Reduced White Matter Microstructure in Bipolar Disorder with and without Psychosis

Abstract

Objectives:

Affective and psychotic features overlap considerably in bipolar I disorder, complicating efforts to determine its etiology and develop targeted treatments. In order to clarify whether mechanisms are similar or divergent for bipolar disorder with psychosis (BDP) and bipolar disorder with no psychosis (BDNP), neurobiological profiles for both groups must first be established. The present study examines white matter structure in the BDP and BDNP groups, in an effort to identify portions of white matter that may differ between bipolar and healthy groups or between bipolar sub-groups themselves.

Methods:

Diffusion-weighted imaging data were acquired from participants with BDP (n=45), BDNP (n=40), and healthy comparisons (HC) (n=66). Fractional anisotropy (FA), radial diffusivity (RD), and spin distribution function (SDF) values indexing white matter diffusivity or spin density were calculated and compared between groups.

Results:

In comparisons between both bipolar groups and HC, FA (FDR < .00001) and RD (FDR = .0037) differed minimally, in localized portions of the left cingulum and corpus callosum, while reductions in SDF (FDR = .0002) were more widespread. Bipolar sub-groups did not differ from each other on FA, RD, or SDF metrics.

Conclusions:

Together, these results demonstrate a novel profile of white matter differences in bipolar disorder and suggest that this white matter pathology is associated with the affective disturbance common to those with bipolar disorder rather than the psychotic features unique to some. The white matter alterations identified in the present study may provide substrates for future studies examining specific mechanisms that target affective domains of illness.

Introduction

Bipolar I disorder is characterized by affective features, including episodes of mania and, typically, major depression. In addition to the affective disturbance required for diagnosis, approximately half of people with bipolar I disorder also experience psychosis ¹. This overlap between affective and psychotic features in bipolar disorder complicates the study of its etiology, which remains unclear. Genetic studies suggest, however, that the etiology of psychotic bipolar disorder may be closer to that of schizophrenia than to that of purely affective illnesses ². Elucidating whether disease mechanisms are similar or divergent for bipolar disorder with psychosis (BDP) and bipolar disorder with no psychosis (BDNP) is critical to developing effective treatments but rests first upon the identification of neurobiological profiles in bipolar disorder.

If BDP and BDNP represent distinct entities, then these two groups should differ on biological measures shown to separate people with psychosis from healthy comparisons (HC). Psychosis-related deviations are apparent in studies of BDP, schizoaffective disorder, and schizophrenia and can be traced through cognitive, neurophysiological, and neuroanatomical domains. Studied together with HC, those with psychosis show global cognitive impairment that increases in severity from BDP to schizoaffective disorder to schizophrenia, thus increasing with the prominence of psychotic versus affective features ³. This impairment extends across subtests of the Brief Assessment of Cognition in Schizophrenia ⁴ and also to specific domains such as inhibitory control, reflected in elevated error rates during antisaccade tasks ⁵. While few studies compare cognition in BDP and BDNP directly, those that do show impairments that are similar in kind but both greater and more consistent in BDP than in BDNP ⁶. Impacted domains include executive function, attention, processing speed, working memory, verbal memory, and social cognition ^7,8. Both bipolar groups show cognitive impairment relative to HC, but this impairment often seems to scale with the presence and/or severity of psychotic symptoms.

Studies of electrophysiology further demonstrate altered functioning in people with psychosis during cognitive and sensory aspects of stimulus processing ⁹. Examined together on an auditory oddball task, people with both BDP and schizophrenia show reduced P300 amplitude in response to infrequent target tones, with these effects more severe in schizophrenia ¹⁰. Similarly, both groups show reduced theta/alpha power in response to steadily-occurring standard tones, while the BDP group alone shows enhanced beta and gamma power relative to standard and target tones, respectively ¹¹. These findings imply deviations in working memory and context updating ¹² for people with psychosis, with additional deviations in salience processing ¹³ and perceptual encoding ¹⁴ for people in the BDP group. Similar studies comparing BDP and BDNP confirm reductions in P300 amplitude to target tones but disagree on the profile of differences: some report reductions for BDP only ¹⁵, while others report P300 reductions for both BDP and BDNP, with these deviations being moderated by family history of psychosis ¹⁶. It remains unclear whether neurophysiological deviations are present exclusively in BDP and thus associated with psychosis or present in both bipolar groups and thus common to affective disturbance.

