NeuroImage: Clinical

Background: Frontotemporal dementia (FTD) is a heterogeneous neurodegenerative disorder associated with frontal and temporal atrophy. Subcortical involvement has been described as well, with early thalamic atrophy most commonly associated with the C9orf72 expansion. However thalamic involvement has not been comprehensively investigated across the FTD spectrum. Methods: We investigated thalamic volumes in a sample of 341 FTD patients (age: mean(standard deviation) 64.2(8.5) years; disease duration: 4.6(2.7) years) compared with 99 age-matched controls (age: 61.9(11.4) years). We performed a parcellation of T1 MRIs using an atlas propagation and label fusion approach to extract left and right thalamus volumes, which were corrected for total intracranial volumes. We assessed subgroups strati ﬁ ed by clinical diagnosis (141 behavioural variant FTD (bvFTD), 76 semantic dementia (SD), 103 progressive non ﬂ uent aphasia (PNFA), 7 with associated motor neurone disease (FTD-MND) and 14 primary progressive aphasia not otherwise speci ﬁ ed (PPA-NOS), genetic diagnosis (24 with MAPT , 24 with C9orf72 , and 15 with GRN mutations), and pathological diagnosis (40 tauopathy, 61 TDP-43opathy, 3 FUSopathy). We assessed the diagnostic accuracy based on thalamic volume. Results: Overall, FTD patients had smaller thalami than controls (8% di ﬀ erence in volume, p < 0.0005, ANCOVA). Stratifying by genetics, C9orf72 group had the smallest thalami (14% di ﬀ erence from controls, p < 0.0005). However, the thalami were also smaller than controls in the other genetic groups: GRN and MAPT groups showed a di ﬀ erence of 11% and 9% respectively (p < 0.0005). ROC analysis showed a relatively poor ability to separate C9orf72 from MAPT (AUC=0.651, p=0.073) and from GRN cases (AUC=0.644, p=0.133) using thalamic volume. All clinical subtypes had signi ﬁ cantly smaller thalami than controls (p < 0.0005), with the FTD-MND group having the smallest (15%), followed by bvFTD (9%), PNFA (8%), PPA-NOS (7%), and lastly SD (5%). In the pathological groups, the TDP-43opathies had an 11% di ﬀ erence from controls, and tauopathies 9%, while the FUSopathies showed only 2% of di ﬀ erence from controls (p < 0.0005). GRN , PPA-NOS and SD were the subgroups showing the highest asymmetry in volumes. Conclusions: The thalamus was most a ﬀ ected in C9orf72 genetically, TDP-43opathies pathologically and FTD-MND clinically. However, thalamic atrophy is a common feature across all FTD groups.


Introduction
Frontotemporal dementia (FTD) is a clinically, pathologically and genetically heterogeneous neurodegenerative disorder, associated with frontal and temporal atrophy. Subcortical involvement has been found in a number of studies (Rohrer et al., 2010a;Whitwell et al., 2012;Seeley et al., 2008;Schroeter et al., 2007), with thalamic atrophy most commonly described in association with the C9orf72 expansion (Whitwell et al., 2012;Sha et al., 2012;Lee et al., 2014;Mahoney et al., 2012), even at the presymptomatic stage Lee et al., unclear whether and to what extent the thalamus is impaired in the other genetic forms of FTD, or across the different clinical and pathological diagnoses.
The thalamus is an important hub within many networks in the brain as it integrates somatosensory, motor, visual and auditory information through reciprocal connections with the cortex. The thalamus is composed of 50-60 different subnuclei and each nucleus has a distinct pattern of cortical and subcortical connectivity (Herrero et al., 2002). The dorsomedial and anteroventral nuclei are part of the dorsolateral prefrontal circuit, related to executive functions and motor programming, and also part of the lateral orbitofrontal circuit, related to personality and mood regulation. Being such a relevant brain structure interconnected to virtually all brain regions, the thalamus is likely to be a key structure involved in FTD. We therefore aimed to investigate thalamic involvement in a large cohort of patients across the whole FTD spectrum, including those with genetic and pathological confirmation.

