudinal fasciculus, and RD in the left anterior corona radiata, left inferior longitudinal fasciculus and right posterior thalamic radiation.In contrast, executive function as measured by Digit Sequencing, was positively associated with AD across the whole brain, with signal peaks in the bilateral superior corona radiata and left inferior longitudinal fasciculus.No significant relationships were identified with language, and this study did not include a comparative HC group.
Oertel-Knöchel et al. ( 2014) also identified a positive correlation between executive function and MD and RD of the fornix, and MD, RD and AD of the right thalamic radiation in 30 euthymic BD patients.However, these authors found no relationship between WM and the other cognitive domain they measured; learning/memory.Additionally, FA of the splenium and MD and RD of the left thalamic radiation were correlated with learning/memory in the 32 HCs that participated in this study, but no correlations were evident in this group for executive function.Bauer et al. (2015) reported findings like this in their BD cohort of 49 using the BAC-A Affective Interference Test, in that WM         microstructure was associated with learning/memory, but not executive function, complex attention or language.Specifically, affective and nonaffective short term memory performance was negatively correlated with MD of the right corticospinal tract and RD of the left superior corona radiata.MD was also positively correlated with MD with delayed affective memory scores.A follow-up study of 38 patients by these authors specifically focused on complex attention did not detect any relationship between this domain and WM microstructure (Bauer et al., 2016).Rather, FA of the inferior fronto-occipital fasciculus and superior longitudinal fasciculus was positively associated with complex attention in the HC sample (n = 24) alone.
The BD findings of Bauer contrast with those of Magioncalda et al. (2016), who found complex attention, as well as language performance, were positively correlated with FA and negatively correlated with MD and RD across numerous WM tracts bilaterally, including the anterior thalamic radiation, cingulum, forceps major and minor, inferior fronto occipital fasciculus, inferior longitudinal fasciculus, superior longitudinal fasciculus and uncinate fasciculus.It should be noted that the sample in this study included 61 mixed mood BD participants (euthymic n = 20, manic n = 21, depressed n = 20), although mood symptom state and severity were covaried in the analyses.
Finally, Abramovic et al. (2018) examined intelligence and tractwise FA and MD, with associations evidence in a combined sample of 257 BD participants and 167 HCs.Specifically, intelligence was positively correlated with FA of the cingulum, fornix, external capsule, posterior thalamic radiation and uncinate fasciculus, and negatively with MD of the corpus callosum, superior and posterior corona radiate, left sagittal striatum, bilateral posterior limb of the internal capsule and the superior longitudinal fasciculus.Subsequent analyses revealed that these correlations were not moderated by diagnosis status.Of those examining only FA, three studies focused their analyses on the corpus collosum.In a study featuring old-age first episode mania patients (n = 22), Ramirez-Bermudez et al. ( 2021) observed a significant negative relationship between executive function and the FA of the left corpus callosum region.Additional correlations between this region and both complex attention and global cognition, were however, not significant.Similarly, Yamada et al. (2015) found no significant associations between projections from this region and complex attention, executive function and learning/memory in 20 euthymic BD patients.Ajilore et al. (2015) also failed to find associations between mean FA of the body, genu or splenium of the corpus collosum and executive function or learning/memory in 22 euthymic BD-I patients.However, positive correlations between body and genu FA and complex attention were evident.