Further evidence of neurobiological deviations that separate people with psychosis according to psychosis severity can be found in studies of neuroanatomy. In terms of gray matter volume, BDP shows reductions in regions spanning frontal, cingulate, insular, temporal, and parietal cortices ¹⁷, as well as the hippocampus subcortically ¹⁸. A similar pattern of reductions is seen in schizophrenia and schizoaffective disorder, but appears more widespread. In terms of white matter volume, McDonald et al. ¹⁹ find that genetic risk for BDP and, separately, for schizophrenia is associated with volumetric reductions in the frontal and left temporo-parietal lobes, as well as corpus callosum. These findings suggest that white matter pathology may be common to psychosis, with less sensitivity to the affective features of bipolar disorder.

Additional neurobiological detail can be gained by examining white matter in terms of microstructural organization, as white matter is comprised of bundled, myelinated axons with specific orientation. White matter microstructure is commonly examined using diffusion-weighted magnetic resonance imaging (DWI), which employs rapidly-shifting gradients to assess in vivo the diffusion of water molecules inherently constrained by the structure of axons ^20,21. The degree of directional diffusion, or anisotropy, indicates whether white matter is intact or compromised ²², with greater fractional anisotropy (FA; scalar values closer to 1) indicating intact white matter architecture and lower FA (values closer to 0) indicating architecture compromised by membrane loss, axonal injury, or demyelination (see Winklewski et al. ²³ for review). Demyelination can also contribute to increases in radial diffusivity (RD) perpendicular to primary fiber bundles ²⁴, such that the combination of increased RD alongside decreased FA is taken to indicate disrupted white matter structure.

These diffusivity metrics index the structure of white matter bundles by capturing the rate of diffusion in a principal direction. Additional metrics calculated from spin distribution functions (SDF) index the connectivity of white matter bundles by capturing the amount of diffusion in multiple directions ²⁵. Reductions in spin density may indicate disruptions in transmission along axons ²⁶, even in the absence of the structural damage indexed by diffusivity metrics. While diffusivity and density results sometimes converge, SDF may be more robust to crossing fibers and partial volume effects ²⁷. Additionally, density measures have shown increased sensitivity in both healthy aging ²⁸ and clinical populations ²⁹.

Investigations of white matter density in bipolar disorder are scarce, but several studies examined white matter diffusivity (FA, RD) in this population. One of the largest studies (N = 311) to compare FA in BDP and schizophrenia shows that mean FA is lower for both of these groups compared to HC in 29 pre-defined white matter tracts, including the internal capsule, corona radiata, superior longitudinal fasciculus, posterior thalamic radiation, and corpus callosum ³⁰. The BDP group also shows higher RD than HC, in tracts including the cingulum, internal capsule, corona radiata, inferior longitudinal fasciculus, inferior fronto-occipital fasciculus, and corpus callosum ³¹. In a recent mega and meta-analysis, Favre et al. ³² found reductions to FA in a combined sample of BDP and BDNP groups compared to HC in the corpus callosum and cingulum. There is also some evidence of reduced FA in the internal capsule, anterior thalamic radiation, and uncinate fasciculus for the combined bipolar group, regardless of the presence of psychotic features ^33,34. Few studies compare BDP and BDNP directly, but those that have found minimal differences between the bipolar subgroups when using traditional FA and RD metrics ^32,35.

Taken together, these findings demonstrate ambiguity surrounding anatomical deviations in bipolar disorder. In some studies, white matter differences seem to be associated with the presence of psychosis, as impairments exist for both BDP and schizophrenia, while in others, white matter metrics do not separate the bipolar with psychosis group from the purely affective bipolar group. The possible contributions of affective disturbance and psychosis to white matter pathology can perhaps be disentangled by examining white matter in both bipolar groups, with and without psychosis, using both diffusivity and density metrics. To this end, the Psychosis and Affective Research Domains and Intermediate Phenotypes (PARDIP) consortium collected diffusion-weighted MRI data in a sample including BDP, BDNP, and HC groups. This provides a unique opportunity to parse potentially divergent contributions from the psychosis features specific to the BDP group and the affective disturbance common to both bipolar groups via a novel connectivity metric.

