Diffusion weighted imaging of prefrontal cortex in prodromal huntington's disease

Abstract

Huntington's disease (HD) is a devastating neurodegenerative disease with no effective disease‐modifying treatments. There is considerable interest in finding reliable indicators of disease progression to judge the efficacy of novel treatments that slow or stop disease onset before debilitating signs appear. Diffusion‐weighted imaging (DWI) may provide a reliable marker of disease progression by characterizing diffusivity changes in white matter (WM) in individuals with prodromal HD. The prefrontal cortex (PFC) may play a role in HD progression due to its prominent striatal connections and documented role in executive function. This study uses DWI to characterize diffusivity in specific regions of PFC WM defined by FreeSurfer in 53 prodromal HD participants and 34 controls. Prodromal HD individuals were separated into three CAG‐Age Product (CAP) groups (16 low, 22 medium, 15 high) that indexed baseline progression. Statistically significant increases in mean diffusivity (MD) and radial diffusivity (RD) among CAP groups relative to controls were seen in inferior and lateral PFC regions. For MD and RD, differences among controls and HD participants tracked with baseline disease progression. The smallest difference was for the low group and the largest for the high group. Significant correlations between Trail Making Test B (TMTB) and mean fractional anisotropy (FA) and/or RD paralleled group differences in mean MD and/or RD in several right hemisphere regions. The gradient of effects that tracked with CAP group suggests DWI may provide markers of disease progression in future longitudinal studies as increasing diffusivity abnormalities in the lateral PFC of prodromal HD individuals. Hum Brain Mapp 35:1562–1573, 2014. © 2013 Wiley Periodicals, Inc.

INTRODUCTION

Huntington's disease (HD) is an autosomal‐dominant, progressive disorder characterized by motor, cognitive, and behavioral disturbances. Age of diagnosis varies inversely with the expanded number of polyglutamine (cytosine‐adenine‐guanine or CAG) repeats in the huntingtin gene [Huntington's Disease Collaborative Research Group, 1993]. Manifest motor onset usually occurs in mid‐life, with a duration of 15 to 20 years after diagnosis [Harper, 1991; Hayden, 1981]. Currently, HD treatments only target symptoms with no pharmacologic solutions for slowing or stopping disease progression [Frank and Jankovic, 2010]. To hasten development of novel treatments that target disease progression, a number of investigators, including the PREDICT‐HD group, are focusing on identifying biomarkers to judge efficacy of new treatments [Paulsen et al., 2006a, 2008].

In the search for a reliable disease marker, volumetric studies using magnetic resonance imaging (MRI) have revealed abnormal brain tissue volumes in prodromal HD (or before symptom onset) patients. Volumetric studies first examined symptomatic HD patients and found prominent atrophy of the caudate and putamen [Jernigan et al., 1991]. HD individuals who had the CAG expansion in the HD gene but no signs/symptoms to warrant a clinical diagnosis (i.e., in the prodrome of HD) demonstrated similar findings of reduced basal ganglia volumes in comparison to healthy controls. Degree of striatal atrophy in prodromal HD individuals also correlated with greater neurological impairment [Campodonico et al., 1998; Harris et al., 1999], poorer performance on cognitive assessments [Campodonico et al., 1998], and years to motor sign/symptom onset [Aylward et al., 1996; Harris et al., 1999]. As for white matter (WM) specifically, cognitive deficits have shown to correlate more with cerebral WM atrophy than striatal atrophy in symptomatic HD individuals [Beglinger et al., 2005]. A significant decrease in total WM volume has also been seen in prodromal HD individuals classified more than 15 years from diagnosis [Paulsen et al., 2006b, 2010] and a disproportionately greater loss of total frontal lobe WM than overall brain volume reductions [Aylward et al., 1998].

WM volume has been shown to correlate with features of disease progression, but volume information alone does not reflect altered white matter integrity. Researchers have thus turned to diffusion‐weighted imaging (DWI) to detect varying levels of anisotropic diffusion that could represent altered WM integrity in diseased tissue [Basser, 1995; Basser and Pierpaoli, 1996; Jones et al., 2012]. A tensor representation is often used to model the diffusion process at each voxel. Rotationally invariant scalars are generated from the resulting eigenvalue decomposition to describe the diffusion anisotropy and magnitude [Basser, 1995; Basser and Pierpaoli, 1996]. Four scalars often used include: fractional anisotropy (FA), mean diffusivity (MD, units = mm²/sec), axial diffusivity (AD, units = mm²/sec), and radial diffusivity (RD, units = mm²/sec). FA reflects anisotropy of the diffusion tensor and is dimensionless, ranging from 0 (isotropic diffusion) to 1 (high anisotropy) [Basser and Pierpaoli, 1996]. MD is the average diffusion magnitude along three principal directions into which diffusion is decomposed [Basser and Pierpaoli, 1996]. AD is the magnitude of diffusion parallel to the principal direction of diffusion, where changes correlate with axonal injury [Song et al., 2003]. Radial diffusivity (RD) is the magnitude of diffusion perpendicular to the principal direction of diffusion, where increases correlate with incomplete myelination [Song et al., 2002] and myelin injury [Song et al., 2003, 2005]. Scalar measures have been used to examine normal‐appearing WM that contains abnormalities (i.e. multiple sclerosis) [e.g. Pagani et al., 2005] and developmental studies to characterize changes associated with aging [e.g. Bucur et al., 2008; Dubois et al., 2008].