ROI analyses in tracts defined
In contrast, Dev et al. ( 2017) examined complex attention and FA of the bilateral superior longitudinal fasciculus and the bilateral uncinate as a function of age in 33 patients with BD.They found that FA in the right uncinate fasciculus was reduced in older patients, and negatively correlated with complex attention performance.McKenna et al. (2015) on the other hand used least angle regression modelling and showed that FA of the corpus collosum genu and splenium as well as the right superior longitudinal fasciculus were associated with complex attention performance in 26 euthymic BD patients.No associations were evident between complex attention and the corpus collosum body, left longitudinal fasciculus or uncinate fasciculus bilaterally.In the HC sample in this study however, FA of the body of the corpus callosum and the left superior longitudinal fasciculus was associated with better complex attention, while FA of the right uncinate fasciculus predicted worse complex attention.No significant FAcognition correlations were observed in the HC samples included in the other three aforementioned Of the studies examining other WM microstructure metrics, Linke et al. (2013) focused on the FA, RD, AD, and mode of anisotropy of the corpus callosum, anterior limb of the internal capsule and uncinate fasciculus in a study of 19 euthymic BD patients and 19 HC.While a negative association between FA of the anterior limb of the internal capsule and the executive functioning was found in the combined BD and HC sample, no BD specific relationships were evident.Martino et al. (2016) re-examined the sample of mixed-mood patients participants (euthymic n = 20, manic n = 21, depressed n = 20) first reported by Magioncalda et al. (2016), but with a specific focus on the cingulum and complex attention only.In contrast to the earlier study's findings, the significant association between complex attention and FA in any cingulum region was not replicated.However, complex attention was correlated with MD and RD of tracts projecting from perigenual anterior cortex to the posterior cingulate cortex, with the relationship present in all BD subgroups but more pronounced in those that were manic.
Finally, the remaining 5 ROI studies did not find significant relationships between cognition and any WM region in BD, HC or combined BD-HC analyses (El Nagar et al., 2022;Haller et al., 2011;Knochel et al., 2016;Knochel et al., 2014;Oertel-Knochel et al., 2014).Knochel et al. (2016) did, however, observe that tract length of the fornix was associated with executive functioning in HC's only (n = 32).

Myelin and axon abnormalities and cognition
Two studies investigated relationships between cognition and myelin and axon abnormalities specifically, using magnetisation transfer imaging (MTI).Lewandowski et al. (2015) used both whole brain MTI and diffusion tensor spectroscopy in 21 BD patients with psychosis, but found no evidence of relationships between these imaging parameters and domains of complex attention, executive function, language and learning/memory.Bruno et al. (2006) was also unable to detect any significant associations between whole brain magnetisation transfer ratio (MTR) and executive function or learning/memory in 25 BD-I and 11 BD-II participants, analysed separately.However, a proxy measure of intellectual trajectory (difference between pre-morbid intelligence and current intelligence) was significantly correlated with reduced MTR in the left and right temporal lobes in the combined BD sample; although the BD-II participants reportedly drove this finding.This proxy measure was also negatively correlated with MTR of the right superior and middle temporal gyri in the BD-II group on its own.No significant relationships were observed in the HC groups included in either of these studies.

Discussion
The aim of this systematic review was to synthesise the available literature investigating the relationship between WM and cognition in BD, to establish a comprehensive overview of the state of this relationship in BD as compared to accompanying healthy control samples included in BD research, and to determine best practice approaches for future studies of this nature.Thirty-seven studies into this relationship had been conducted to date.No associations between WM and cognitive performance were consistently reported across the 14 studies investigating WM hyperintensities, volumes and thickness.In contrast, 15 of the 26 studies measuring WM microstructure identified associations between specific microstructural regions and specific cognitive domains.Within these, FA was consistently linked to cognition in a positive direction and MD and RD in a negative direction, which is in line with evidence that reduced FA and increased RD and MD are indicators of reduced white matter integrity (Alexander et al., 2007;Chanraud et al., 2010).Nonetheless, there was generally limited consistency in terms of the specific microstructural measures or tracts associated with particular cognitive domains, although a few important tract-relevant patterns did emerge.Below we summarise the key results of the review, first based on studies examining WM macrostructure, and then those of WM microstructure.An overview of the main methodological limitations of the existing literature is then presented, followed by discussion of the limitations of the current review, and recommendations for future research.

