Thalamic nuclei in frontotemporal dementia: Mediodorsal nucleus involvement is universal but pulvinar atrophy is unique to C9orf72

Abstract Thalamic atrophy is a common feature across all forms of FTD but little is known about specific nuclei involvement. We aimed to investigate in vivo atrophy of the thalamic nuclei across the FTD spectrum. A cohort of 402 FTD patients (age: mean(SD) 64.3(8.2) years; disease duration: 4.8(2.8) years) was compared with 104 age‐matched controls (age: 62.5(10.4) years), using an automated segmentation of T1‐weighted MRIs to extract volumes of 14 thalamic nuclei. Stratification was performed by clinical diagnosis (180 behavioural variant FTD (bvFTD), 85 semantic variant primary progressive aphasia (svPPA), 114 nonfluent variant PPA (nfvPPA), 15 PPA not otherwise specified (PPA‐NOS), and 8 with associated motor neurone disease (FTD‐MND), genetic diagnosis (27 MAPT, 28 C9orf72, 18 GRN), and pathological confirmation (37 tauopathy, 38 TDP‐43opathy, 4 FUSopathy). The mediodorsal nucleus (MD) was the only nucleus affected in all FTD subgroups (16–33% smaller than controls). The laterodorsal nucleus was also particularly affected in genetic cases (28–38%), TDP‐43 type A (47%), tau‐CBD (44%), and FTD‐MND (53%). The pulvinar was affected only in the C9orf72 group (16%). Both the lateral and medial geniculate nuclei were also affected in the genetic cases (10–20%), particularly the LGN in C9orf72 expansion carriers. Use of individual thalamic nuclei volumes provided higher accuracy in discriminating between FTD groups than the whole thalamic volume. The MD is the only structure affected across all FTD groups. Differential involvement of the thalamic nuclei among FTD forms is seen, with a unique pattern of atrophy in the pulvinar in C9orf72 expansion carriers.

Given the extensive heterogeneity across the FTD spectrum, it is likely that the nuclei of the thalamus are differentially involved in the various forms of FTD and that some of the symptoms are related to the function of the nuclei affected. However, to date, no study has methodically looked at the specific thalamic nuclei in the FTD spectrum, with prior investigations focused on volumetry of the whole thalamus.
Due to recent advances in segmentation methods, it is now possible to measure individual thalamic nuclei in vivo on structural magnetic resonance (MR) scans (Iglesias et al., 2018). We therefore aimed to investigate the specific patterns of atrophy in the thalamic nuclei in a large cohort of FTD patients, to determine which nuclei are impaired across the different clinical, genetic, and pathological forms of FTD.      Volumetric MRI scans were first bias field corrected and wholebrain parcellated using the geodesic information flow (GIF) algorithm (Cardoso et al., 2015), which is based on atlas propagation and label fusion. Volumes of the whole thalamus and of its nuclei were subsequently segmented using a customised version of the module We performed a stepwise discriminant analysis between pairs of genetic, pathological, and clinical FTD subgroups for the thalamic nuclei and a second discriminant analysis for the whole thalamus.

| RESULTS
Stratifying by genetics, the three groups showed significantly smaller thalamic nuclei than controls, except for VM for the GRN group and the pulvinar for both the GRN and MAPT groups. The pulvinar was only significantly smaller in C9orf72 than controls (16% difference in volume, p < .0005 ANOVA) (Table 3 and Figure 1A). The MD and LD nuclei were particularly affected in all groups (21-32% and 28-38%,  (Table 3 and Figure 1B). The LD was also particularly affected in TDP-43 type A (47%, p < .0005) and tau-CBD (44%, p < .0005). These two groups had the most nuclei affected followed by tau-Pick's disease, with sparing of VPL, VM, and the pulvinar in each of these groups.
Stratifying by clinical diagnosis, the MD was affected in all subgroups, with FTD-MND being the most affected (33%), and svPPA and PPA-NOS the least (17-18%, p < .0005). The LD was also affected in all subgroups, especially in FTD-MND (53%, p < .0005).
The pulvinar was spared in all groups. FTD-MND was the group with the smallest volumes overall, with PPA-NOS the least affected group (Table 3 and Figure 1C).
Comparisons between the disease groups for each of the three analyses are reported in the Table S1.
We also repeated the above analyses in a purely sporadic cohort, excluding the 21 genetic cases. The results for the clinical and pathological subgroups showed a similar pattern of nuclei involvement despite a reduction in the sample sizes and therefore the statistical power. The sporadic FTD-MND group still showed the highest volumetric differences from controls in the LP, MD, and midline nuclei (- Table S2), and the sporadic TDP-43 type A cases mainly showed volume differences in the MD, midline, LD, and LP (Table S3).
F I G U R E 1 Pattern of atrophy in the thalamic nuclei in the (a) genetic; (b) pathological; and (c) clinical FTD groups. The cartoon is a schematic representation of an axial view of the thalamic nuclei and is not anatomically accurate. Colour bar denotes the % difference in volume from controls T A B L E 4 Discriminant analysis between pairs of genetic, pathological and clinical FTD subgroups for the thalamic nuclei and the whole thalamus Notes: "p-Values" represents the Wills' Lambda significance for prediction accuracy, "% CC" represents the percentage of correctly classified subjects (where the first number refers to the reference group in the first column, and the second number to the group in the second column), while "Predictors" and "Correlation" represent respectively the nuclei included in the prediction models and their standardized canonical discriminant function coefficients.
The volumetric comparisons between the FTD groups are reported in the Table S3.
Results of the discriminant analysis are shown in Table 4. Overall, the accuracy to correctly classify the groups was higher when using one or a combination of thalamic nuclei, than the whole thalamus. Among the genetic groups, the best classification was between MAPT and C9orf72 using the pulvinar volumes, which correctly classified 70% of the MAPT carriers and 79% C9orf72 (p < .0005). Among the pathological groups, 100% correct classification was obtained between FUS and tau-PSP using VA, VLp, and intralaminar nuclei (p < .0005), between tau-Pick's and tau-PSP using LD, intralaminar, and LGN (p < .0005), between tau-PSP and TDP-43 type C using AV, VM, and intralaminar (p < .0005), and between TDP-43 type A and TDP-43 type B using LD, VLa, and midline nuclei (p = .001). Among clinical groups, classification accuracy was no better than using the whole thalamus (Table 4).