As for DWI studies involving HD participants, many have focused on WM of the motor loop [Bohanna et al., 2011; Della Nave et al., 2010; Rosas et al., 2006; Stoffers et al., 2010], periventricular region [Mascalchi et al., 2004], corpus callosum [Bohanna et al., 2011; Della Nave et al., 2010; Di Paola et al., 2012; Dumas et al., 2012; Müller et al., 2011; Rosas et al., 2006, 2010; Sritharan et al., 2010; Weaver et al., 2009], corona radiata [Bohanna et al., 2011; Della Nave et al., 2010; Stoffers et al., 2010; Weaver et al., 2009], and whole brain [Mascalchi et al., 2004; Rosas et al., 2006]. Overall, scalar studies on prodromal and symptomatic HD participants demonstrate diffusivity changes in white matter to explain increased motor signs with disease progression.

Another region of interest in HD disease progression is the prefrontal cortex (PFC) due to its prominent striatal connections. The dorsolateral PFC projects to the central to dorsal caudate (dorsal loop), while the orbital PFC and rostral anterior cingulate cortex projects to the ventromedial caudate and ventral striatum (ventral loop) [Alexander et al., 1986; Arikuni and Kubota, 1986]. Based on the striatal dorsal‐to‐ventral progression of cell death [Hedreen and Folstein, 1995], Lawrence et al. hypothesized that functions associated with the dorsal PFC‐striatal loop may be impaired before motor sign/symptom onset, followed by impairment of functions associated with the ventral loop as neuronal loss increases [Lawrence et al., 1998]. Despite possibly containing valuable information on disease progression, PFC diffusivity in HD individuals has yet to be explored in great detail. Most diffusion tensor scalar studies in the frontal lobe report findings in voxel clusters in regions of frontal lobe WM [Magnotta et al., 2009; Reading et al., 2005; Rosas et al., 2006] or associated major WM bundles [Della Nave et al., 2010]. However, diffusion properties of voxel clusters may not be representative of the larger surrounding region. The only study that has examined diffusivity in an entire sub‐region of PFC WM did so in the superior frontal cortex [Dumas et al., 2012]. Dumas et al. found decreased mean FA and increased mean MD in WM fibers running through the superior frontal cortex in early HD participants [Dumas et al., 2012]. These diffusion findings were supported by the average MD in the WM fibers running through the superior frontal cortex negatively correlating with Stroop word reading task performance [Dumas et al., 2012], a test that is sensitive to cognitive deficits in prodromal HD participants [Stout et al., 2011].

This study strives to build upon past prodromal HD studies by examining WM diffusivity in sub‐regions of the PFC. It was hypothesized that diffusivity differences would be seen in PFC WM regions among CAP groups relative to controls. It was also hypothesized that the difference relative to controls would be a function of CAP group with the high group showing the greatest difference.

METHODS

Participants

This analysis used structural images, diffusion‐weighted images, and clinical data from the first time point of a larger longitudinal functional MRI study, “Cognitive and Functional Brain Changes in Preclinical Huntington's Disease” (HD‐fMRI; NS054893: P.I. J.S. Paulsen). This was a two‐site collaboration whose goal is to utilize neurobiological and clinical markers to understand the progression of HD before diagnosis and to provide candidate disease markers to assist future preventive HD clinical trials. Consent was obtained in accordance with the Institutional Review Board at each site. Controls were participants from HD families but who were free of the CAG‐expansion (i.e., CAG ≤ 35). Thirty‐four healthy controls (11 male/23 female, mean age 49.1, SD = 10.4) and 53 prodromal CAG‐expanded individuals were recruited from the HD Registries at the University of Iowa and the Cleveland Clinic. Prodromal CAG‐expanded individuals were stratified into low (n = 16; CAP < 287.16), medium (n = 22; 287.16 < CAP < 367.12), and high (n = 15; CAP > 367.12) groups based on their CAG‐Age Product or CAP designation, as previously described [Zhang et al., 2011]. CAP groups are used to reflect the individual's progression through the disease process, from presymptomatic through manifest HD, based on CAG and age. It is meant to encompass terms such as “disease burden” and “genetic burden” that have been used in previous literature. The formula for CAP is as follows:

$$\mathrm{CAP}={\mathrm{Age}}_0\times (\mathrm{CAG}-33.6600)$$

where Age₀ represents age of the participant at the time of scan for this study (i.e., baseline) [Zhang et al., 2011].