The primary goal of the present study is to identify white matter alterations that distinguish between bipolar and healthy groups or between bipolar sub-groups using both traditional diffusivity metrics (FA, RD) and a unique density metric (SDF). Diffusivity metrics are analyzed by comparing group means across pre-defined white matter tracts, and then both diffusivity and density metrics are analyzed by comparing group associations across neighboring voxels of the local connectome. The connectome analysis allows for localization of possible differences to portions of white matter rather than entire tracts, potentially improving the specificity of the findings, while the inclusion of both diffusivity and density metrics allows for inference about underlying biophysical profiles of white matter differences. Differences between the combined bipolar groups and HC on these metrics would indicate white matter pathology associated with affective instability alone, while differences between bipolar sub-groups might indicate white matter pathology uniquely associated with psychosis. Given evidence of neural deviations that scale with psychosis severity, we expected that white matter structure would decrease in each group, from HC, to BDNP, to BDP, demonstrating white matter pathology that marks psychosis and perhaps divergent mechanisms within bipolar illness.

Materials and Methods

Participants

Through the multi-site PARDIP consortium, participants in the BDP, BDNP, and HC groups, ages 18–60 years, were recruited across three sites (Boston, MA; Dallas, TX; Hartford, CT) using newspaper and community advertising. All participants provided written informed consent and were compensated for their participation. Study protocols were approved by institutional review boards at each site.

Subjects in the two bipolar groups were stable, often medicated outpatients with a lifetime diagnosis of BDP or BDNP. Diagnoses were based on the Structured Clinical Interview for DSM-IV Axis I Disorders/Patient Edition (SCID-I/P) ³⁶, administered by experienced raters at each site, and confirmed by consensus between at least two raters. The Young Mania Rating Scale ³⁷ and the Montgomery-Åsberg Depression Rating Scale ³⁸ were also administered to subjects in the bipolar groups. Healthy participants had no personal history of psychotic, bipolar, or recurrent major depressive disorders, and no family history of psychotic disorders in first-degree relatives, as determined by self-report. All participants were required to achieve a standardized score of ≥60 on the Wide-Range Achievement Test-IV (WRAT-IV) Word Reading subtest (or the WASI Matrix Reasoning subtest, for non-native English speakers) and to present free of major neurological disorders (including traumatic brain injury with loss of consciousness greater than 30 minutes); major medical disorders affecting central nervous system functioning; and DSM-diagnosed alcohol/illicit substance abuse within one month or dependence within three months. Participants were included in the current analyses only if they provided the clinical data required for diagnosis and also completed a diffusion weighted MRI (DWI) session. Seven participants were excluded due to scanning issues of incomplete coverage or incorrect scan parameters. A total of 151 participants (45 BDP, 40 BDNP, 66 HC) were included in analyses; demographic characteristics of the sample are provided in Table 1.

Table 1.

Sociodemographic and Clinical Characteristics of the Sample

	DSM-IV-TR Diagnosis
Variable	BDP (n = 45)		BDNP (n = 40)		HC (n = 66)		Analysis
Sociodemographic	Mean	SD	Mean	SD	Mean	SD	F	df	p
Age (years)	40.6	10.2	43.4	11.4	35.7	12.6	6.0	2, 148	.003a
Education (years)	14.7	2.7	15.0	2.8	15.6	1.9	2.4	2, 145	.098
	N	%	N	%	N	%	χ2	df	p
Male gender	23	51	13	33	38	58	6.4	2	.041b
Handedness							6.5	4	.168
Right	41	91	33	83	59	89
Left	4	9	6	15	3	5
Ambidextrous	0	0	1	3	4	6
Hispanic ethnicity	6	13	4	10	6	9	0.5	2	.781
Race							9.9	4	.042c
Caucasian	30	67	29	73	40	61
African-American	14	31	9	23	14	21
Other	1	2	2	5	12	18
Site							18.8	4	.001d
Boston	14	31	6	15	30	46
Dallas	24	53	21	53	15	23
Hartford	7	16	13	33	21	32
Clinical	Mean	SD	Mean	SD	Mean	SD	t	df	p
YMRS score	11.8	9.6	10.1	8.0			0.9	79	.376
MADRS score	18.2	13.2	16.0	10.1			0.8	79	.408
Concomitant medication	N	%	N	%			χ2	df	p
Off psychotropic	1	2	1	3	-	-	0.1	1	.933
Antipsychotics	38	84	17	43	-	-	16.3	1	<.001e
Mood stabilizers	35	78	28	70	-	-	0.7	1	.414
Lithium	19	42	10	25	-	-	2.8	1	.095
Anticonvulsants	26	58	21	53	-	-	0.2	1	.625
Antidepressants	24	53	29	73	-	-	3.3	1	.069
Anticholinergics	4	9	0	0	-	-	3.7	1	.053
Stimulants	2	4	1	3	-	-	0.2	1	.628
Sedatives	15	33	13	33	-	-	0.1	1	.935

Note. BDP = bipolar disorder with psychosis; BDNP = bipolar disorder no psychosis; HC = healthy comparison; YMRS = Young Mania Rating Scale; MADRS = Montgomery-Åsberg Depression Rating Scale.