| CONCLUSIONS
In a large cohort of FTD patients we have shown that the MD is the only thalamic nucleus affected across all FTD groups. There is differential thalamic involvement among the FTD forms, with unique involvement of the pulvinar in C9orf72 expansions carriers. Involvement of thalamic nuclei was more in the genetic forms of FTD than the sporadic forms with only MD involvement (and no other nuclei) in TDP-43 type C and FUSopathies.
The MD is connected to several brain regions typically affected in FTD (Table 1), particularly prefrontal, temporal, and limbic areas that play a role in executive function as well as emotional and behavioural regulation. Prior small pathological studies have shown that the MD is specifically affected by chronic degenerative changes (e.g., neuronal depletion, gliosis, and astrocytosis) in FTD (in a single case with bvFTD (Radanovic et al., 2003) and in a group of ubiquitin-positive cases (Mackenzie et al., 2006)).
The LD nucleus is also significantly smaller across the genetic and clinical groups: this is another nucleus with a key role in the limbic system, and is strongly connected to regions commonly affected across the FTD spectrum (Schmahmann, 2003) (Table 1).
Our findings of unique pulvinar atrophy in C9orf72 expansion carriers are in line with the literature on imaging (Lee et al., 2014;Lee et al., 2016) and with pathological studies (Vatsavayai et al., 2016;Yang, Halliday, Hodges, & Tan, 2017) that have previously shown pulvinar involvement in C9orf72 carriers: one histological study showed that almost all the neurons in the pulvinar had inclusions containing TDP-43 and dipeptide repeat proteins (Vatsavayai et al., 2016). The pulvinar is a key region for limbic functions and intramodality integration of sensory information (Schmahmann, 2003) ( Table 1). Atrophy in this nucleus can lead to altered processing of pain, hallucinations, and both affective and psychotic symptoms. This is in line with frequently reported clinical symptoms in C9orf72 carriers that tend not to be found in other forms of FTD (Ducharme, Bajestan, Dickerson, & Voon, 2017;Fletcher et al., 2015). We also found that the LGN was particularly affected in C9orf72, an area previously linked to visual hallucinations which are a common feature of this genetic group (Ducharme et al., 2017;Kertesz et al., 2013).
We showed that by using the volume of a single thalamic nucleus or a combination of them, the accuracy in distinguishing between pairs of FTD subgroups was considerably higher than the accuracy in using the whole thalamus alone, especially for the genetic and pathological groups. This suggests that measuring individual thalamic nuclei rather than the whole thalamus may prove to be a better diagnostic biomarker for FTD, potentially as part of a wider set of volumetric measures.
Limitations of this study include the use of different scanners (three manufacturers, two different magnetic fields: 1.5 T and 3 T) with slightly different MRI sequence types. Even if we correct for scanner type and gender in the statistical model, we cannot completely remove some of the intrinsic heterogeneity due to these variables. However the algorithm operates at an internal resolution of 0.33 mm, which might compensate for the different native resolutions of the scans. Moreover, due to its unsupervised model for image intensities and to its Bayesian nature, this segmentation method is agnostic to the contrast of the MR images, and it is thus robust to the contrast changes between the scans acquired on different scanners.
Whilst we used an automated method to extract the thalamic nuclei volumes, which is not as accurate as manual segmentation on dedicated MRIs or on brain tissue post-mortem, we combined the 52 regions in the initial segmentation into 14 nuclei in order to decrease the effect of a less reliable delineation on T1-weighted MRI.
Furthermore, manual segmentation is extremely time-consuming and labour-intensive in such a large cohort. As there is no reliable measure of disease severity for FTD, and there is heterogeneity across its forms in the rate of disease progression, another limitation is the difficulty in characterising the level of disease severity between groups.
Further studies with longitudinal data and both diffusion-weighted and functional MRI are needed to understand the differential involvement of thalamic nuclei over the course of the disease, and the changes in thalamic connectivity to other regions of the brain. Particularly important will be the investigation of presymptomatic mutation carriers in whom the earliest disease changes can be seen. However, this study has already highlighted both common and unique features of thalamic nuclei involvement across the FTD spectrum, adding to our understanding of the heterogeneity of this neurodegenerative illness.

CONFLICT OF INTEREST
J.D.R. has been on a Medical Advisory Board for Wave Life Sciences and Ionis Pharmaceuticals.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are not publicly available due to ethical restrictions.