Measures

Participants were evaluated by clinicians experienced in the administration of the Unified Huntington's Disease Rating Scale (UHDRS) and certified by the Huntington Study Group (HSG). Formal diagnosis of HD was based on the Diagnostic Confidence Level rating of four indicating the examining clinician felt the participant showed “unequivocal presence of an otherwise unexplained extrapyramidal movement disorder” with ≥99% confidence [Huntington Study Group, 1996]. Participants with a rating of DCL = 4 were excluded to restrict this particular analysis to prodromal HD subjects. The sum of all the individual motor ratings from the UHDRS (total motor impairment score) is reported as well [Huntington Study Group, 1996]. Several cognitive measures were assessed alongside the imaging measures and included the Symbol Digit Modalities Test (SDMT), the Stroop Color Word Test, and the Trail Making Test (TMT). The SDMT measures psychomotor speed and working memory by counting the number of correct matches between numbers to their designated symbol based on a key [Smith, 1991]. The Stroop Color Word Test measures processing speed and executive functions by counting the number of correct responses in three conditions: color‐naming (name colors), word‐reading (read color names), and interference (inhibition of dominant reading response while naming color) [Stroop, 1935]. The TMT measures psychomotor speed and executive function by recording the time it takes participants to connect numbers alone (TMT Part A, TMTA) and connect alternating numbers and letters (TMT Part B, TMTB) both in ascending order [Reitan, 1958]. A greater time required to complete the TMT results in a higher score, which indicates worse performance or poorer function. A summary of participant characteristics is provided in Table 1.

Table 1.

Summary of demographic and clinical data for study participants

	Controls (mean; SD (N))	Low (mean; SD (N))	Medium (mean; SD (N))	High (mean; SD (N))
Age	49.1; 10.4 (34)	32.1; 8.8 (16)	39.4; 10.8 (22)	47.8; 12.2 (15)
Education (yr)	15.6; 2.0 (34)	14.8; 2.5 (16)	15.0; 2.2 (22)	13.7; 2.7 (15)
Gender	11M/23F (34)	3M/13F (16)	6M/16F (22)	2M/13F (15)
UHDRS Total Motor	4.7; 3.3 (32)	3.7; 2.7 (16)	7.1; 9.3 (22)	15.2; 10.5 (14)
SDMT	54.0; 11.0 (31)	55.1; 9.8 (15)	54.7; 12.5 (21)	46.2; 9.7 (15)
Stroop color	84.4; 11.7 (31)	84.1; 11.1 (15)	83.4; 13.8 (20)	64.8; 15.6 (14)
Stroop word	106.1; 18.1 (31)	106.9; 13.5 (15)	99.3; 18.3 (20)	77.6; 19.2 (14)
Stroop interference	49.2; 9.4 (31)	52.4; 13.4 (15)	50.5; 12.8 (20)	36.5; 10.9 (14)
TMTA	22.1; 5.6 (30)	20.9; 6.5 (15)	21.5; 7.5 (21)	27.6; 9.1 (15)
TMTB	55.2; 24.4 (30)	46.9; 16.6 (15)	52.4; 24.7 (21)	78.6; 38.3 (14)

UHDRS Total Motor Score = sum of all items of the Motor Assessment scale; SDMT = Symbol Digit Modalities Test; TMTA = Trail Making Test A; TMTB = Trail Making Test B.

Imaging

Imaging data were collected at two large medical research universities (University of Iowa and Cleveland Clinic). Both sites used a Siemens 3T TIM Trio scanner. Structural imaging consisted of T1‐ and T2‐weighted images both collected in the coronal plane. T1‐weighted images had the following parameters: TI = 900 ms, TE = 3.09 ms TR = 2,530 ms, flip angle = 10°, NEX = 1, bandwidth = 220 Hz/pixel, FOV = 256 × 256 × 220 mm, matrix = 256 × 256 × 220. T2‐weighted images had the following range of parameters: TE ≈ 440 ms, TR = 4800 ms, bandwidth = 590 Hz/pixel, FOV = 220 × 256 × 224 mm, matrix = 214 × 256 × 160 mm. A diffusion‐weighted sequence (71 noncollinear diffusion‐weighting gradients with diffusion‐weighting of b = 1,000 sec/mm² and eight b = 0 sec/mm² acquisitions, 256 × 256 mm FOV, 128 × 128 matrix, 50 2‐mm thick axial slices with zero gap, TE = 92 ms, TR = 7,700 ms (CCF) or 8,000 ms (Iowa), and bandwidth = 1,562 Hz/pixel (CCF) or 1,565 Hz/pixel (Iowa)) was acquired three times. All scans were transferred to The University of Iowa for processing and analysis.

Structural Image Preprocessing

Structural image preprocessing was performed using a derivative of the fully‐automated BRAINS (Brain Research: Analysis of Image, Networks, and Systems) AutoWorkup software package [Pierson et al., 2011]. T1‐ and T2‐weighted images for each subject were anterior commissure (AC)‐posterior commissure (PC) aligned. The AC‐PC‐aligned images were then bias‐field corrected using an atlas‐based classification algorithm. The preprocessed T1‐weighted images were used in FreeSurfer (version 5.1.0) image analysis suite (documented and freely available for download online at http://surfer.nmr.mgh.harvard.edu/) for volumetric segmentation of the cortical WM regions. An illustration of the WM labels used in this study is provided in Figure 1.

Figure 1.

(A) WM labels generated by FreeSurfer on T1‐weighted images shown in sagittal and axial views. Radiologic convention is used for the axial view (left is right, right is left). Each color represents the same region in both hemispheres: caudal middle frontal (white), frontal pole (dark brown), lateral orbitofrontal (light brown), medial orbitofrontal (red), pars opercularis (pink), pars orbitalis (yellow), pars triangularis (green), rostral middle frontal (blue), superior frontal (purple). (B) Significant CAP group differences in mean diffusivity (MD) in the left rostral middle frontal and right lateral orbitofrontal regions in comparison to controls (top). Significant CAP group differences in radial diffusivity (RD) in the left lateral orbitofrontal, left pars opercularis, left pars triangularis, left rostral middle frontal, right lateral orbitofrontal, right pars opercularis, right pars orbitalis, and right pars triangularis regions in comparison to controls (bottom). *P < 0.05, **P < 0.01, ***P < 0.005, ****P < 0.001, *****P < 0.0005.

