Dementia with Lewy bodies (DLB) is the second most common form of dementia after Alzheimer’s disease (AD), accounting for 15–20 % of neuropathologically defined cases [1]. Diagnostic classification of DLB is based on revised consensus criteria, with core diagnostic features of DLB being (1) recurrent visual hallucinations, (2) cognitive fluctuations and (3) spontaneous motor features of parkinsonism [1]. The presence of two or three of these core signs is sufficient for a diagnosis of probable DLB [1] at the stage of dementia, and this overlaps with the recent nomenclature update in the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-V), where probable DLB is now comes under the heading of ‘major neurocognitive disorder with Lewy bodies’ [2]. In addition, DSM-V has the diagnostic classification of mild neurocognitive disorder with Lewy bodies. Thus, there is formal recognition that there is a pre-dementia state of this disease [2], similar to AD, where there is underlying neurodegenerative disease but function has not yet been compromised. This state has also been referred to as prodromal DLB (pro-DLB), a term which we use in this paper.

Distinguishing DLB from AD at the dementia stage is difficult because of overlapping clinical and neuropathological features between the two conditions, as well as because specific symptoms of DLB, such as hallucinations or fluctuations, are not spontaneously described by the patient and the caregiver. At early or prodromal stages, the challenges are even greater, given the subtlety of symptoms present. Whilst cognitive patterns in pro-DLB have been described as different from prodromal AD (pro-AD) [3, 4], with patients with DLB at this early stage having more visuospatial and fluency deficits than those with AD, and patients with AD having more amnesic impairments than those with DLB [4, 5], the neuropsychological pattern of pro-DLB appears to be more heterogeneous than that in pro-AD [5]. However, verbal memory [6] and visual memory impairments are frequent at the early stage of DLB, as in AD [7], and that is the reason why the differentiation of AD and DLB is difficult at the individual level. The behaviour modifications are also common in the two diseases: anxiety and depressive symptoms are frequent and do not permit differentiate of the two diseases [8, 9].

However, accurate differentiation of DLB and AD, regardless of the stage of the disease, is important clinically, given different management trajectories for each disease (e.g., avoidance of neuroleptics in DLB, but likely better response to cholinesterase inhibitors [10]) as well as prognosis [11]. Furthermore, accurate early subtype diagnosis of the underlying neurodegenerative cause is becoming increasingly important for ensuring that future disease-modifying treatments can be targeted in individuals before substantive neurodegenerative deficits have occurred. Similarly, identification and validation of early biomarkers of pro-DLB will further assist in the development of pro-DLB criteria [12]. Specifically, for purposes of our analyses, and in alignment with our previous work [13], we consider individuals with pro-DLB to be those patients who meet the revised diagnostic criteria for DLB, but, instead of dementia [1], fit the criteria for mild cognitive impairment (MCI) [14].

Structural neuroimaging represents one potential biomarker area, and a well-established method is the use of voxel-based morphometry (VBM) to study atrophic change in dementia. Comparison of VBM and a manual method for hippocampal volumetry to detect temporal lobe atrophy in AD showed that VBM with or without diffeomorphic anatomical registration through exponentiated lie algebra (DARTEL) registration is equivalent or more accurate [15, 16]. But VBM-DARTEL seems to be more accurate at detecting hippocampal atrophy in subtle cases, such as in depression [17]. This image-processing technique is also useful in many others neurological or psychiatric diseases including depression, schizophrenia, temporal lobe epilepsy, [18] but also MCI Parkinson’s disease [19].

Previous VBM studies in DLB at the stage of dementia have demonstrated diminished volume in the insulae bilaterally and in the lateral temporal lobes, frontal lobe and precuneus [20], and others have reported grey matter (GM) volumetric reductions in the mid-brain, the substantia innominata, the hypothalamus and the right insula [21], and these findings are confirmed by a voxel-wise meta-analysis on cortical atrophy of patients with DLB at the stage of dementia whose authors found bilateral insulae and basal ganglia atrophy [22]. Furthermore, this resonates with recent findings from our group. We demonstrated that the cortical thickness of patients with pro-DLB is diminished in the right anterior part of the insula [13]. This novel finding is significant, since patients at the early stage of the disease frequently have neurovegetative symptoms such as constipation, orthostatic dizziness or increased saliva, symptoms domains [23] which have been linked to insula. At the early stage, patients also frequently have cognitive fluctuations and hallucinations, and insular dysfunction has also been implicated in these symptoms [24, 25]. Contrarily, given that parkinsonism is rarely obvious at the beginning of the DLB, one might expect the putamen, pallidum and substantia nigra regions would be less structurally affected earlier in the disease [26]. However, data on subcortical structure were previously unexplored, since cortical in thickness studies researchers analysed only the cortex and not the hippocampi or basal ganglia [13].