WM macrostructure
The literature exploring relationships between WM macrostructure and cognition identified limited and inconsistent associations.In regard to WM hyperintensities, only two of six studies found evidence of relationships with cognition in their BD samples (Gildengers et al., 2015;Rolstad et al., 2016), and one further study observed a WM hyperintensity-cognition relationship in their HC sample only (Rej et al., 2014).Of note, while the study of Rolstad et al. (2016) identified a relationship between WM hyperintensities and executive function in BD, it was stronger in individuals with low crystalised intelligence, indicating that intermediary factors, such as those that map onto cognitive reserve, may be relevant to macrostructure-cognition associations.Similar relationships have been found in healthy samples (Baker et al., 2017;Dufouil et al., 2003;Nebes et al., 2006) and in other psychiatric populations in the context of grey matter morphology (Van Rheenen et al., 2020a).
Regarding WM volume and cognition, only three of seven studies identified significant associations.One study reported these associations only in the circumscribed region of the left superior temporal sub-gyrus and only in patients with BD II (Bruno et al., 2006), while the two others employed clustering approaches and found significantly reduced whole brain WM volume in BD patients with global cognitive impairment (Chakrabarty et al., 2021;Macoveanu et al., 2021).
In combination, these standalone, non-specific findings do not present strong evidence of a relationship between WM volume and cognition in BD at first glance, particularly when several studies did not report any associations, including in HC groups (Alves et al., 2018;Francis et al., 2016;Sarnicola et al., 2009;Zimmerman et al., 2006).However, it should be noted that the investigations that did report significant associations between WM hyperintensities and cognition not only had larger sample sizes than their comparators, but also accounted for relevant intermediary factors such as older age, cognitive reserve and mood stabilising treatments.Similarly, the most compelling evidence of WM volume-cognition relationship in BD were derived from studies that employed cognitive clustering techniques to stratify their analyses, which suggests that it is only at more severe levels of cognitive impairment that the association of WM becomes more apparent.This is in line with a recent systematic review that found evidence of greater grey matter morphological abnormalities in cognitively impaired versus cognitive intact patients with BD and schizophrenia spectrum disorders (Karantonis et al., 2022).
Notably, there was minimal evidence of macrostructure-cognition relationships in the healthy comparison groups included in the BD studies reviewed, which is inconsistent with that observed in some studies in the broader literature.For example, the longstanding Rotterdam Scan Study (Vermeer et al., 2003) and Framingham Heart Study (Au et al., 2006) found associations between macrostructural pathologies and cognitive performance in healthy, albeit elderly populations.These are large prospective studies, which further highlights the possibility that the absence of such relationships in many clinical and HC samples in BD research may relate to the much smaller sample sizes that have generally been used.Alternatively, given the focus of the Rotterdam Scan Study and the Framingham Heart Study on elderly individuals, it is possible that WM macrostructure-cognition relationships become more apparent with increasing age.Relevantly, while age was statistically controlled in some BD studies, the moderating role of age on the WM macrostructure-cognition relationship was not examined in any study.Should future BD studies seek to extend work in this area, statistical power and the age of participants in the cohort should be more carefully considered.