Diffusion‐Weighted Image Preprocessing

Each DWI scan was visually inspected individually to identify artifacts. Repeat DWI scans from the same subject from a single scan session were concatenated (resulting in 3× redundancy of each gradient directions) before quality control checking with DTIPrep [Liu et al., 2010]. DTIPrep performs several quality assurance steps and removes volumes within a scan that do not meet its minimal quality criteria. DTIPrep first checked the diffusion imaging parameters to ensure the diffusion‐weighted gradients were in the directions as expected, along with image size and spacing. Intensity artifacts were then detected by comparing normalized correlation values of corresponding neighboring slices across all volumes within a scan. If a pair of slices possessed a normalized correlation value outside the designated number of standard deviations from the average normalized correlation value, the volume containing those slices was removed. Interlace artifacts were detected in a similar manner where a single normalized correlation value was calculated between interleaving slices for each volume. Motion between multiple baseline volumes was removed by rigidly registering and signal‐to‐noise ratio was maximized by averaging the registered baselines. The averaged baseline served as a reference for eddy‐current and head motion artifact correction through estimation of affine transforms between each diffusion‐weighted image and the averaged baseline. Directions of diffusion weighting were updated based on the rotational component of the affine transformation. The final quality assurance step removed any volumes that possessed residual motion or translation relative to the averaged baseline. The final dataset contained an averaged baseline image and only those diffusion‐weighted images that passed all quality assurance tests [Liu et al., 2010].

Imaging Variables in Regions of Interest

The output files from DTIPrep were used to estimate the tensor images, and subsequently the fractional anisotropy (FA), mean diffusivity (MD), axial diffusivity (AD), and radial diffusivity (RD) images were computed from the tensor images using components of the GTRACT software [Cheng et al., 2006]. A visual inspection of all FreeSurfer labels revealed that voxels posterior to the caudate were often included in the segmentation of both left and right medial orbitofrontal WM regions. To create a consistent regional definition, the medial orbitofrontal WM segmentation for each subject was edited by removing all voxels posterior to the centroid of the ipsilateral caudate. The FreeSurfer WM labels and brain masks were both resampled into DWI space using a B‐Spline transformation from the T2‐weighted image to the averaged b0 baseline image from the output DTIPrep file and visual inspections of registration quality were performed. The resampled FreeSurfer WM labels were used to obtain measurements of volume and mean rotationally invariant scalar measures. The ratio of label volume to intracranial volume will be referred to as the WM volume throughout the remainder of this manuscript. Mean FA, MD, AD, and RD values were computed in FreeSurfer WM labels intersected with the thresholded FA binary mask (subject's FA image containing FA values above 0.1) using components of SimpleITK (http://www.itk.org/Wiki/ITK/Release_4/SimpleITK).

Statistical Analysis

Statistical analyses were performed using general linear models (GLM) with PROC GLM in SAS 9.2. For each FreeSurfer defined PFC region, differences in WM volume and mean FA, mean MD, mean AD, and mean RD among groups determined by CAP designation were investigated using analysis of covariance models with age, years of education, gender, and site of data collection as covariates. Partial Pearson correlations were computed between regional WM volume, mean FA, mean MD, mean AD, and mean RD and SDMT, Stroop Word, Stroop Color, Stroop Interference, TMTA, and TMTB scores for prodromal HD subjects only with age, years of education, gender, and site of data collection as covariates. In the GLM and correlation analyses, a false‐discovery rate (FDR) correction was performed to adjust for multiple comparisons across ROIs using the procedures of Benjamini and Hochberg 1995 as implemented in PROC MULTTEST. FDR correction was used for the GLM omnibus test of any group difference. A criterion of q < 0.05 was used to elevate omnibus statistical significance, with q being the FDR‐adjusted P value. For each significant result based on the q value, unadjusted P values were used to evaluate pair‐wise group differences. A criterion of P < 0.05 was used to evaluate pair‐wise statistical significance.

RESULTS

GLM Groups Analysis

The results for the GLM group analysis are listed in Tables 2 through 6. In each table, the omnibus results are presented in three columns for mean FA (Table 2), MD (Table 3), RD (Table 4), AD (Table 5), and WM volume (Table 6). As the tables show, differences among groups that remained significant after FDR correction included those measuring diffusivity (Tables 3 and 4) as opposed to volume (Table 6). Model‐based group means (adjusted for covariates) for regions whose differences among groups that remained significant after FDR correction are plotted in Figure 1. Figure 1 illustrates that differences among groups that remained significant after FDR correction were mainly in regions of the inferior and lateral frontal lobe.

Table 2.

