Clinicopathological correlations in behavioural variant frontotemporal dementia

Abstract

Accurately predicting the underlying neuropathological diagnosis in patients with behavioural variant frontotemporal dementia (bvFTD) poses a daunting challenge for clinicians but will be critical for the success of disease-modifying therapies. We sought to improve pathological prediction by exploring clinicopathological correlations in a large bvFTD cohort. Among 438 patients in whom bvFTD was either the top or an alternative possible clinical diagnosis, 117 had available autopsy data, including 98 with a primary pathological diagnosis of frontotemporal lobar degeneration (FTLD), 15 with Alzheimer's disease, and four with amyotrophic lateral sclerosis who lacked neurodegenerative disease-related pathology outside of the motor system. Patients with FTLD were distributed between FTLD-tau (34 patients: 10 corticobasal degeneration, nine progressive supranuclear palsy, eight Pick's disease, three frontotemporal dementia with parkinsonism associated with chromosome 17, three unclassifiable tauopathy, and one argyrophilic grain disease); FTLD-TDP (55 patients: nine type A including one with motor neuron disease, 27 type B including 21 with motor neuron disease, eight type C with right temporal lobe presentations, and 11 unclassifiable including eight with motor neuron disease), FTLD-FUS (eight patients), and one patient with FTLD-ubiquitin proteasome system positive inclusions (FTLD-UPS) that stained negatively for tau, TDP-43, and FUS. Alzheimer's disease was uncommon (6%) among patients whose only top diagnosis during follow-up was bvFTD. Seventy-nine per cent of FTLD-tau, 86% of FTLD-TDP, and 88% of FTLD-FUS met at least 'possible' bvFTD diagnostic criteria at first presentation. The frequency of the six core bvFTD diagnostic features was similar in FTLD-tau and FTLD-TDP, suggesting that these features alone cannot be used to separate patients by major molecular class. Voxel-based morphometry revealed that nearly all pathological subgroups and even individual patients share atrophy in anterior cingulate, frontoinsula, striatum, and amygdala, indicating that degeneration of these regions is intimately linked to the behavioural syndrome produced by these diverse aetiologies. In addition to these unifying features, symptom profiles also differed among pathological subtypes, suggesting distinct anatomical vulnerabilities and informing a clinician's prediction of pathological diagnosis. Data-driven classification into one of the 10 most common pathological diagnoses was most accurate (up to 60.2%) when using a combination of known predictive factors (genetic mutations, motor features, or striking atrophy patterns) and the results of a discriminant function analysis that incorporated clinical, neuroimaging, and neuropsychological data.

Keywords: Alzheimer’s disease; Pick’s disease; corticobasal degeneration; frontotemporal dementia; frontotemporal lobar degeneration.

Figure 1

Patient ascertainment by bvFTD diagnostic certainty. Dx = diagnosis.

Figure 2

Pathological diagnoses for all patients and grouped by clinician bvFTD diagnostic certainty. AD = Alzheimer’s disease.

Figure 3

Grey matter atrophy maps and frequency of behavioural features by pathological diagnosis. Imaging shows maps of grey matter atrophy at a threshold of W > 1.5 (red) and W > 3 (yellow). Radial plots display the frequency (0–100%) of meeting each of the six FTDC diagnostic criteria at first presentation and mean NPI subscale scores (0–12). Right side of coronal and axial images corresponds to the right side of the brain. AD = Alzheimer’s disease; Ag = agitation; Anx = anxiety; Ap = apathy; Com = compulsions; Del = delusions; Dep = depression; Dis = disinhibition; Eat = eating behaviour; Emp = loss of empathy; Eup = euphoria; Hall = hallucinations; Irr = irritability; Mot = aberrant motor behaviour; NP = neuropsychological profile; Slp = sleep.

Figure 4

Overlap in grey matter atrophy. Imaging shows the number of patients or pathological subtypes with atrophy (W > 1.5) at each voxel. Top: Overlap in the 10 top pathological subtypes. Middle: Overlap in individual subjects (n = 82). Bottom: Overlap in patients with high confidence bvFTD (n = 42). Right side of coronal images corresponds to the right side of the brain. Right: Mean W for each diagnosis in regions of greatest overlap (≥8 diagnoses for top right, ≥65 patients for middle right, and ≥39 patients for bottom right). AD = Alzheimer’s disease; PiD = Pick’s disease.

Figure 5

PCA of grey matter W-score maps. The top 25% of voxels contributing to the first 10 components from the principal component analysis. With each is a colour bar representing the mean score for each component of all included subjects with each of the top 10 diagnoses. n = 82. AD = Alzheimer’s disease; PiD = Pick’s disease.

Figure 6

A priori algorithm for bvFTD pathological prediction. The algorithm is shown on the left, with branches leading to a list of likely pathological diagnoses. On the right are the results from applying the algorithm to the bvFTD cohort, including the numbers of patients whose diagnoses were consistent or inconsistent with the algorithm’s prediction. *Sixteen patients could not be fully classified by the algorithm because of lack of imaging. AD = Alzheimer’s disease.