WM microstructure
Compared to WM macrostructure, WM microstructure -cognition associations were more extensively explored in the reviewed literature.However, no studies analysed WM microstructure-cognition relationships using an unbiased whole brain approach, and only three analysed global WM microstructure, finding associations with either global cognition or complex attention and language (Abramovic et al., 2018;Gildengers et al., 2015;Magioncalda et al., 2016).More prolific were studies employing an ROI approach, either by i) defining ROI's a-priori (n = 12), or ii) post-hoc (n = 11) based on differences in WM between patients and HCs.The largest proportion of significant relationships between WM tracts and cognitive domains were observed within ROI's defined using the latter method, suggesting that relationships between WM tracts and cognition are more prominent when the integrity of the tract is objectively reduced.This is in line with broader findings suggesting that the presence of WM tract pathology is associated with more variability in cognitive performance, which in turn, is more likely to be detected by correlation analyses (Forkel et al., 2022).Further complicating this variability in a-priori versus post-hoc ROI selection across studies, were differences in the selection of cognitive domains of interest.Ultimately, this culminated in numerous ROI tract-cognitive domain combinations across the results that were difficult to synthesise.To provide some organisation and a framework from which some interpretations of the findings could be made, we focussed on WM tracts that were examined in at least two studies of BD, and discuss these below.
Tracts examined in only one study were not included given no pattern of findings could be synthesised.Fig. 4 presents an overview of the tract based findings. 24.2.1.1.Corpus callosum.The corpus collosum was examined in 14 ROI studies in which associations with complex attention, executive function, intelligence and learning/memory were of interest.In six of nine studies examining associations between corpus callosum microstructure and executive function, and five of six studies examining its association with learning/memory, no significant correlations were observed.However, the integrity of the corpus callosum was most consistently associated with complex attention, with six of the 10 studies probing this relationship finding that greater FA and/or reduced MD and RD in this region was associated with better performance on tasks subsumed within this domain ie., those assessing sustained, divided and selective attention, or processing speed and psychomotor coordination (Ajilore et al., 2015;Masuda et al., 2020;McKenna et al., 2015;Melloni et al., 2019;Poletti et al., 2015;Shahab et al., 2019).
The corpus callosum is one of the most pathologically implicated WM tracts in BD (Favre et al., 2019) and is also the largest WM structure in the brain, accounting for a major part of the its overall microstructural integrity (Rodrigue et al., 2018).Thus, evidence of the corpus callosum's specific relationship with cognition in BD is well aligned with the findings reported by Abramovic et al. (2018); Gildengers et al. (2015); Magioncalda et al. (2016) of a correlation between global FA and cognitive performance.
That the most consistent evidence of the role of the corpus callosum in cognition in BD was with regard to complex attention, fits with evidence that the corpus callosum plays a crucial role in maintaining interhemispheric connectivity and efficient network communication (Filley and Fields, 2016).Such a role appears central to the integration of skills involved in complex attention and suggests that callosal microstructure is particularly relevant to the more fundamental aspects of cognition, rather than higher-order skills (Prunas et al., 2018).Relevantly, these associations do not appear to be specific to BD, as several general population studies have also shown evidence of a relationship between the corpus collosum and complex attention (Hinkley et al., 2012;van der Knaap and van der Ham, 2011), as have studies of several neuropsychiatric (Holleran et al., 2020;Meinert et al., 2021) and neurodegenerative disorders (Frederiksen and Waldemar, 2012;Llufriu et al., 2012;Wang et al., 2020).
The absence of relationships between the corpus callosum and cognition amongst the other four studies probing this tract (Haller et al., 2011;Knochel et al., 2016;Ramirez-Bermudez et al., 2021;Yamada et al., 2020) may relate to the following; firstly, low statistical power may provide an explanation as these studies all comprised small samples ranging from between 19 and 22 BD patients.No associations were observed in the HC samples of these studies, which were also generally small in size.Second, all four of these studies selected the corpus callosum as an a-priori ROI, with no significant differences in its integrity observed in group-level comparisons.This may reflect earlier suggestions that significant WM microstructure-cognition relationships are more likely to be detected when the microstructure has a greater degree of pathology.This observation may indicate a new avenue for this research in BD; whereby the magnitude or threshold of WM degradation that is actually impactful upon cognitive functioning should be explored.

Cingulum.
The cingulum was examined in 13 ROI studies, in which associations with complex attention, executive function, intelligence, language and learning/memory were of interest.The cingulum is a medial associative bundle extending within the cingulate gyrus to the corpus callosum (Catani and Thiebaut de Schotten, 2008).With limbic and parahippocampal connections, it has proposed involvement in cognitive functions such as attention and memory (Bubb et al., 2018).However, the findings pertaining to its relationship with cognition in BD did not entirely support this.That is, although three studies observed positive associations between WM integrity and complex attention (Magioncalda et al., 2016;Martino et al., 2016;Poletti et al., 2015), seven found no significant relationship (Alves et al., 2018;Haller et al., 2011;Knochel et al., 2016;Knochel et al., 2014;Oertel-Knochel et al., 2014;Ramirez-Bermudez et al., 2021).Similarly, a consistent lack of associations between the cingulum and both executive function and learning/memory were noted across these same studies in both BD and HC samples.
A common feature that distinguished the studies that observed significant relationships between the cingulum and cognition, and those that didn't, was patient mood symptomatology.Indeed, the three studies by Magioncalda et al. (2016); Martino et al. (2016);and Poletti et al. (2015) included BD patients in an active mood episode, particularly with higher levels of depression, while the other studies found no association in euthymic patient samples.The cingulum is embedded within the limbic circuits that are well established to play a role in negative emotionality, apathy, hostility, motivation and obsessional thinking (Bubb et al., 2018).Hence, our findings suggest that the cingulum may only be relevant to cognition in BD as a proxy of its association with the heightened emotionality evident during active BD mood episodes, which are themselves known to perpetuate cognitive dysfunction (Buoli et al., 2014;Vieta et al., 2018;Volkert et al., 2016).This interplay between mood, WM microstructure and cognition is another essential avenue for future research to better delineate.