Therefore, the primary aim of this study was to investigate cortical and subcortical grey matter (GM) atrophy patterns in patients with pro-DLB, and we report magnetic resonance imaging (MRI) patterns of GM atrophy in subjects with DLB at the stage of MCI (pro-DLB) and AD at the stage of MCI (pro-AD) compared with healthy elderly control subjects (HC). We hypothesised that, in pro-AD, the pattern of GM atrophy would involve predominantly the temporal lobe and parietal association cortices in keeping with the significant body of evidence supporting volumetric losses in these areas in pro-AD [27]. In contrast, we expected that the pattern of GM atrophy in pro-DLB would be less marked and more specific to structures such as the insulae.

We report distinct GM atrophy in pro-DLB and pro-AD compared with HC. Compared with HC the group, the pro-DLB group was characterised by cortical atrophy in insulae, structures and anterior cingulate cortex. Compared with HC, pro-AD was characterised by a widespread cortical atrophic pattern which included the hippocampi and temporal, parietal and frontal structures. The comparison of patients with pro-AD with patients with pro-DLB demonstrated more atrophy the right parietal lobe in the former than in the latter group.

The results we obtained with patients with pro-AD are in accord with the literature. The progression of atrophy with time in pro-AD was previously described as beginning in the medial temporal lobes [37], then extending to the parietal and finally the frontal lobes [27, 38, 39]. On one hand, the strong involvement of the frontal lobe is unusual in MCI-AD [38]. On the other hand, more impaired MCI [40] and MCI converters to AD dementia [41] have greater atrophy in the frontal lobe at the initial scan.

The atrophy found in patients with pro-DLB compared with HC has not been described previously, to our knowledge. Notably, we obtained similar results with the same patients on the basis VBM-DARTEL as we did with FreeSurfer methods [13], showing clear atrophic involvement of the anterior insulae in pro-DLB. Previously, we observed that cortical thinning was more apparent in the right insula, whereas in the present study, we observed GM insular atrophy bilaterally. This may reflect subtleties between MRI techniques when quantifying volumetric changes [42]. However, the discovery of the insula change in pro-DLB with two techniques in the same cohort reinforces the link between pro-DLB and early changes in the insula. This overlaps with voxel-wise meta-analytic data on cortical atrophy of patients with DLB at the stage of dementia, based on findings of bilateral insular and basal ganglia atrophy [22]. The insula is involved in integrating somatosensory, autonomic and cognitive-affective information to guide behaviour [43], and specifically the anterior insula has been described as part of a ‘salience network’ due to its consistent activation during cognitively demanding tasks. It also has been implicated in switching brain networks involved in cognition, including the central executive and default mode network [44]. The anterior insula has an abundance of specific neurons, namely the von Economo neurons (VENs), located in layer 5 of the cortex with a predominance in the right hemisphere, the same region we have found to be atrophic and thinner in pro-DLB [45]. Notably, in the right anterior cingulate cortex, an area which we found to be atrophic in this study, is also rich in VENs. Because of the larger size of VENs compared with pyramid neurons, they are purported be involved in the fast assessment of complex situations [45], as well as in the salience network function [46], and thus it might be hypothesised that deficits in this region might be pertinent to the cognitive slowing and attentional deficits which characterise DLB. This is supported by the observation that increasing functional dyssynchrony between frontal areas inclusive of the insula and parietal areas, as measured using resting state MRI, is associated with increasing cognitive fluctuations in patients with DLB [24]. Whether there is a specific vulnerability of VENs in DLB remains to be confirmed, however.

When we compared pro-DLB with pro-AD, we found subtle right parietal GM volume reductions in patients with pro-AD, and this is in line with our previous findings demonstrating cortical thinning of this region in AD compared with DLB [13]. Notably, we did not find evidence of any relative atrophy of the hippocampi in pro-AD compared with pro-DLB. However, this is consistent with our visual rating data which demonstrated that both groups had relatively mild hippocampal atrophy at the prodromal stage, with only subtle differences between the two diseases. Longitudinally, however, it might be expected that there would be an increasing divergence between pro-AD and pro-DLB cohorts, with greater hippocampal loss in the former; certainly, longitudinal comparisons in established dementia demonstrate that marked temporal and hippocampal thinning is a feature of AD but not DLB [47].