^aAge: BDNP > HC.

^bMale gender: disproportionate number of males in HC, females in BDNP.

^cRace: disproportionate number of participants with race other than Caucasian or African-American in HC.

^dSite: disproportionate number of BDP and BDNP from Dallas, HC from Boston.

^eAntipsychotics: disproportionate number of BDP on antipsychotics.

Diffusion MRI acquisition

Diffusion-weighted images were acquired on 3T magnets at three sites (Boston, MA; Dallas, TX; Hartford, CT, USA). Images were acquired using a diffusion weighted imaging protocol and echo planar imaging sequence (e.g. Dallas site: acquisition matrix = 128 × 125, 45 slices, voxel size = 1.72 × 1.75 × 3 mm, FOV = 220 × 220 mm, TR = 6,300 ms, TE = 85 ms, 1 non-weighted b = 0 image, 32 diffusion-weighted images, b = 1000 s/mm²). Sequence parameters were comparable between sites (Supplemental Table 1).

Diffusion MRI preprocessing

Raw diffusion images were converted from DICOM (General Electric, Siemens scanners) or PAR/REC (Philips scanners) formats to NIFTI using the dcm2niix function ³⁹. All volumes from each participant were visually inspected for motion artifacts, and those reflecting motion were removed from participants’ images and the corresponding b-value and b-vector tables (1.4% of total volumes removed; average number of volumes/participant removed < 1). Preprocessing for diffusion tensor analysis was then conducted using the FMRIB Software Library (FSL) ⁴⁰. First, diffusion images were corrected for eddy current-induced distortions and registered to the first non-weighted (b=0) image using affine transformation. Non-brain tissue was removed using the Brain Extraction Tool ⁴¹. Finally, a single FA image and single RD image were created for each participant using FSL’s Diffusion Toolbox to fit a tensor to each white matter voxel.

Diffusion tensor imaging tract-specific analysis

All FA images were analyzed using Tract-Based Spatial Statistics ⁴². Participants’ FA images were aligned to MNI152 space and combined to calculate a mean FA image for the sample, which was then used to generate an FA skeleton representing the core of fiber bundles common across the sample. Whole brain voxelwise FA values were then restricted to voxels containing 18 major white matter tracts using a combination of three masks, as outlined by Schaeffer et al. ⁴³. First a mask based on each participant’s FA image was used to select voxels with FA values greater than 0.2. Second, the Johns Hopkins University white matter tractography atlas ⁴⁴ was used to create a binary mask tract to select voxels with greater than 5% probability of containing a given tract. Last, a mask was applied to individual scans to select only voxels common to the sample’s FA skeleton. Voxels included in all three masks were retained for analysis. For each participant, mean FA and mean RD were calculated across all retained voxels in each of 18 major tracts. These values were then compared between groups in SPSS version 26 (IBM Corp., Armonk, NY, USA), using ANCOVA to include a covariate of site, and post-hoc Bonferroni adjustment to account for multiple comparisons.

Diffusion connectometry reconstruction and voxel-wise analysis

Each participant’s final NIFTI file was loaded into DSI Studio (http://dsi-studio.labsolver.org) and reconstructed in standard MNI space, using q-space diffeomorphic reconstruction ⁴⁵ to obtain spin distribution functions (SDF) ⁴⁶ and a diffusion sampling length ratio of 1.25 mm to cover more than 80% of the diffusion spins. SDFs from all participants were registered to a standard fiber atlas (HCP-1021) ²⁶ in DSI Studio to create a local connectome matrix. For each participant, SDFs were sampled by local fiber directions from the standard atlas to estimate the local connectome, measured as the magnitude of SDFs in all fiber directions. The local connectome of each participant was arranged into a row vector, and all vectors were compiled into a single local connectome matrix (rows = one participant’s connectome, columns = fiber directions).

