Insular atrophy at the prodromal stage of dementia with Lewy bodies: a VBM DARTEL study

Diffuse atrophy including the insula was previously demonstrated in dementia with Lewy bodies (DLB) patients but little is known about the prodromal stage of DLB (pro-DLB). In this prospective study, we used SPM8-DARTEL to measure gray matter (GM) and white matter (WM) atrophy in pro-DLB patients (n = 54), prodromal Alzheimer’s disease (pro-AD) patients (n = 16), DLB patients at the stage of dementia (mild-DLB) (n = 15), and Alzheimer’s disease patients at the stage of dementia (mild-AD) (n = 28), and compared them with healthy elderly controls (HC, n = 22). Diminished GM volumes were found in bilateral insula in pro-DLB patients, a trend to significance in right hippocampus and parahippocampal gyrus in pro-AD patients, in left insula in mild-DLB patients, and in medial temporal lobes and insula in mild-AD patients. The comparison between prodromal groups did not showed any differences. The comparison between groups with dementia revealed atrophy around the left middle temporal gyrus in mild-AD patients. Reduced WM volume was observed in mild-DLB in the pons. The insula seems to be a key region in DLB as early as the prodromal stage. MRI studies looking at perfusion, and functional and anatomical connectivity are now needed to better understand the role of this region in DLB.

storage impairment than DLB 10,11 . These findings are in keeping with the cognitive profiles of AD and DLB patients when the dementia becomes manifest 12 , although the neuropsychological pattern in prodromal DLB (pro-DLB) has been reported as more heterogeneous than in pro-AD 11 .
Nevertheless, early identification of DLB, particularly in the prodromal phase (i.e., pro-DLB), will be highly relevant to the development and testing of future disease modifying treatments and thus there is an urgent need to develop viable and sensitive biomarkers that can detect DLB in its early stages. Furthermore, determination of early biomarkers in DLB is necessary to guide the operationalization of future consensus criteria for pro-DLB 8 .
Structural neuroimaging represents one potential biomarker area: in particular voxel-based morphometry (VBM) is a classical approach to detect brain atrophy. In AD compared to healthy elderly controls, VBM studies described atrophy of the entorhinal cortex and hippocampus as early as the prodromal stage of AD [13][14][15][16][17] , which was also shown to be a good predictor of cognitive decline and conversion to dementia 15,[18][19][20][21] . In AD at the stage of dementia, in addition to the medial temporal lobe, other atrophies were reported in temporal, parietal, posterior cingulate and frontal cortices as well as insula and rolandic operculum 14,[22][23][24] . Previous VBM studies in DLB at the stage of dementia (mild-DLB) have demonstrated heterogeneous results. The Newcastle team showed a diminished gray matter (GM) volume in the lateral temporal lobes, frontal lobe, insulae, and precuneus 25 . On the other hand, the Mayo Clinic team showed atrophy in midbrain, substantia innominata, hypothalamus and right insula 23 . A voxel-wise meta-analysis on cortical atrophy in mild-DLB patients found bilateral insula and basal ganglia atrophy 26 . We recently demonstrated in a multicentre study that the cortical thickness of pro-DLB patients is diminished in the right anterior part of the insula 22 . This first result is of high interest since patients at the early stage of the disease frequently present neurovegetative disorders such as constipation, orthostatic dizziness, or increased saliva 27 , which could be linked to insula dysfunction 28 . Pro-DLB patients also frequently have fluctuations and hallucinations that could also be linked to the insula 29,30 . In contrast, parkinsonism is rarely obvious at the beginning of DLB, and such symptoms are known to be linked to basal ganglia and particularly the substantia nigra. While cortical thickness and VBM are both related to gray matter, they are not fully consistent 31 . Moreover, we aimed to compare patients that have been screened in a homogeneous way.