Methods
We reviewed the UCL Dementia Research Centre FTD MRI database to identify 341 patients with a usable T1-weighted magnetic resonance (MR) scan and with a diagnosis of behavioural variant FTD (bvFTD) , semantic dementia (SD), progressive nonfluent aphasia (PNFA) , FTD with associated motor neurone disease (FTD-MND), or a primary progressive aphasia not otherwise specified (PPA-NOS) (Harris et al., 2013). 99 cognitively normal subjects, with a similar age to the patients and with a usable T1weighted MRI, were identified as controls. The study was approved by the local ethics committee and written informed consent was obtained from all participants.
MRIs were acquired from 1992 to 2017 with three different manufacturer scanners: 216 on a 1.5T Signa MRI scanner (GE Medical systems, Milwaukee, WI), 188 on a 3T Trio MRI scanner (Siemens, Erlangen, Germany), and 36 on a 3T Prisma MRI scanner (Siemens, Erlangen, Germany). When more than one MRI per participant was available, we selected the MRI closest to symptom onset. We reviewed the MRIs to make sure we excluded individuals with moderate to severe vascular disease or other brain lesions such as tumours.
For 54 patients, post-mortem confirmation of the underlying neuropathology was available: pathological examination of brain tissue was carried out according to standard histopathological methods at the Queen Square Brain Bank for Neurological Disorders, UCL Institute of Neurology. 67 patients were carriers of a mutation in one of the FTDlinked genes: microtubule-associated protein tau (MAPT) (Hutton et al., 1998;Ghetti et al., 2015), progranulin (GRN) (Baker et al., 2006;Cruts et al., 2006), chromosome 9 open reading frame 72 (C9orf72) (DeJesus-Hernandez et al., 2011;Renton et al., 2011), TANK-binding kinase 1 (TBK1) (Freischmidt et al., 2015;Gijselinck et al., 2015;Le Ber et al., 2015;Pottier et al., 2015), and sequestosome 1 (SQSTM1) (Rubino et al., 2012;Le Ber et al., 2013;Miller et al., 2015). We divided the patient group based on their clinical diagnosis (141 bvFTD, 76 SD, 103 PNFA, 7 FTD-MND, 14 PPA-NOS), their genetic diagnosis (24 MAPT, 24 C9orf72, 15 GRN), and their pathological diagnosis (40 tauopathy, 61 TDP-43opathy, 3 FUSopathy). Within the tauopathy group, we included individuals who had tau pathology at post-mortem or were MAPT mutation carriers, while within the TDP-43opathy group, we included individuals who had definite TDP-43 pathology or were carriers of mutations in GRN, C9orf72, TBK1 (n = 2), and dual mutations in GRN/ C9orf72 (n = 1) or C9orf72/SQSTM1 (n = 1) (   Thalamic volumes were extracted as part of the parcellation on T1weighted volumetric MRI scans as previously described , using an atlas propagation and label fusion strategy . Volumes are expressed as a percentage of total intracranial volume (TIV), computed with SPM12 v6470 (Statistical Parametric Mapping, Wellcome Trust Centre for Neuroimaging, London, UK) running under Matlab R2014b (Math Works, Natick, MA, USA) (Malone et al., 2015). All segmentations were visually checked for quality. Statistical analyses were performed in SPSS software (SPSS Inc., Chicago, IL, USA) v22.0, between control and patient groups, using the ANCOVA test adjusting for scanner type, gender and age. When comparing between different patient subgroups, we also adjusted for disease duration. Results were corrected for multiple comparisons (Bonferroni's correction), p < 0.008 for the clinical groups, p < 0.013 for the genetic and pathological groups. To assess the accuracy of the thalamic volume in discriminating between different diagnoses, we performed a Receiver Operating Characteristic (ROC) analysis. We also investigated asymmetry by calculating an Asymmetry Index (AI), defined as the absolute difference between the left and right thalamic volumes in relation to the total bilateral volume: |(Left − Right)|/ (Left + Right).

Results
Sociodemographic and clinical data are reported in Table 1. The mean disease duration for the whole FTD group at the time of the scan was 4.6 years (standard deviation 2.7) with an average age at onset at 59.6 (8.6). There was no significant difference in age between FTD and controls (p = 0.067, t-test), or for scanner type (p = 0.785, Chi square test), but there were more males in the FTD group than in the control group (60% vs 42%, p = 0.002, Chi square test). Across the different clinical, genetic and pathological diagnoses, there was no difference for scanner type (p = 0.266, p = 0.508 and p = 0.390, Chi square test). There was a significant difference in disease duration across the genetic groups (p = 0.016, ANOVA), with GRN carriers showing the shortest duration, but no difference across the pathological nor clinical groups (p = 0.433 and p = 0.075).
Investigating the control group, there was a weak but significant negative correlation of thalamic volume with age (Spearman's rho: −0.444, p-value < 0.0005).
Stratifying by genetics, C9orf72 group had the smallest thalami (left: 15 and right: 13% difference from controls, p < 0.0005). However, the thalami were also smaller than controls in the other genetic groups: GRN (13 and 9%, p < 0.0005) and MAPT (12 and 7%, p < 0.0005) groups. There were no significant differences between the disease groups. Note that excluding the genetic cases and analyzing the sporadic cases alone (n = 274) also showed a similar pattern of smaller thalami than controls (4% difference on the left, 9% on the right, and 7% in total, p < 0.0005 ANCOVA -Supplementary Table 2 for more detailed analysis).
Among the clinical groups, the ROC analysis showed the highest diagnostic accuracy between FTD-MND and SD (right thalamus, AUC = 0.921, p < 0.0005) ( Table 2). In the genetic groups the ROC analysis showed a poor ability to separate C9orf72 from MAPT, with the highest AUC value for the right thalamus (AUC = 0.679, p = 0.034) and from GRN cases (AUC = 0.644, p = 0.133 for the sum of right and left) ( Table 2). For the pathological groups, the highest accuracy was when differentiating between TDP-43 and tau for the left thalamus volume (AUC = 0.869, p-value = 0.032).
When investigating the asymmetry of the thalamus, the FTD group as a whole was significantly more asymmetric than controls (0.032 (0.024) versus 0.015 (0.012), < 0.00005, ANCOVA). However, consistent with previous findings, the controls had a non-zero asymmetry index. PPA-NOS and SD were the most asymmetric clinical groups, and GRN the most asymmetric among the genetic groups ( Fig. 2 and Table 3). FUS, FTD-MND and C9orf72 were the only groups not showing significant asymmetry.