Summary of general linear model results, regional FA findings

Region	F valuea	Raw P value	FDR q value
Left caudal middle frontal	1.106	0.352	0.463
Left frontal pole	1.350	0.264	0.416
Left lateral orbitofrontal	2.520	0.064	0.230
Left medial orbitofrontal	2.372	0.077	0.230
Left pars opercularis	1.309	0.277	0.416
Left pars orbitalis	1.451	0.234	0.416
Left pars triangularis	3.244	0.026	0.158
Left rostral middle frontal	1.840	0.147	0.369
Left superior frontal	1.490	0.224	0.416
Right caudal middle frontal	0.509	0.677	0.677
Right frontal pole	0.901	0.445	0.500
Right lateral orbitofrontal	1.748	0.164	0.369
Right medial orbitofrontal	1.085	0.360	0.463
Right pars opercularis	3.333	0.024	0.158
Right pars orbitalis	2.991	0.036	0.161
Right pars triangularis	4.447	0.006	0.110
Right rostral middle frontal	1.001	0.397	0.476
Right superior frontal	0.704	0.553	0.585

^aThe F‐test reported in the table represents the main effect of CAP group from an analysis of covariance model that includes four groups (controls, low, medium, and high CAP groups) and age, years of education, gender, and site as covariates. df ₁ = 3 and df ₂ = 79 for all F‐tests.

Table 6.

Summary of general linear model results, regional WM volume findings

Region	F valuea	Raw P value	FDR q‐value
Left caudal middle frontal	0.247	0.863	0.971
Left frontal pole	0.321	0.810	0.971
Left lateral orbitofrontal	0.892	0.449	0.850
Left medial orbitofrontal	0.847	0.472	0.850
Left pars opercularis	2.654	0.054	0.488
Left pars orbitalis	0.996	0.399	0.850
Left pars triangularis	1.713	0.171	0.770
Left rostral middle frontal	0.470	0.704	0.917
Left superior frontal	0.490	0.690	0.917
Right caudal middle frontal	1.876	0.140	0.770
Right frontal pole	0.014	0.998	0.998
Right lateral orbitofrontal	0.491	0.690	0.917
Right medial orbitofrontal	0.457	0.713	0.917
Right pars opercularis	4.034	0.010	0.181
Right pars orbitalis	0.953	0.419	0.850
Right pars triangularis	1.002	0.397	0.850
Right rostral middle frontal	0.155	0.926	0.980
Right superior frontal	1.115	0.348	0.850

^aThe F‐test reported in the table represents the main effect of CAP group from an analysis of covariance model that includes four groups (controls, low, medium, and high CAP groups) and age, years of education, gender, and site as covariates. df ₁ = 3 and df ₂ = 79 for all F‐tests.

Table 3.

Summary of general linear model results, regional MD findings

Region	F valuea	Raw P value	FDR q value
Left caudal middle frontal	2.370	0.077	0.126
Left frontal pole	0.598	0.618	0.655
Left lateral orbitofrontal	4.181	0.008	0.051
Left medial orbitofrontal	2.553	0.061	0.118
Left pars opercularis	2.911	0.040	0.089
Left pars orbitalis	1.560	0.206	0.265
Left pars triangularis	3.647	0.016	0.055
Left rostral middle frontal	4.883	0.004	0.033
Left superior frontal	1.312	0.276	0.332
Right caudal middle frontal	2.971	0.037	0.089
Right frontal pole	0.102	0.959	0.959
Right lateral orbitofrontal	4.949	0.003	0.033
Right medial orbitofrontal	1.255	0.296	0.333
Right pars opercularis	3.933	0.011	0.051
Right pars orbitalis	2.498	0.066	0.118
Right pars triangularis	1.926	0.132	0.198
Right rostral middle frontal	3.537	0.018	0.055
Right superior frontal	1.847	0.145	0.201

^aThe F‐test reported in the table represents the main effect of CAP group from an analysis of covariance model that includes four groups (controls, low, medium, and high CAP groups) and age, years of education, gender, and site as covariates. df ₁ = 3 and df ₂ = 79 for all F‐tests.

Table 4.

Summary of general linear model results, regional RD findings

Region	F valuea	Raw P value	FDR q value
Left caudal middle frontal	2.889	0.041	0.067
Left frontal pole	0.436	0.728	0.771
Left lateral orbitofrontal	4.430	0.006	0.022
Left medial orbitofrontal	3.142	0.030	0.054
Left pars opercularis	4.186	0.008	0.025
Left pars orbitalis	2.024	0.117	0.176
Left pars triangularis	4.550	0.005	0.022
Left rostral middle frontal	4.657	0.005	0.022
Left superior frontal	1.793	0.155	0.186
Right caudal middle frontal	1.901	0.136	0.182
Right frontal pole	0.207	0.891	0.891
Right lateral orbitofrontal	5.091	0.003	0.022
Right medial orbitofrontal	1.162	0.330	0.371
Right pars opercularis	6.345	0.001	0.012
Right pars orbitalis	3.594	0.017	0.039
Right pars triangularis	3.788	0.014	0.035
Right rostral middle frontal	3.200	0.028	0.054
Right superior frontal	1.868	0.142	0.182

^aThe F‐test reported in the table represents the main effect of CAP group from an analysis of covariance model that includes four groups (controls, low, medium, and high CAP groups) and age, years of education, gender, and site as covariates. df ₁ = 3 and df ₂ = 79 for all F‐tests.

Table 5.