When assessing white matter (WM) integrity, a meta-analysis 32 showed that AD was mainly characterized by WM medial temporal lobe atrophy, although monocentre studies describe reduced WM in other parts of the brain such as the parietal 33,34 and frontal 35 lobes, together with the inferior longitudinal fasciculus 33 and corpus callosum 32,33 . To our knowledge, compared to GM atrophy, VBM-based WM atrophy has been less investigated in DLB. Lee et al. 36 reported a loss of WM at the stage of dementia in mainly the posterior areas, but at an uncorrected statistical threshold. Only two studies have compared WM between DLB and AD: although the first study 34 did not showed any reduction of WM volume in DLB, the second study 37 showed an atrophy in DLB in the dorsal midbrain, pons and the cerebellum. No investigations have been performed yet at the prodromal stage.
Therefore, the aim of this study was to investigate, in a monocentre study, GM and WM atrophy patterns in pro-DLB and in mild-DLB patients. To this end, GM and WM volumes were compared to those of healthy elderly controls (HC) and those of AD patients at either the prodromal stage (pro-AD) or the mild dementia stage (mild-AD). According to patterns of atrophy in both the prodromal and dementia stages in AD and DLB, we made the following assumptions. First, we hypothesized that, compared to HC, pro-DLB patients would have GM atrophy predominantly in the anterior insula, but pro-AD patients would have GM atrophy in the entorhinal cortex and the hippocampus. As a consequence, we expected that the comparisons of the two prodromal groups of patients would provide a similar pattern of atrophy. At the stage of dementia, we hypothesized that the GM atrophy of insula in DLB patients would enlarged to the entire insula, and expand to basal ganglia and frontal cortex, whereas the medial temporal lobe atrophy in AD patients would spread to insula, temporal, posterior cingulate and frontal cortices. Accordingly, we expected that the comparisons of DLB and AD patients at the stage of dementia would not lead to statistical differences in the insula. If present, we also expected slight WM atrophy in the brainstem for prodromal DLB patients and in the medial temporal lobe for prodromal AD patients, which would enlarge at the stage of dementia.

Material and Methods
Participants. Information about the goals of the prospective longitudinal AlphaLewyMA study is freely available on the Clinical Trials web site (http://clinicaltrials.gov/ct2/show/NCT01876459). Briefly, the primary end point was to assess synuclein in the cerebrospinal fluid, and the secondary end point, which concerned the present study, was to create biomarker for DLB and AD using MRI. One hundred forty-four patients with either DLB or AD were consecutively recruited from the Memory Resources and Research Center of the University Hospital of Strasbourg, France, together with 22 elderly healthy controls. Nine participants were excluded due to motion during the MRI acquisition (n = 9). The remaining 135 patients were classified according to their diagnosis and the level of their cognitive deficit. Consequently, the study comprised 54 patients with pro-DLB (pro-DLB group), 15 patients with DLB at the dementia stage (mild-DLB group), 16 patients with pro-AD (pro-AD group), 28 patients with AD at the dementia stage (mild-AD group), and 22 healthy elderly controls (HC group). Demographic data are presented in Table 1. Patients presenting both DLB and AD clinical features were excluded from the analysis.
The five groups were examined by clinicians with expertise (tertiary center) in dementia to perform a complete anamnesis and medical examination. Using the Unified Parkinson's Disease Rating Scale (UPDRS) III score 38 , akinesia, rigidity, and tremor at rest were rated from 0 to 4 (0: no symptoms to 4: serious impairment). Fluctuations were assessed with the Mayo Clinic Fluctuations scale (patients with a score higher than or equal to two were considered as having fluctuations) and the Newcastle-upon-Tyne Clinician Assessment of Fluctuation scale. Cognitive functions were evaluated, and the cognitive profile of the pro-DLB patients of the AlphaLewyMA group has been reported by Kemp et al. 39 .