Differences in voxel-level FA, RD, and SDF were then obtained from the matrix according to participants’ clinical group, using a regression model. A total of 2000 randomized permutations were applied to the group label to obtain the null distribution of the track length and to estimate the false discovery rate (FDR); tracks with length greater than 20 mm and FDR less than 0.05 were retained. Because connectometry holds that differences should propagate along a common pathway, the group difference was tracked and localized to fiber subcomponents using a deterministic fiber tracking algorithm ²⁷. Fiber components were clustered in DSI Studio and retained if mean FA, RD, or SDF values for the cluster showed group differences in an ANCOVA model with a covariate of site.

Results

Tract-specific analysis

Mean FA and RD values for each of 18 white matter tracts were compared between BDP, BDNP and HC groups. Three tracts showed a main effect of group (statistics shown in Table 2; remaining tracts listed in Supplemental Table 2): left cingulum cingulate gyrus part, left corticospinal tract, and forceps minor. For the left cingulum cingulate gyrus part, there was an effect of group on both FA and RD metrics; post-hoc comparisons were not significant for FA (p > .05), but mean RD was higher for the BDP group than HC (p = .041). For the left corticospinal tract, mean FA was lower for the BDNP group than HC (p = .028). Finally, for the forceps minor, there was an effect of group on FA, but post-hoc comparisons were not significant (p > .05).

Table 2.

White Matter Diffusivity in Tracts with a Significant Effect of Group (p < .05)

				Estimated Marginal Mean (SE)
Tract	Metric	F (df)	p	BDP	BDNP	HC
CGC, L	FA	3.56 (2, 146)	.031	0.515 (0.005)	0.514 (0.005)	0.529 (0.004)
	RD	3.19 (2, 146)	.044a	0.509 (0.005)	0.502 (0.006)	0.491 (0.005)
CST, L	FA	3.52 (2, 146)	.032b	0.598 (0.003)	0.592 (0.004)	0.604 (0.003)
	RD	1.88 (2, 146)	.156	0.431 (0.004)	0.431 (0.004)	0.423 (0.003)
F. minor	FA	3.30 (2, 146)	.040	0.518 (0.004)	0.516 (0.004)	0.529 (0.004)
	RD	2.32 (2, 146)	.102	0.533 (0.006)	0.531 (0.006)	0.518 (0.005)

Note.

BDP = bipolar disorder with psychosis; BDNP = bipolar disorder no psychosis; HC = healthy comparison; CGC = cingulum cingulate gyrus part; L = left; FA = fractional anisotropy; RD = radial diffusivity; CST = corticospinal tract; F = forceps. For RD, estimated marginal means and SE are both multiplied by 1,000. Post-hoc tests were conducted for significant (p < .05) omnibus tests using a Bonferroni adjustment for multiple comparisons.

^aBDP > HC.

^bBDNP < HC.

Voxel-wise analysis

Voxel-level FA, RD and SDF were assessed by testing for group associations first between bipolar groups and HC, and then between BDP and BDNP groups. White matter streamlines significantly associated with group in the bipolar versus HC comparisons are shown in Figure 1. Comparisons between bipolar sub-groups were not significant for FA (FDR = 1), RD (FDR = 1), or SDF (FDR = 0.089).

Figure 1.

White matter streamlines associated with group (only streamlines significantly associated with group are shown). Colors indicate direction: red fibers are oriented left-right, green fibers are oriented anterior-posterior, blue fibers are oriented superior-inferior. a) FA is lower for the bipolar group than the healthy comparisons (HC) in portions of left cingulum cingulate gyrus part and corpus callosum (FDR < .00001). b) RD is higher for the bipolar group than HC in portions of bilateral fornix, left superior longitudinal fasciculus, and corpus callosum (FDR = .0037). c) SDF is lower for the bipolar group than HC in portions of bilateral cingulum cingulate gyrus part, superior longitudinal fasciculus, inferior longitudinal fasciculus, inferior fronto-occipital fasciculus, corona radiata, and cerebellum; right anterior thalamic radiation; and corpus callosum (FDR = .0002).