Summary of general linear model results, regional AD findings

Region	F valuea	Raw P value	FDR q value
Left caudal middle frontal	1.323	0.273	0.639
Left frontal pole	0.650	0.585	0.676
Left lateral orbitofrontal	3.007	0.035	0.131
Left medial orbitofrontal	1.082	0.362	0.656
Left pars opercularis	0.862	0.465	0.676
Left pars orbitalis	0.657	0.581	0.676
Left pars triangularis	1.289	0.284	0.639
Left rostral middle frontal	3.368	0.023	0.131
Left superior frontal	0.625	0.601	0.676
Right caudal middle frontal	3.581	0.017	0.131
Right frontal pole	0.024	0.995	0.995
Right lateral orbitofrontal	2.982	0.036	0.131
Right medial orbitofrontal	1.076	0.364	0.656
Right pars opercularis	0.793	0.502	0.676
Right pars orbitalis	0.858	0.466	0.676
Right pars triangularis	0.386	0.763	0.808
Right rostral middle frontal	3.161	0.029	0.131
Right superior frontal	1.534	0.212	0.637

^aThe F‐test reported in the table represents the main effect of CAP group from an analysis of covariance model that includes four groups (controls, low, medium, and high CAP groups) and age, years of education, gender, and site as covariates. df ₁ = 3 and df ₂ = 79 for all F‐tests.

As seen in Table 3, there were statistically significant differences in MD among groups in the left rostral middle frontal (q = 0.033) and right lateral orbitofrontal (q = 0.033) regions. Figure 1 (upper right) illustrates mean MD increased with CAP group, as shown by significantly higher MD values in the left rostral middle frontal region for both medium (P < 0.01) and high CAP (P < 0.005) groups and in the right lateral orbitofrontal region for the high CAP (P < 0.005) group in comparison to controls. As seen in Table 4, the left rostral middle frontal (q = 0.022) and right lateral orbitofrontal (q = 0.022) regions also had statistically significant differences in RD among groups, along with the left lateral orbitofrontal (q = 0.022) and all inferior frontal lobe regions (left pars opercularis, q = 0.025; left pars triangularis, q = 0.022; right pars opercularis, q = 0.012; right pars orbitalis, q = 0.039; right pars triangularis, q = 0.035) bilaterally except for the left pars orbitalis. RD also increased with progression. Most regions had significantly higher RD values for both medium (P < 0.01–0.05) and high CAP (P < 0.0005–0.01) groups in comparison to controls, except for the left pars opercularis (P < 0.005), right lateral orbitofrontal (P < 0.001), and right pars orbitalis (P < 0.01) regions that had higher RD values for the high CAP group only (Fig. 1).

Cognitive Variable Partial Correlations

After the application of FDR correction to all correlations between cognitive and imaging variables, TMTB was the only cognitive variable that showed significant partial correlation with two imaging variables in several regions. Amongst the regions that demonstrated significant differences in imaging variables among groups, the mean FA in two regions (right pars opercularis and right pars triangularis) in addition to the right medial orbitofrontal region negatively correlated with TMTB score (all q = 0.037). TMTB score also positively correlated with mean RD in the right pars triangularis region (q = 0.044) (Supporting Information Table 3). Complete summaries on the correlations between imaging variables and the SDMT (Supporting Information Table 1), TMTA (Supporting Information Table 2), TMTB (Supporting Information Table 3), Stroop Word (Supporting Information Table 4), Stroop Color (Supporting Information Table 5), and Stroop Interference (Supporting Information Table 6) can be found in Supporting Information Tables 1 through 6.

DISCUSSION

The main goal of this study was to build upon past prodromal HD studies on the frontal lobe by examining focused regions of PFC WM in prodromal HD individuals using four commonly used measures of diffusivity (FA, MD, RD, and AD) and WM volume. Mean measures of diffusivity and WM volume for each region were compared across four groups (controls and three prodromal HD groups) and correlated with several measures of cognitive performance. In this study, much like the differences in cognitive performance seen in prodromal HD subjects at varying stages before diagnosis [Stout et al., 2011], statistically significant increases in MD and RD in CAP groups relative to controls were seen in inferior and lateral PFC regions. In comparison to controls, a gradient of effects was seen in MD and RD, where the smallest effect was seen in the low group and the largest effect in the high group. Significant correlations between TMTB score and mean fractional anisotropy (FA) and/or RD paralleled the group differences in mean MD and/or RD in several right hemisphere regions. The gradient effect of lower anisotropy with CAP group could be explained by larger axon diameter or lower packing density of axons that both discourage anisotropic diffusion [Takahashi et al., 2002]. Specifically, significant differences in RD in the presence of no findings in AD has been seen in an animal study that attributed this effect to demyelination [Song et al., 2002]. In addition, changes in diffusivity that reflect a loss of directionality in diffusion seen in two other animal studies (lower FA, higher RD) [Song et al., 2003, 2005] were demonstrated in the same regions of the right lateral PFC that showed group differences in MD and RD and correlated with a poorer performance on one of the cognitive tests used in this study (TMTB). In summary, this study detected changes in diffusivity for the first time in a region that has not been closely examined in the context of prodromal HD. The meanings of these changes in diffusivity were further supported by correlating with scores on a cognitive test (TMTB) that has a documented ability to detect cognitive deficits in prodromal HD subjects. The gradient of effects suggests DWI can provide reliable markers of disease progression in the form of increasing diffusivity changes in the lateral PFC of prodromal HD individuals. Therefore, the results of this study suggest that mean RD in regions of the right lateral PFC could serve as a reliable biomarker to monitor disease progression in the prodromal HD stage.

