We reviewed the UCL Dementia Research Centre FTD MRI database to identify patients with a diagnosis of svPPA [7] and a usable 3 T T1-weighted magnetic resonance (MR) scan. Twenty-four patients were identified, all with left-temporal predominant disease. Seventy-two cognitively normal subjects with a usable volumetric 3 T T1-weighted MRI were identified as controls. The study was approved by the local ethics committee, and written informed consent was obtained from all participants. The study was conducted in accordance with the Helsinki Declaration of 1975.
Based on their scores on a test of semantic knowledge (the British Picture Vocabulary Scale, BPVS, a word-picture matching task) [8], we split the svPPA patients into three equal subgroups (n = 8 per group) of early (BPVS > 110/150), middle (BPVS = 55–110/150) and late disease stage (BPVS < 55/150). Patients were negative for mutations in all FTD-related genes. Two patients received post-mortem confirmation of the underlying neuropathology, both TDP-43 type C.
All patients underwent a detailed neuropsychological examination including tests of fluid intelligence (WASI Matrices), single-word comprehension (WASI Vocabulary), naming (Graded Naming Test), reading (National Adult Reading Test), verbal memory (Recognition Memory Test for Words), visual memory (Recognition Memory Test for Faces), short-term memory (forwards digit span), working memory (backwards digit span), calculation (Graded Difficulty Calculation Test), visuoperceptual function (Visual Object and Space Perception battery Object Decision subtest) and executive function (inhibition—D-KEFS Color-Word Ink Naming Test; abstract reasoning—WASI Similarities). A percentile score based on standard norms was generated for each patient, with a mean percentile score created for the early, middle and late stage groups. Assessment of behavioural symptoms was performed using the revised version of the Cambridge Behavioural Inventory (CBI-R) [9]: six subscores were used (difficulties with self-care, abnormal sleep, hallucinations/delusions, disinhibition, abnormal eating behaviour, obsessive-compulsive behaviour, apathy and loss of empathy) with a percentage of the total possible subscore generated for every patient; for each stage, a mean percentage score was created. We report the cognitive and behavioural profiles at each stage for illustrative purposes (Fig. 1 and Additional file 1: Table S1).
T1-weighted MRIs were acquired using a 3-T scanner, either a Trio (Siemens, Erlangen, Germany, TR = 2200 ms, TI = 900 ms, TE = 2.9 ms, acquisition matrix = 256 × 256, spatial resolution = 1.1 mm) or a Prisma (Siemens, Erlangen, Germany, TR = 2000 ms, TI = 850 ms, TE = 2.93 ms, acquisition matrix = 256 × 256, spatial resolution = 1.1 mm). Individuals with moderate to severe vascular disease or space-occupying lesions were excluded.
Volumetric MRI scans were first bias field corrected and whole-brain parcellated using the geodesic information flow (GIF) algorithm [10], which is based on atlas propagation and label fusion. The hippocampal subfields and amygdalar subregions were subsequently segmented using a customized version of the module available in FreeSurfer 6.0 [11, 12], to adapt the output of GIF to the FreeSurfer format. For the hippocampal subfields, we focused on seven areas: CA1, CA2/CA3, CA4, dentate gyrus, subiculum, presubiculum and the tail. We excluded from the analysis the hippocampus-amygdala transition area, the parasubiculum, the molecular layer of the hippocampus, the fimbria and the hippocampal fissure, as they were too small, or not reliably delineated on T1-weighted images. For the amygdalar subnuclei, we focused the analysis on five regions, by combining the smallest subnuclei, based on an anatomical subdivision [13]: lateral nucleus, basal and paralaminar nucleus, accessory basal nucleus, cortico-amygdaloid transition area and the superficial nuclei (central nucleus, cortical nucleus, medial nucleus, anterior amygdaloid area).
For comparison with the medial temporal subregions, we extracted volumes of the following cortical regions from GIF: temporal (medial, lateral, supratemporal, temporal pole), frontal (orbitofrontal, prefrontal), parietal, occipital, insular and cingulate (anterior and posterior). We also extracted volumes of subcortical structures for the pallidum, putamen, caudate, nucleus accumbens and thalamus.
Left and right volumes were corrected for 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) [14]. All segmentations were visually checked for quality.
Statistical analyses were performed on brain volumes (as a percentage of TIV) in STATA v14 (Stata-Corp, College Station, TX), between control and patients (early, middle and late stage groups), using a linear regression test adjusting for scanner type, TIV, gender and age. The results were corrected for multiple comparisons (Bonferroni correction): p < 0.006 for amygdalar subnuclei and subcortical structures, p < 0.005 for hippocampal subfields and p < 0.0035 for cortical regions.