An etiologic diagnosis of the neurocognitive disorder for each patient was made using Dubois' criteria for pro-AD (i.e., MCI AD) and mild-AD 7 , and McKeith's criteria (probable DLB, i.e. at least two core symptoms) for mild-DLB 1 . Pro-DLB patients were defined as patients with MCI (according to Petersen criteria 40 ), preservation of independence (assessed by the Instrumental Activities of Daily Living [IADL]) and by McKeith's criteria (meeting probable DLB criteria except presence of dementia) 1 . All patients had formal assessment of their diagnosis by three independent expert clinicians. Patients with concomitant AD and DLB, i.e., meeting both McKeith's (for probable DLB) and Dubois' criteria, were also excluded.
The control group consisted in elderly healthy and cognitively intact (no MCI) subjects who were recruited via advertisements in local community newsletters in Strasbourg, and via the listing of controls of the local clinical investigation center ("Centre d'Investigation Clinique") in charge of any type of medical research of the University Hospital of Strasbourg. In the Memory Resources and Research Center, controls underwent similar clinical and cognitive assessments than patients to exclude any who may have had occult MCI or dementia.
Exclusion criteria for participation in the study included contraindications for MRI, history of alcohol/substance misuse, evidence suggesting alternative neurological or psychiatric explanations for their symptoms/cognitive impairment, focal brain lesions on brain imaging, and the presence of other severe or unstable medical illness.
The study was approved by the local Ethics Committee (Comité de Protections des Personnes, Est IV, Strasbourg, France), and all experiments were performed according to the guidelines and regulations provided by this committee. Controls and patients gave written informed consent.

Data acquisition. A 3D MPRAGE T1-weighted image was acquired on a Siemens Verio 3 T scan-
ner equipped with a 32-channel head coil (Siemens, Erlangen Germany). Parameters were: sagittal orientation; repetition time = 1900 ms; echo time = 2.53 ms; inversion time = 900 ms; flip angle = 9°; imaging matrix 192 × 192 × 176; 1 mm 3 isotropic voxel; acquisition time = 7 min 38 s. The image was acquired such that the axial slice orientation was aligned with the AC-PC line.
Data processing. Images were processed using SPM8 (Welcome Department of Cognitive Neurology, London, UK). According to the algorithm of the New Segment toolbox of SPM8, images were first segmented into five tissue probability maps in the native space. Two GM DARTEL templates were then computed from GM and WM probability maps of either a subset that included the pro-DLB, pro-AD and HC groups, or another subset that included the mild-DLB, mild-AD and HC groups. The GM and WM probability maps of each participant were spatially normalized to the Montreal National Institute (MNI) space according to the transformation parameters from the corresponding DARTEL template (since healthy participants were including in the two subsets, the normalization was processed twice according to each DARTEL template). During this procedure, images were modulated, then smoothed with a Gaussian filter of full width at half maximum of 8 mm.
Statistical analysis. Patients with either MCI or dementia were analyzed separately, i.e. a first set of analyses compared the pro-DLB, pro-AD and HC groups, whereas another compared the mild-DLB, mild-AD and HC groups.
Analyses were performed on Matlab (R2014b, Mathworks, Natick, MA). The distribution of gender and age values between groups was assessed by chi-squared tests and a one-way ANOVA, respectively. Prior to ANOVA, homoscedasticity and normality of the distribution of age values were verified by Levene's test and Lilliefors tests to ensure that our conditions really met the assumptions of ANOVA as a parametric test. Post-hoc analyses were performed using Tukey-Kramer multi-comparison test.