Discussion
Using an automated and robust segmentation method to segment the thalamus in a large cohort of FTD patients, we demonstrated that thalamic volumes were lower than in controls in all clinical, genetic, and pathological FTD groups except those with FUSopathies, and that FTD-MND, C9orf72, and TDP-43 were the subgroups for which the thalamus was particularly affected. Our results support the existing literature on C9orf72 (Whitwell et al., 2012;Sha et al., 2012;Lee et al., 2014;Mahoney et al., 2012), and on TDP-43 pathology (Rohrer et al., 2010b). However, these results show that thalamic atrophy is not just a characteristic of C9orf72 (Yang et al., 2017), but of the whole FTD spectrum, and it is not possible to accurately discriminate among different forms of FTD based on thalamic volumes.
The overall difference in thalamic volumes compared with controls was smaller than in the two neuropathological studies of thalamus atrophy that have been performed [28-37% in Mann and South, 1993;46-49% in C9orf72 carriers and~25% in sporadic FTD due to TDP-43 pathology in Yang et al., 2017], but our measurements were done in vivo on MRI and as close as possible to the diagnosis. They are likely to be lower therefore than any pathological studies where the disease will be at a more severe stage.
The most significant clinical, genetic and pathological groups overlap in our study, as FTD-MND is usually a TDP-43opathy and is commonly associated with a C9orf72 mutation. Four out of 7 of the FTD-MND cases here had either a single C9orf72 mutation (2 cases) or a dual mutation of C9orf72 with another gene (2 cases). However it is clear that lower thalamic volumes are not just driven by the C9orf72 status with overlap of values with other mutations and pathologies.
Previous studies have shown involvement of the thalamus in patients with ALS without FTD (e.g. in C9orf72-ALS: Bede et al., 2013) and it will be useful to study the FTD/MND continuum further in larger cohorts, given the relatively limited size of the FTD-MND group in this study.
Of all the groups, the FUSopathies seems to have the most intact thalamus, consistent with previous imaging studies . However, this is a rare pathological cause of FTD and there were only 3 patients in this grouplarger studies will be required to investigate this further.
Clinically, while the combined right and left thalamic volume was highest for PPA-NOS and SD, this was driven by the asymmetrical nature of the disease, with much lower volumes of the left compared with the right thalamus (and the highest asymmetry index). The GRN mutation group also showed asymmetry consistent with previous studies showing that this is generally a very asymmetrical disease (Rohrer et al., 2010a).
The study included a large control group of 99 healthy individuals. Our study supported two key findings in the literature of firstly, a correlation between age and thalamic volume [previous studies showing an R 2 ranging from 0.31 to 0.60: Sullivan et al., 2004;Hughes et al., 2012], and evidence for asymmetry of thalamic volume [e.g. in a study of over 15,800 people by Guadalupe et al., 2016, AI was 0.021, comparable with the finding here of 0.015].
The thalamus is a key hub in several brain networks and it is therefore not surprising that it is involved in all the different forms of FTD. We only investigated the whole thalamus volume here, but each thalamic nucleus has a distinct pattern of cortical and subcortical connectivity. Future studies, including functional and structural connectivity MR analyses of the thalamus, will be needed to investigate the different subnuclei and their connections in order to better understand the role of this key structure in each of the different forms of FTD. Furthermore, improved understanding of the role of the thalamus in different cognitive and behavioural functions may allow better stratification clinically of the FTD syndromes (e.g. those that have abnormal pain and temperature processing (Fletcher et al., 2015)) and therefore clearer correlation with specific subnuclei. Lastly, the findings of this study suggest that thalamic atrophy may be a useful volumetric imaging biomarker in future disease modifying therapy trials as it is easily measured and universal to all FTD subtypes.