The lack of findings for FA and MD in this study emphasizes the importance of investigating directional measures of diffusivity in addition to rotationally invariant diffusivity measures. FA and MD are commonly used measures of diffusivity because they summarize general shape and magnitude of diffusion, respectively, by accounting for diffusion magnitudes along three orthogonal directions at once [Basser and Pierpaoli, 1996]. The three orthogonal directions are numbered as eigenvectors based on the descending order of their corresponding diffusion magnitudes (first, second, and third eigenvalues) [Basser, 1995; Basser and Pierpaoli, 1996]. In comparison to other summary measures of diffusivity (e.g. volume ratio and relative anisotropy), FA is less susceptible to noise and provides the highest signal‐to‐noise ratio (SNR) [Papadakis et al., 1999]. However, when changes in diffusion are subtle and in one or two of the orthogonal directions, these changes may not be reflected in summary measures that normalize or average across all diffusion magnitudes. It may be more helpful to examine these subtle changes using directional measures of diffusivity to see diffusion magnitudes perpendicular and parallel to the first eigenvector. For example, Acosta‐Cabronero et al. demonstrated increases in AD, RD, and MD that were more highly significant and sensitive to white matter changes in early Alzheimer's patients than reductions in FA [Acosta‐Cabronero et al., 2010]. In addition, the increases in AD, RD, and MD were located in areas where tract degeneration was expected to occur based on prior gray matter lesion studies, further challenging the notion that reduced FA alone is able to fully capture changes in axonal integrity in Alzheimer's disease [Acosta‐Cabronero et al., 2010].

When using measures of directional diffusivity, it is common to see changes in both RD and AD in a given region because the processes that cause changes in these measures (axonal death and demyelination) often occur in close proximity [Song et al., 2003]. As mentioned earlier, AD describes diffusion along the largest eigenvector [Basser and Pierpaoli, 1996]. Animal studies have demonstrated that a decrease in AD is associated with axonal injury and degeneration because normal parallel diffusion along axons is being hindered by dysfunctional tissue [Song et al., 2003]. RD describes diffusion perpendicular to the first eigenvector. In contrast to AD, an increase in RD is associated with demyelination since diffusion perpendicular to the axon is increased when there is less myelination [Song et al., 2003, 2005].

In this study, only increases in RD were seen in the prodromal HD individuals. It is important to remember that since RD is the mean of two eigenvalues it will be less noisy than AD, a measure that consists of a single eigenvalue. Therefore, in this study RD may have been more sensitive to tissue changes and AD has yet to reach significance. An increase in RD with no change in AD has been documented in a very specific type of myelin pathology called dysmyelination [Song et al., 2002]. Dysmyelination is the incomplete myelination of functional axons, as opposed to demyelination that is the complete loss of myelination [Song et al., 2002]. Song et al. examined diffusivity changes in the setting of dysmyelination by using Shiverer mice. Shiverer mice are homozygous for a recessive autosomal mutation for myelin basic protein, causing incomplete myelination in the central nervous system [Inoue et al., 1981; Privat et al., 1979; Rosenbluth, 1980; Shen et al., 1985]. Song et al. showed that major WM tracts in Shiverer mice have increased RD but identical AD in comparison to the same tracts in control mice [Song et al., 2002]. At this point in time, it is not possible to unambiguously interpret increased RD in the lateral PFC without AD changes in this study as a dysmyelintation process in prodromal HD individuals without longitudinal or histological data to pinpoint the exact process affecting diffusivity [Jones et al., 2012]. However, it must be emphasized that this study demonstrated a consistent gradient effect of increased RD without AD changes throughout the lateral PFC bilaterally.

The lack of WM volume findings in this study was initially surprising, given that previous studies have shown decreases in WM volume in prodromal HD [Aylward et al., 2011] and correlations between morphological abnormalities and cognitive deficits in early HD subjects [Beglinger et al., 2005]. It must be noted that if abnormal WM volume findings in the literature are specific to the frontal lobe (either prodromal or symptomatic HD), they tend to be in the entire frontal lobe [Aylward et al., 1998, 2011; Halliday et al., 1998]. An aspect of this study that could have prevented frontal lobe WM volume findings was that a precentral gyrus region was not included in any part of the analysis. Perhaps WM volume abnormalities in prodromal HD individuals are specific to the precentral gyrus. It was not possible to include the precentral gyrus because the DWI data used here did not consistently include the most superior portions of the frontal and parietal lobes. Although the FreeSurfer WM definition has been shown to produce similar mean FA values in the same WM regions defined by other methods [Fjell et al., 2008; Tamnes et al., 2010] and variability within regions are replicated across groups [Salat et al., 2009], it still may not be the true volume of WM associated with the cortical region.