Uncinate fasciculus.
The uncinate fasciculus was the focus of 11 ROI studies that examined its association with complex attention, executive function, intelligence, language and learning/memory.Significant relationships between this tract and any of these cognitive domains were, however, observed in only four of the 11 reports, with no clear domain-specific patterns emerging (Abramovic et al., 2018;Dev et al., 2017;Magioncalda et al., 2016;Poletti et al., 2015).
2 Given that no pattern of findings could be ascertained from WM tracts that were examined in only one study, these tracts were not included in Figure 4 or the discussion of key WM microstructural findings in text.
The anatomy of the uncinate fasciculus is well established, characterised by long-range bi-directional connections that link the anterior temporal lobe with the orbitofrontal cortex (Catani and Thiebaut de Schotten, 2008).Knowledge of its function is, however, limited.It has been proposed that the uncinate fasciculus plays a role in episodic memory, social-emotional processing, and language (Papagno, 2011).Specifically, its bi-directionality is thought to support the temporal-lobe modulation of mnemonic stimuli; creating associations between people's names, faces, voices and one's feelings about others, and directing subsequent decision making (Von Der Heide et al., 2013).Thus, it is possible this WM tract may be more relevant to these socio-emotional processes in BD than the more traditional 'cold' domains of neurocognition (Furlong et al., 2021), for which no robust associations were evident across the reviewed studies.
4.2.1.4.Fornix, corona radiata and inferior longitudinal fasciculus.The corona radiata, fornix and inferior longitudinal fasciculus were explored in six, five and four studies respectively.The corona radiata connects the brainstem and thalamus with the cortex (Mori et al., 2009).In contrast, the inferior longitudinal fasciculus connects fronto-temporal brain regions and the fornix connects the hippocampus to other subcortical structures (Catani and Thiebaut de Schotten, 2008).Across two studies, neither the fornix nor the corona radiata were associated with intelligence (which was not studied in relation to the inferior longitudinal fasciculus).However, all tracts were associated with complex attention, executive function, and learning and memory in a predominantly positive direction in at least 10 studies in combination, although null results (n = 5) were evident in relation to associations with complex attention and learning and memory as well (Alves et al., 2018;Melloni et al., 2019;Oertel-Knöchel et al., 2014;Poletti et al., 2015;Shahab et al., 2019).That complex attention and executive function were non-specific in their links to these underlying WM tracts likely reflects the diversity of complex information processing captured within these domains, and the integrated nature of WM tracts throughout the neural network that then underpin fundamental cognitive processes (Filley and Fields, 2016).That said, there is a sparsity of studies explicitly examining each of the domain-tract associations, which makes it difficult to draw firm conclusions about the robustness of the pattern of findings observed here.