Using the randomize function of FSL 5.0 (FMRIB Analysis Group, Oxford, UK), non-parametric permutation tests with threshold-free cluster enhancement were performed to compare GM and WM volumes between groups, by running 5,000 permutations with variance smoothing of 8 mm. Age, gender and total intracranial volume were regarded as regressors of non-interest. The total intracranial volume was assessed in the patient's space (i.e., prior to spatial normalization) by summing thresholded GM, WM and CSF probability maps (threshold = 0.2) followed by a count of non-zero voxels. Statistical values were corrected for multiple comparisons and thresholded at p corrected < 0.05. Unthresholded corrected maps in nifti format are freely available for consultation and download at http://neurovault.org/collections/2466. Since we used whole-brain analysis, we verified that each significant cluster certainly arose from the tissue map of whom it depended so that white matter changes were not due to gray matter changes for instance. All clusters were anatomically defined according to the AAL atlas.

Results
Demographic data. Demographic data are presented in Table 1. According to results from ANOVA and post hoc analysis in the subset with prodromal patients (F[2,89] = 5.32; p < 0.01), the pro-AD group was significantly older than HC (p < 0.01) and we noted a trend compared to the pro-DLB group (p = 0.062), but the pro-DLB group did not differ from HC. No significant difference in gender was observed, but pro-AD patients were less likely to be female than pro-DLB patients (χ 2 , p = 0.066). The total intracranial volume did not differ between these three groups (ANOVA, F[2,89] = 0.08). An age effect was observed in the subset with patients with dementia (ANOVA, F[2,62] = 6.26; p < 0.01): HC were younger than mild-DLB (p < 0.05) and mild-AD patients (p < 0.01) but mild-DLB patients did not differ from those with mild-AD. There was no gender effect. The total intracranial volume differed between groups (ANOVA, F[52,62] = 3.37; p < 0.05) since the mild-AD group had smaller values than the mild-DLB group (p < 0.05) but did not differ from HC. Total intracranial volume did not differ between mild-DLB patients and HC.

VBM-DARTEL
Prodromal patients. Unthresholded statistical images are freely available for consultation and download at http://neurovault.org/collections/2466 (the viewer has a threshold slider for which we recommend to set the left value to 0). Pro-DLB patients had a reduced GM volume in right medial frontal gyrus, and in bilateral insula and right claustrum relative to controls (Table 2, Fig. 1). Pro-AD patients did not differ from HC at p corrected < 0.05, although a tendency towards a GM loss in right hippocampus was observed at a lower statistical threshold (p corrected = 0.07 and p uncorrected = 0.0002). Comparisons between prodromal groups did not reveal any decrease of GM volume. Controls did not show any atrophy that exceeded that in prodromal patients. The effect of age did occur either in the insula or in the medial temporal lobe.
In terms of WM volume, no differences between groups were evidenced. The effect of age occurred in all lobes of the brain.
Patients at the stage of dementia. Unthresholded Fig. 2). In mild-AD patients compared to HC, the atrophy occurred extensively in parahippocampal gyrus, amygdala and hippocampal complex, temporal pole, but also insula and putamen, all bilaterally. In mild-AD patients relative to mild-DLB patients, left middle temporal gyrus was atrophied in mild-AD patients, whereas no significant differences were apparent in the mild-DLB group relative to the mild-AD group. In HC, no GM losses were apparent that exceeded that in patients at the stage of dementia. The effects of age did not occur either in the insula or in the medial temporal lobe. In terms of WM volume, mild-DLB patients had a loss of WM in the right part of the pons compared to HC (Fig. 3. Size in mm3: 556; mean p-value: 0.046; coordinates in the MNI space [x, y, z]: 8, −17, −29). No other differences reached statistical significance. The effect of age occurred in the frontal cortex.