In this study, significant group differences relative to controls in MD and RD were mostly located in the lateral PFC, specifically in the ventrolateral or lateral left and right inferior regions. Traditionally, the left inferior frontal area is known to be involved in language, where lesions to the posterior portion cause Broca's aphasia [Fuster, 2009]. However, the lateral inferior regions are broadly implicated in a number of higher‐order executive processes [Perry et al., 2011]. Therefore, the significant correlations between FA, RD, and TMTB in most of the same right inferior regions containing group differences may further suggest a link between the lateral inferior regions and higher‐order executive processes. Overall, the TMT is a cognitive measure that has a documented ability to differentiate among prodromal HD individuals in the CAP groups considered in this analysis [O'Rourke et al., 2011]. Specifically, TMTB involves subjects connecting alternating letters and numbers to test cognitive flexibility and working memory, where the score is the time necessary to complete the task. Poor performance on the TMTB is reflected in a longer completion time [O'Rourke et al., 2011]. TMTB scores negatively correlating with FA and positively correlating with RD in this study can possibly be interpreted together as prodromal HD individuals experiencing greater impairment in executive functioning, processing speed, and working memory with disease progression due to a white matter disease process that can be detected by measures of diffusivity [Jones et al., 2012].

As for the significant findings in the dorsolateral and orbitofrontal PFC regions, these results may be explained by the dorsal‐to‐ventral progression cell death in the striatum observed in HD [Hedreen and Folstein, 1995] affecting components of corticostriatal loops [Lawrence et al., 1998]. Specifically, Lawrence et al. hypothesized that functions associated with the dorsal PFC‐striatal loop may be impaired before motor symptom onset, followed by impairment of functions associated with the ventral loop as neuronal loss increases with disease progression [Hedreen and Folstein, 1995; Lawrence et al., 1998]. Based on anatomical studies done by Alexander et al. and Arikuni et al., the dorsal PFC striatal loop includes dorsolateral PFC projections to the central to dorsal caudate, while the ventral loop includes orbitofrontal PFC projections to the ventromedial caudate [Alexander et al., 1986; Arikuni and Kubota, 1986]. Therefore, the significant findings in the dorsolateral (increased MD and RD in the left rostral middle frontal region) and the orbitofrontal (increased MD and RD in the right lateral orbitofrontal and increased RD in the left lateral orbitofrontal regions) PFC in this study may be explained by the pattern of cell death in the striatum affecting components of the corticostriatal loops as implied by changes in diffusivity [Jones et al., 2012].

The main limitation of this study was that WM regions of the PFC were only explored with WM volume and a limited set of scalar diffusivity measures derived from the tensor model. Another metric for detecting differences in diffusivity among tissue types in future studies of white matter integrity in Huntington's Disease is diffusional kurtosis imaging (DKI). Diffusional kurotosis values quantify diffusional non‐Gaussianity as a consequence of tissue structure creating barriers and compartments [Jensen et al., 2005]. Diffusional kurtosis values may provide greater sensitivity to differences between largely isotropic tissues and have been used in ischemic stroke [Helpern et al., 2009; Lätt et al., 2009], aging [Falangola et al., 2008], schizophrenia [Ramani et al., 2007], and attention deficit disorder [Helpern et al., 2007]. Analyzing WM regions derived from WM fiber tracts that connected cortical gray matter to the striatum instead of WM regions based on proximity to cortical gray matter would have provided a means for specifically examining corticostriatal tracts. Additionally, using methods more sophisticated than the tensor model, such as high angular resolution diffusion imaging (HARDI) to resolve multiple fiber orientations in white matter containing crossing fibers, would be important to examine in the future with the number diffusion‐weighted gradients per scan used in this study [Tuch et al., 2002]. The additional information on multiple fiber orientations per voxel could possibly make scalar diffusivity measures more sensitive to changes in white matter and assist with more reliable fiber tract reconstructions in future studies [Tuch et al., 2002]. Another limitation of this study was the incomplete coverage of superior frontal and parietal lobes in DWI scans mentioned earlier that led to clipping in the superior frontal, caudal middle frontal, and rostral middle frontal regions. The rostral middle frontal areas were the least affected and perhaps that is why findings were strongest there. However, it is uncertain whether the superior frontal and caudal middle frontal regions contain findings in this study as these regions were visibly substantially clipped.

Future directions include expanding upon these findings in the PFC in the form of more complex analyses. The next step is to perform cross‐sectional fiber tracking to obtain representations of the PFC WM that can be analyzed for changes along each WM region. In addition to WM in the PFC, it may also be useful to examine WM extending to the PFC from the striatum and beyond to characterize how HD affects corticostriatal loops in their entirety. Ultimately, the above analyses will be expanded to characterize changes in individual subjects longitudinally.

CONCLUSION

The main goal of this study was to build upon past prodromal HD studies on the frontal lobe by examining focused regions of PFC WM in three groups of prodromal HD individuals stratified by baseline progression (low, medium, and high groups) using four commonly used measures of diffusivity (FA, MD, RD, and AD) and WM volume. In summary, this study was able to detect differences in diffusivity based on baseline disease progression for the first time in the lateral PFC, a region that has not been closely examined in the context of prodromal HD. The meaning of these changes in diffusivity were further supported by correlating WM measures with scores on a cognitive test that has a documented ability to detect cognitive deficits in prodromal HD. Therefore, the results of this study suggest that mean RD in regions of the right lateral PFC could serve as a reliable biomarker to monitor disease progression in the prodromal HD stage in future longitudinal studies.

Supporting information

Supporting Information Tables.