Executive summary of WM microstructure-cognition relationships in BD.
In the WM microstructure literature, there is some, albeit limited, evidence, that global measures of FA, and of certain key microstructural regions, are associated with cognition in BD.Complex attention appears tied to most WM tracts, particularly the corpus collosum.However, Note.Only WM tracts that were tested in at least two microstructural studies of BD were summarised.Given that no pattern of findings could be ascertained from WM tracts that were examined in only one study, these tracts were not included in the figure or discussion.n = number of studies.Green text denotes a significant positive relationship; Red text denotes a significant negative relationship; Blue text denotes no significant relationship.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)more detailed and robust evidence is still required to confirm the exact nature of these relationships in BD.
Notably, the two studies that used MTI observed no significant relationship between myelin and axon abnormalities and cognition (Bruno et al., 2006;Lewandowski et al., 2015).In-vivo measurement of myelin and axon integrity remains to be a severely understudied aspect of WM microstructural biology and pathology, requiring further investigation for any synthesis or conclusions to be drawn.Further, whilst relationships with specific cognitive domains were observed in some studies, as indicated previously, many of these same studies also did not find associations with the other cognitive domains they had specified as being of interest.Likewise, seven studies did not identify any associations between WM microstructure and cognition in BD across any of the cognitive domains and tracts they measured at all (Bauer et al., 2016;Haller et al., 2011;Knochel et al., 2016;Knochel et al., 2014;Linke et al., 2013;Oertel-Knochel et al., 2014;Yamada et al., 2020).Although all of these studies received moderate-high general quality assessment scores, they were still amongst the lower scoring of all included in our review (range 85-90%) and had some of the smallest individual sample sizes (ranging from 19 to 38 BD patients).Thus, it is possible they may not have had enough power to detect effects.
Another commonality across these studies that may have contributed to their complete absence of results, is that all of them made a-priori selections of WM ROIs.As mentioned previously, a-priori ROI selection may be less sensitive to WM-cognition relationships than those selected post-hoc.Indeed, a dominant pattern amongst the studies that did observe associations between WM microstructure and cognition in BD was the post-hoc selection of ROI tracts based on significant BD group reductions in WM integrity.This raises the possibility that the targeting of tracts that differ between BD and HCs can inform the development of BD specific WM-cognition models as a result of better indexing pathophysiological processes occurring in BD.Overlaying this approach with the repeated implication of complex attention as the cognitive domain most commonly associated with WM could thus imply that pathophysiological changes to WM integrity initially impact lower-order component processes within this domain; attention, processing speed and psychomotor coordination.
It is also pertinent to acknowledge that it was the few studies of BD patients in active mood states that commonly contributed to the most consistent findings across all key WM tracts (Magioncalda et al., 2016;Melloni et al., 2019;Poletti et al., 2015).Of the seven studies in which no WM-cognition findings were evident across any of the domains analysed, all had low symptom rating scores in their patient cohort.This suggests a possible role of current BD symptomatology in associations between WM and cognition, which should be explored to give insight into the state or trait nature of these associations.

Methodological limitations and future directions
Prevailing across the studies in this review was heterogeneity as well as methodological limitations, which impeded the comparison and synthesis of findings and made it challenging to delineate robust from spurious findings.In describing these limitationsthroughout this section, we also identify several best-practice recommendations for future developments and studies in the field (summarised in Fig. 5).
Firstly, the neuroimaging and statistical techniques used to study WM were varied, with several different brain atlases, analysis pipelines and ROI approaches used across the studies, which hampered the comparability of findings and limited our ability to interpret effects.In most of the ROI studies, be they a-prioi or post-hoc, no global (or relative regional) microstructural measures were covaried (or incorporated) in statistical analyses, and it is thus unclear whether region-specific relationships with cognition occur over and above that of global microstructural effects.It is also of note that most of the a-priori ROI studies of both macro and microstructure yielded a null finding.This may indicate that the theories informing a-priori ROI choices to date lack precision.Going forward, a more standardised approach in measuring and assessing WM is required, with consortia such as ENIGMA DTI or ENIGMA BD offering a useful direction that may more accurately uncover WM-cognition associations by proxy of homogenizing methodological approaches to study them (Holleran et al., 2020).Future work on this topic could also include emerging diffusion neuroimaging techniques, such as MTI and high resolution FLAIR DTI, that more powerfully measure microstructural integrity by mitigating standard DTI limitations such as low spatial resolution, image distortion and misregistration of WM structures (Valdés Cabrera et al., 2022).
The approaches taken to assess cognition also differed greatly between studies.Future studies in this area would benefit from the use of a standardised cognitive battery such as the MCCB/ISBD-BANC (Miskowiak et al., 2017;Yatham et al., 2010).At the very least, researchers should aim to be explicit with what cognitive domain their task of choice is assessing, to ensure accuracy in terms of what the task has been designed, standardised, and validated for.This, along with ensuring that composite cognitive domain scores are calculated with appropriate analyses will improve comparability of findings.In addition, consideration of how secondary factors related to cognitive function, such as intraindividual variability, may impact its relationship with WM in BD, would be useful (Fjell et al., 2011;Gallagher et al., 2015).
Moreover, in addition to only a limited number of studies exploring current mood state, several studies did not report relevant illness chronicity variables such as age of illness onset, illness duration, number of mood episodes and medication regime.Illness duration is well established as individually impactful on both WM microstructure and cognition in BD (Berk et al., 2011;Cardoso et al., 2015;Kapczinski et al., 2017;Van Rheenen et al., 2020b), and the number of mood episodes Use consistent and replicable gold-standard neuroimaging approaches, taking guidance from large internaƟonal consortia such as ENIGMA DTI or ENIGMA BD Assess cogniƟve performance through validated and standardised cognitive batteries in BD such as the MCCB/ISBD-BANC, which specify the ideal tasks for measuring each domain Employ large-scale databases and leverage their sample size, in order to comprehensively explore this relaƟonship and all its potenƟal factors with increased statistical power InvesƟgate intermediary variables that could be associated with WM-cogniƟon relaƟonships in BD, parƟcularly clinical factors such as symptoms, illness chronicity and medicaƟon.Integrate these into analyses to explore their involvement G.F. Caruana et al. experienced in BD have been shown to increase the severity of cognitive deficits (Hellvin et al., 2012;Passos et al., 2016).
Medication may also affect the relationship between WM and cognition in BD, and although documented in the description of patient cohorts, was not included as a covariate or even in secondary analyses in most studies.Lithium in particular is a first-line treatment for BD, and evidence of its neuroprotective effects on both brain structure and function has begun to accumulate (Berk et al., 2017;Dodd et al., 2013;Kessing et al., 2010).A recent systematic review highlighted that in 11 of 16 studies, lithium treatment significantly increased either WM macrostructural volumes and/or microstructural integrity in people with BD (Espanhol and Vieira-Coelho, 2022).Additionally, lithium is neuroprotective against cognitive impairment not only in BD (Abramovic et al., 2018;Burdick et al., 2010;Gildengers et al., 2015;Rybakowski et al., 2018) but also in non-psychiatric populations at-risk of dementia (Donix and Bauer, 2016;Forlenza et al., 2011;Kessing et al., 2017;Mauer et al., 2014;Nunes et al., 2007) and in developed Alzheimer's disease (Matsunaga et al., 2015).Overall, this research demonstrates how BD illness course and treatments may reduce the presence of WM alterations and cognitive deficits, and in turn contribute to the heterogenous findings pertaining to their relationship with each other.
Finally, the majority of studies in this review were conducted with small, cross-sectional datasets, creating low statistical power, with only 35% of the reviewed studies using samples of >50 participants.Of note, is that many of the studies in which WM-cognition relationships were evident tended to be those employing larger samples, which emphasises the need to conduct future investigations of this nature at a larger scale.Future studies could indeed seek to leverage large study populations such as that in the UK Biobank and use this increased statistical power to evaluate WM-cognition relationships and all their potential confounders and covariates in BD more comprehensively.