Discussion
Using VBM analysis, this study sought to assess patterns of GM atrophy in DLB at both prodromal and dementia stages compared to healthy elderly controls and patients with AD. Our assumptions were the followings: (1) pro-DLB patients would have GM atrophy predominantly in the anterior insula, but pro-AD patients would have GM atrophy in the entorhinal cortex and the hippocampus in such a way that patients group comparison would provide a similar pattern of atrophy; (2) at the stage of dementia, the GM atrophy of insula in DLB patients would enlarged to the entire insula, and expand to basal ganglia and frontal cortex, whereas the medial temporal lobe atrophy in AD patients would spread to insula, temporal, posterior cingulate and frontal cortices, and patients group comparison would provide a similar pattern of atrophy. If present, we also expected slight WM atrophy in the brainstem for prodromal DLB patients and in the medial temporal lobe for prodromal AD patients, which would enlarge at the stage of dementia. We report that early atrophy in DLB involved bilateral insula whereas it occurred rather in the medial temporal lobe in pro-AD (although at a lower statistical threshold). Comparisons of pro-DLB and pro-AD did not show differences in these structures. When the pathology worsened, the pattern of atrophy enlarged with deeper atrophy at the stage of mild dementia. Whereas in mild-DLB volume loss was observed in insula but not in hippocampus, mild-AD showed atrophy in both hippocampus and insula. Comparisons of patients with mild dementia revealed left middle temporal gyrus atrophy in AD relative to DLB.
The pattern of atrophy that we found in AD patients is consistent with previous studies reporting early atrophy of the medial temporal lobe, supporting as a consequence our hypotheses. The entorhinal cortex has been reported as reduced as early as the subjective cognitive impairment 13 , then the progression reached hippocampus in pro-AD [13][14][15][16][17] , and AD patients with dementia 14,41 . Such losses of GM were shown to be a good predictor of cognitive decline and conversion to dementia 15,[18][19][20][21] . In addition to entorhinal cortex and hippocampus, we report a loss of gray matter in parahippocampal gyrus and amygdala in mild-AD, which has already been described in MCI 14,21,42 , although the atrophy occurred also in other brain structures known to be impaired in AD, such as the temporal pole, the inferior and superior temporal gyrus 23,41,43 , the fusiform gyrus 43 , but also the putamen and insula 43 .
Attesting to our hypothesis, atrophy in DLB mainly occurred in the bilateral insular cortex as early as the prodromal stage, and also included the claustrum. The deterioration of the insular cortices in DLB at the stage of dementia has been described in some previous studies 23,25 . Moreover, a meta-analysis on GM atrophy in DLB patients at the stage of dementia found an atrophy of bilateral insula and basal ganglia 26 . However, we are the first to our knowledge to describe the occurrence of insular atrophy in DLB early in the development of the pathology. Moreover, the significant clusters were well localised and demonstrate that both the anterior and posterior parts of the insulae are impaired. This pattern of early atrophy does not support our assumption of an anteroposterior evolution of insular atrophy, although further longitudinal studies will provide new information about its. The topological distinction between anterior and posterior insula (sometimes subdivided into three: anterior, middle and posterior parts) was done according to cytoarchitectonic 44 and histological 45 analyses, task-based functional MRI 46 , functional connectivity MRI [47][48][49] , structural connectivity MRI 50-52 , or even multimodal imaging 53 . Briefly, these studies reported that anterior insula is connected to prefrontal and anterior cingulate cortices, whereas posterior insula is rather connected to motor cortex, superior temporal gyrus and the posterior parts of the brain. Accordingly, they also demonstrated that anterior insula is involved in emotional processes, saliency, attention and cognitive control, whereas posterior insula would be implicated in somatosensory and motor processing (including interoception and visceral motor, although other studies reported interoception involving the anterior insula 28,54 ). These data about insula are consistent with the known cognitive pattern of DLB at the early stage, including executive function impairment 55 and attention deficit 8,11,56 . Moreover, stimulations of the anterior insula in epileptic patients were responsible for clinical responses including motor modifications such as tremor, hallucinations, and neurovegetative symptoms 28 . These exceptional clinical experiments are consistent with the known symptoms of DLB even at the prodromal stage 8 . In the same way, stimulation of a particular region between the anterior insula and the claustrum is responsible for impairment of consciousness, with staring, unresponsiveness, and slowing, and sometimes with confusion 57 , which are fully consistent with the known cognitive and alertness fluctuations in DLB. The claustrum, which is atrophied in the left hemisphere of mild-DLB patients together with the insula, is also involved in visual processing 58,59 and is connected to the occipital cortex 60,61 , suggesting that the insula and the claustrum could be involved in the visual disorders in DLB 62 . According to our results and those of previous studies on DLB, and in view of the functions of the insula and its connections to other brain areas, the insula in DLB seems to be (1) affected at an early stage, (2) the main atrophy at the stage of dementia, and (3) related to the clinical profile of DLB, making the insula a key region for further investigations on DLB. Still, insula alone may not be all responsible for the disorders in DLB that we related to insula, but may play an essential role.