We found an association between GM loss in the left cuneus and visual hallucinations in our pro-DLB group. The cuneus is a secondary visual area (Brodmann area 18) and is of importance for recognition and extraction of object features (shape, colour, movement). Dysfunction of the cuneus is responsible for errors in visual processing [48]. The involvement of this visual area in hallucinations of patients with DLB is consistent with previous studies [48–50]. The involvement of the cuneus is congruent with the top-down/bottom-up models on hallucinations [51], since we have found a key region for visual processing (bottom-up).

Our study has some limitations. In the present study, we have used a combination of pre-existing criteria for MCI and DLB, but a primary issue is that operational consensus criteria for pro-DLB remain to be established. In addition, whilst patients continue to be followed longitudinally, our cohort is relatively recent in its inception, and the definite diagnostic trajectory of patients with prodromal impairment remains to be established either clinically or neuropathologically, particularly with regard to DLB. Moreover, the possibility that an incorrect classification of patients confounded the results cannot be excluded. However, we used the McKeith et al. criteria, which have an excellent specificity (>95 %) [52, 53] compared with gold standard neuropathological diagnosis. In addition, we excluded other pathologies, such as psychiatric illness, other neurological diseases and co-occurrence of AD and DLB (see Fig. 1 flowchart). Furthermore, the majority of our patients with pro-DLB had RBD (>50 %), which enhances the specificity of the diagnosis [54]. We also systematically looked for subtle clinical symptoms such as anosmia/hyposmia, constipation and other autonomic features (data not shown) [55], as these have previously been demonstrated to improve the diagnostic specificity of patients with pro-DLB [55]. Finally, we used CSF analysis for most of the patients and obtained similar results for patients with AD and patients with DLB.

From a technical perspective, it could also be argued that a drawback of our study is the fact that data were collected from two sites with differing imaging protocols. Unfortunately, we were not able to include MRI sequences in the model, as the relative numbers in each group were disproportionately represented (i.e., 30 HC at NCL, 3 at SXB; 1 AD at NCL, 26 at SXB; and 2 DLB at NCL, 26 at SXB). Therefore, it was difficult to obtain accurate parametric estimates without considerable error. However, centre effects were less likely to have influenced the between-patient contrast, as the results did not significantly differ when we examined only SXB patients in the analysis. A single-centre study should be conducted to confirm these results. Last, in a possible future analysis using resampling/subsampling techniques, the reliability of the prodromal results could be examined for validation purposes, although this would require larger cohorts.

Our data suggest that patients with prodromal disease have different patterns of atrophy depending on the pathology. Patients with pro-DLB seem to have more restricted areas of atrophy inclusive of the insulae and anterior cingulate cortex, and patients with pro-AD have more diffuse areas. New studies using MRI are acutely needed, particularly those considering perfusion, functional connectivity and structural connectivity, as well as other imaging modalities, such as positron emission tomography metabolic and amyloid radiotracers, to better understand the specificity of patients with pro-DLB.

Aβ_1–42, amyloid-β_1–42; AD, Alzheimer’s disease; ANOVA, analysis of variance; BA, Brodmann area; CAF, Clinician Assessment of Fluctuation; CDR, Clinical Dementia Rating; ChEI, cholinesterase inhibitor; CSF, cerebrospinal fluid; 3D, three-dimensional; DARTEL, diffeomorphic anatomical registration through exponentiated lie algebra; DLB, dementia with Lewy bodies; Dopa, levodopa or dopaminergic agonists; DSM-V, Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition; FWE, family-wise error; GM, grey matter; HC, healthy elderly control subjects; MCI, mild cognitive impairment; MMSE, Mini Mental State Examination; MNI, Montreal Neurological Institute; MRI, magnetic resonance imaging; NCL, old age psychiatry, geriatric medicine or neurology services from Newcastle upon Tyne; NL, neuroleptic; pro-AD, prodromal Alzheimer’s disease; pro-DLB, prodromal dementia with Lewy bodies; RBD, rapid eye movement sleep behaviour disorder; SXB, tertiary memory clinic of Strasbourg; TE, echo time; TIV, total intracranial volume; TMT, Trail Making Test; TR, repetition time; UPDRS, Unified Parkinson’s Disease Rating Scale; VBM, voxel-based morphometry; VEN, von Economo neuron; WM, white matter