Comparison between the combined bipolar groups and HC showed reduced FA for the bipolar groups (FDR < .00001). FA was lower for the bipolar groups in portions of the left cingulum cingulate gyrus part and corpus callosum. RD was higher for the bipolar groups (FDR = .0037) in portions of left and right fornix, left superior longitudinal fasciculus, and corpus callosum. In terms of SDF, differences between the bipolar groups and HC were widespread. SDF was lower for the bipolar groups (FDR = .0002) in portions of bilateral cingulum cingulate gyrus part, superior longitudinal fasciculus, inferior longitudinal fasciculus, inferior fronto-occipital fasciculus, corona radiata, and cerebellum, as well as right anterior thalamic radiation and corpus callosum. There were no streamlines in which SDF was higher for the bipolar groups than HC. These findings of reduced SDF correspond with the direction of the diffusivity findings (reduced FA and increased RD) in the bipolar groups.

Discussion

The present study examined white matter microstructure in bipolar disorder with and without psychotic features, in order to determine whether white matter pathology might mark affective disturbance generally or psychosis uniquely. Traditional FA and RD metrics separated both bipolar groups from HC, but not from each other, in select white matter streamlines, while a more novel SDF metric separated the bipolar groups from HC in several white matter streamlines. These results both demonstrate a novel profile of white matter differences in bipolar disorder and suggest that this white matter pathology is associated with the affective disturbance common to both bipolar groups rather than the psychotic features unique to some.

In both tract-specific and voxel-wise analyses, FA was lower for both bipolar groups than HC in select white matter streamlines. First when FA values were averaged in pre-defined tracts, mean FA was lower for the BDNP group versus HC in the left corticospinal tract (p = .028), with no difference between BDNP and BDP in this tract (p = .687) or any others. Similarly, when FA values were examined at the voxel level, FA was lower for the combined bipolar group than HC in portions of left cingulum cingulate gyrus part and posterior corpus callosum (FDR < .00001) but did not differ between the BDP and BDNP groups in any streamlines (FDR = 1). These findings of reduced FA are in line with previous reports in bipolar disorder ³⁰, although the spatial extent of the reductions is more limited in the present study. Lower FA indicates a reduction in diffusion parallel to fiber bundles, which should be high when white matter is structurally intact. This decrease in FA suggests a change in white matter structure that could be a result of membrane loss, axonal injury, or demyelination ^23,47.

Similarly, RD differed between the bipolar groups and HC in limited white matter streamlines. When RD values were averaged in pre-defined tracts, mean RD was greater for the BDP group than HC in the left cingulum cingulate gyrus part only (p = .041), while BDP and BDNP did not differ in this tract (p = .999) or any others. When RD was examined at the voxel level, RD was higher for the combined bipolar groups than HC in portions of bilateral fornix (which was not assessed in the tract-specific analysis), left superior longitudinal fasciculus, and corpus callosum; there were no significant RD differences between BDP and BDNP in any streamline (FDR = 1). These findings of increased RD are in line with previous reports in bipolar disorder ³¹, although the spatial extent of the increases is more limited in the present study. Higher RD indicates an increase in diffusion perpendicular to fiber bundles, which should be minimal when restricted by intact axons. This increase in RD suggests a change in white matter structure that could be a result of demyelination ^24,48.

Together, FA and RD metrics demonstrated a pattern of white matter alterations that were few and focal, distinguishing both bipolar groups from HC in small sections of few white matter tracts, but not from each other. In the case of spin density, as indexed by the SDF metric, the pattern of differences between the bipolar groups and HC was more extensive. The white matter implicated still included portions of bilateral cingulum cingulate gyrus part and the corpus callosum, but to a much greater extent (Figure 1c). Additional reductions in SDF were seen in portions of white matter projections between fronto-temporal regions and more posterior regions – including bilateral superior longitudinal fasciculus, inferior longitudinal fasciculus, and inferior fronto-occipital fasciculus – as well as portions of the anterior thalamic radiation, corona radiata, and cerebellum. It should be noted that these results stem from the voxel-wise analysis of the local connectome, equipped to identify specific subcomponents of white matter that differ by group rather than entire tracts that differ at undetermined locations within their boundaries. This increases our confidence that the extensive differences identified in the current sample are indeed specific to white matter streamlines impacted; the streamlines impacted are just numerous.