Review limitations and conclusions
The narrative synthesis conducted allowed for detailed consideration of each study's sample characteristics and methodologies, but the substantial heterogeneity of these studies, the lack of effect size reporting and the diverse outcomes made conducting a meta-analysis unfeasible.Whilst we had to forgo the objectivity that can be facilitated by quantitative meta-analyses, our narrative synthesis was taken with careful consideration of the Cochrane guidelines set by Higgins et al. (2022).However, future work is required to determine the magnitude and quantitative effect of WM-cognition relationships in BD, when data amenable to such statistical solutions become available.
Further, whilst our reporting of participant demographics was detailed, there may be additional clinical variables that influence the relationship between WM and cognition that we did not consider.For example, physical comorbidities, BMI, smoking and cannabis use, lifestyle factors and socio-demographic indicators such as years of education and current employment may influence cohort variability and in turn, the nature of the relationship between WM and cognition.Hence, this review highlights general links between the brain and behaviour in BD, but future reviews should seek to examine the impacts of additional clinical variables on these overarching relationships.Finally, as an objective of this review was to understand the relationship between WM and cognition in BD, any findings that combined WM and grey matter, or that explored structural and functional neurobiology, were not reported on, to maintain the WM focus and preclude any confounders.
In conclusion, this systematic review finds that there is some limited evidence of associations between WM macrostructure and cognition in BD, as well as some replicated links between the integrity of individual WM tracts, such as the corpus callosum, and foundational cognitive processes, such as complex attention.Findings regarding other cognitive domains are mixed.Given the methodological constraints mentioned above, the literature is unlikely to progress without more considered and statistically powerful research designs.As described by our suggested best practice guidelines, the use of standardised neuroimaging techniques and cognitive batteries in better powered datasets are necessary to establish robust evidence of the presence and nature of WM-cognition relationships in BD.The application of these guidelines will also support the capacity to investigate more complex aspects of WM-cognition relationships in BD; such as whether these relationships are state or trait in nature, the moderating role of clinical features, and the overall degree of WM pathology likely to impact cognition in BD.Ultimately, we suggest that value remains in exploring structural brain anatomy in the context of cognitive performance in BD if more robust methodologies are adopted going forward.Doing so could further our understanding of the psychopathological and functional nuances of BD and inform therapeutic approaches.