Compared to healthy controls, we found that atrophy in AD started with medial temporal lobe atrophy, then the pattern enlarged to other brain areas, including insula. In contrast, DLB started with insula atrophy and patients did not develop medial temporal atrophy, although we cannot rule out the possibility that more seriously affected DLB patients might have a decrease of GM volume in the medial temporal lobe 63 . In contrast to our assumptions, the comparison between DLB and AD did not revealed any loss of GM in AD in the medial temporal lobe but in the middle temporal gyrus, and no structure were more significantly atrophied in DLB compared to AD. Even though these results are in agreement with the clinical profile of patients, they do not support the hypothesis of a dissociation between an early atrophy of the medial temporal lobe in AD, which is followed by atrophy of the insula, and an early atrophy of the insula in DLB without loss of GM in the medial temporal lobe. While relevant, exploring this hypothesis is outside the scope of our study and will require longitudinal investigations. At the stage of dementia, both AD and DLB had atrophy of insula, and this might relates to the clinical symptoms that could be shared between DLB and AD.
Comparisons of WM volumes demonstrated only a reduction in the pons in mild-DLB compared to HC according to our hypothesis, which is in line with previous comparisons between DLB and AD 37 . Since this part of the brain stem is involved in arousal, autonomic functions, sleep, and acts as a relay between the cerebrum and the cerebellum, its atrophy would support some of the clinical signs of DLB, such as parkinsonism, Rapid Eye Movement sleep behavior disorder 11,64 , dysautonomia and fluctuations 65 . Using fractional anisotropy or mean diffusion might provide additional information about WM changes in DLB, overcoming the restricted sensitivity of WM VBM analyses, and providing in addition a description of WM microstructural changes.
While the present results are consistent with the literature, there are still some limitations in our study. First, notwithstanding the statistical analyses accounting for the sample size, the groups are small, limiting the generalization of the results to a larger population. However, the cohort of pro-DLB is to our knowledge the largest to date in an MRI study. Second, age was not equally distributed across groups, even though we limited this effect by including it as a covariate in the statistical model. At the prodromal stage of DLB, patients did not differ in age from HC but were atrophied in insula. Moreover, no effect of age on GM and WM volumes were observed in insula and in the part of the pons which was atrophied in pro-DLB patients, respectively. Accordingly, the atrophies in DLB were not mistaken effects due to age but related to structural modifications due to DLB.

Conclusion
This paper describes a loss of GM volume of bilateral insula in DLB, which started as early as the prodromal stage. According to the functions of the insula and its relationship with other brain areas, this atrophy is consistent with the clinical and cognitive profile of patients with DLB. In contrast, patients with AD were rather characterized by a decrease of GM in the medial temporal lobe, as described in several previous studies. Nevertheless, in contrast to our hypothesis, comparisons between DLB and AD revealed external temporal atrophy in AD. A potential dissociation between DLB and AD atrophy at the early stage, i.e. starting with either insula or medial temporal lobe, has still to be explored. Since the insula seems to play a key role in DLB, the repercussions of its impairment on the functioning of other brain areas, and on cognitive deficits as a consequence, should be further explored. Functional and structural connectivity, perfusion, and their correlations with cognitive performances could be such an approach.