The extensive nature of the streamlines showing reduced SDF for both bipolar groups indicates a widespread change in diffusion density, the amount of water able to transfuse along white matter axons ⁴⁵. Considered alongside the diffusivity results, this suggests a two-fold biophysical profile of white matter alterations in bipolar disorder: white matter structure may be altered in very specific regions yet remain intact overall, while connectivity between intact axons is altered more extensively. Such an explanation appears consistent with previous reports comparing BDP and other psychosis syndromes. In terms of white matter structure, the BDP group only shows reduced FA within a subset of the voxels impacted in people with schizophrenia ³⁰. On other measures such as resting-state connectivity, however, the BDP group shows the same reductions as the schizophrenia group ⁴⁹. There is evidence, then, that white matter architecture may be altered in BDP to a lesser extent than other psychosis syndromes, while connectivity is altered more comparably.

The profile of white matter differences described in this study likely applies to bipolar disorder with and without psychosis, as bipolar sub-groups were not distinguished by FA, RD, or SDF metrics. These results are in line with previous studies of both white matter structure ^33,34 and neurophysiology ¹⁶, which show respective reductions in FA and electrophysiological responses in people with bipolar disorder, regardless of psychosis state. The current study provides an additional context in which BDP and BDNP differ biologically from HC but not from each other, suggesting that biological differences may be common to affective disturbance rather than unique to psychosis. These differences may still be apparent in other psychosis syndromes but at least partially attributable to their affective components; affective features are present in schizoaffective disorder by definition and often even in schizophrenia, to a lesser extent.

This study was uniquely equipped to parse potentially divergent contributions from the affective features common to bipolar disorder and the psychotic features specific to BDP, given the inclusion of both the BDP and BDNP groups in our sample. This allowed for direct comparison of white matter between participants experiencing affective symptoms in isolation and in combination with psychosis. Both BDP and BDNP groups were similar in terms of affect, as scores did not differ on the YMRS or the MADRS scales (Table 1). This helped to ensure that affective symptomatology was common to both groups, while psychosis was specific to the BDP group. In addition to its comprehensive evaluations, this study was also unique in its methodological approach. We examined both traditional diffusivity (FA, RD) and more novel density metrics (SDF), using complementary tract-specific and voxel-wise analyses. While tract-specific analysis identifies mean differences in pre-defined white matter tracts end to end, local connectome analysis tracks voxel-wise differences in FA, RD, or SDF metrics as they propagate along white matter fibers, allowing for identification of specific white matter subcomponents that differ between groups. Adopting the latter approach helped to confirm that the few FA/RD differences seen in the tract-specific analyses were truly focal in nature and that the SDF differences, while widespread, were still specific to our group variable.

Examination of the SDF metric provided evidence of a novel profile of white matter differences in bipolar disorder. Potential limitations include modest sample size (BDP=45, BDNP=40, HC=66). Additionally, participants with BDP were diagnosed based on lifetime presence of psychosis symptoms, making it possible that psychosis symptoms were minimal or remitted prior to participation in the study. For both the BDP and BDNP groups, the age of symptom onset and illness duration were not analyzed and might have an impact on white matter. Finally, the current results may have been impacted by medication, as all but two participants reported taking psychotropic medication during the course of the study (Table 1). Medication effects are difficult to control in studies of chronic illness but are perhaps less likely here given that BDP and BDNP groups were similarly medicated, across all medication classes (including mood stabilizers), with the exception of antipsychotics. Antipsychotic usage was higher in the BDP group (84%), than in the BDNP group (43%). (Table 1). It should also be noted that reductions to FA have been found in first-episode, medication naive psychosis patients ^50,51. These observations increase our confidence that psychotropic medication did not have undue influence on our white matter results.

The primary goal of the current study was to identify white matter alterations associated with affective disturbance and/or psychosis, through examination of both traditional diffusivity (FA, RD) and more novel diffusion density (SDF) metrics in bipolar disorder. We demonstrated a pattern of white matter alterations in the members of the bipolar group with and without psychosis, characterized by focal changes in diffusivity alongside widespread changes in diffusion density. These results suggest that white matter structure may be largely intact in bipolar disorder, while white matter connectivity and axonal transmission are altered – a pattern not detected previously with traditional diffusivity metrics alone. Given that this white matter pathology is shared between BDP and BDNP, this neurobiological profile may be associated with the shared phenomenology of affective disturbance, independent of psychosis manifestation. This indicates further that white matter connectivity may serve as a promising target for understanding affective disturbance as a basic domain rather than bipolar disorder as a discrete syndrome. Future studies of functional connectivity and electrophysiology in BDP and BDNP, as well as additional affective syndromes, may help to further elucidate the implications of reduced diffusion density in affective disturbance.