Fig. 1 .
Fig. 1.Hierarchy of cognitive domains and subskills.Global cognition and intelligence do not perform within this hierarchy, but rather as overarching general constructs (American Psychiatric Association, 2013).

Fig. 2 .
Fig. 2. PRISMA flow diagram depicting the search strategy for this review.

Fig. 3 .
Fig. 3. Overview of the number of studies investigating each cognitive domain, based on white matter approach.Note.WM, white matter.

Fig. 4 .
Fig. 4. Summary of replicated microstructural findings in studies investigating a priori and post-hoc ROI tracts.Note.Only WM tracts that were tested in at least two microstructural studies of BD were summarised.Given that no pattern of findings could be ascertained from WM tracts that were examined in only one study, these tracts were not included in the figure or discussion.n = number of studies.Green text denotes a significant positive relationship; Red text denotes a significant negative relationship; Blue text denotes no significant relationship.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 5 .
Fig. 5. Recommendations for future studies of WM-Cognition relationships in BD.
in different regions and moderated by different clinical characteristics.
Note: BD, bipolar disorder; HC, healthy controls; n, number of studies with reported data point; SD, standard deviation.

Table 2
Summary of studies investigating relationships between white matter macrostructure and cognition in bipolar disorder.

Table 2
(continued ) (continued on next page) G.F.Caruana et al.

Table 2
(continued ) (continued on next page) G.F.Caruana et al.

Table 2
(continued ) (continued on next page) G.F.Caruana et al.

Table 2
(continued ) Inventory of Depressive Symptomatology; MADRS, Montgomery-Asberg Depression Rating Scale; PANSS, Positive and Negative Syndrome Scale; PANAS, Positive and Negative Affect Schedule; ROI, region of interest; SAPS, Scale for the Assessment of Positive Symptoms; USA, United States of America; WMH, white matter hyperintensities; YMRS, Young Mania Rating Scale.than those reported in this table, i.e. schizophrenia, schizoaffective disorder, and major depressive disorder.However, as the focus of this reviews is on adults with bipolar disorder, only bipolar disorder and healthy control groups are presented in this table.

Table 3
Summary of studies investigating relationships between white matter microstructure and cognition in bipolar disorder.
(continued on next page) G.F.Caruana et al.

Table 3
(continued ) N/A (continued on next page) G.F.Caruana et al.

Table 3
(continued ) (continued on next page) G.F.Caruana et al.

Table 3
(continued ) IFOR, inferior fronto occipital fasciculus; IDS-SR30, Inventory of Depressive Symptomatology; ILF(L), left inferior longitudinal fasciculus; Le; tract length; LMTR, left medial temporal region; MADRS, Montgomery-Asberg Depression Rating Scale; MTR, Magnetisation Transfer Ratio; NoFT, number of fibre tracts; PANSS, Positive and Negative Syndrome Scale; PANAS, Positive and Negative Affect Schedule; PCR(L), left posterior corona radiata; PTR(L), left posterior thalamic radiation; RIFA, right inferior frontal area; ROI, region of interest; SACC, subgenual anterior cingulate cortex; SAPS, Scale for the Assessment of Positive Symptoms; S-CC, splenium of the corpus callosum; SCR, superior corona radiata; SLF; superior longitudinal fasciculus; SS(L), left sagittal striatum; TR, thalamic radiation; UF, uncinate fasciculus; USA, United States of America; vol; tract volume YMRS, Young Mania Rating Scale.+ Indicates a significant positive correlation; ¡ indicates a significant negative correlation; N/A indicates not applicable.^Study included groups other than those reported in this table, i.e. schizophrenia, schizoaffective disorder and major depressive disorder.However, as the focus of this reviews is on adults with bipolar disorder, only bipolar disorder and healthy control groups are presented in this table.