Distinct [18F]THK5351 binding patterns in primary progressive aphasia variants

Purpose To assess the binding of the PET tracer [18F]THK5351 in patients with different primary progressive aphasia (PPA) variants and its correlation with clinical deficits. The majority of patients with nonfluent variant (NFV) and logopenic variant (LV) PPA have underlying tauopathy of the frontotemporal lobar or Alzheimer disease type, respectively, while patients with the semantic variant (SV) have predominantly transactive response DNA binding protein 43-kDa pathology. Methods The study included 20 PPA patients consecutively recruited through a memory clinic (12 NFV, 5 SV, 3 LV), and 20 healthy controls. All participants received an extensive neurolinguistic assessment, magnetic resonance imaging and amyloid biomarker tests. [18F]THK5351 binding patterns were assessed on standardized uptake value ratio (SUVR) images with the cerebellar grey matter as the reference using statistical parametric mapping. Whole-brain voxel-wise regression analysis was performed to evaluate the association between [18F]THK5351 SUVR images and neurolinguistic scores. Analyses were performed with and without partial volume correction. Results Patients with NFV showed increased binding in the supplementary motor area, left premotor cortex, thalamus, basal ganglia and midbrain compared with controls and patients with SV. Patients with SV had increased binding in the temporal lobes bilaterally and in the right ventromedial frontal cortex compared with controls and patients with NFV. The whole-brain voxel-wise regression analysis revealed a correlation between agrammatism and motor speech impairment, and [18F]THK5351 binding in the left supplementary motor area and left postcentral gyrus. Analysis of [18F]THK5351 scans without partial volume correction revealed similar results. Conclusion [18F]THK5351 imaging shows a topography closely matching the anatomical distribution of predicted underlying pathology characteristic of NFV and SV PPA. [18F]THK5351 binding correlates with the severity of clinical impairment. Electronic supplementary material The online version of this article (10.1007/s00259-018-4075-3) contains supplementary material, which is available to authorized users.

Although there is some concordance between each pathology and its clinical presentation, the clinical diagnosis does not provide a reliable indication of the underlying pathology in the individual PPA patient [3]. For instance, although the likelihood of a primary FTLD tauopathy is higher in patients with NFV PPA, AD pathology may also be a cause [3,4]. Moreover, in many patients, a tauopathy cannot be reliably distinguished from a TDP-43 proteinopathy as the underlying cause based on clinical grounds [3,4]. The ability to discriminate between these two distinct causes in vivo is essential for the development of disease-modifying therapies. Furthermore, molecular targets, which can serve as noninvasive markers of drug target engagement and disease progression, are needed for clinical drug development.
[ 18 F]AV1451 binds robustly to AD-related tauopathy, while binding to 4R tau is inconclusive [10][11][12][13][14]. By contrast, [ 18 F]THK5351 shows more prominent binding in FTLD (CBD and PSP) than in AD tauopathy [9,[15][16][17][18][19][20]. In PPA patients, [ 18 F]AV1451 has been demonstrated to be able to discriminate among the three subtypes [21]. A principal components analysis has shown that two components allow reliable discrimination: the degree of increase and the frontotemporal gradient. Standardized uptake value ratios (SUVR) were found to be higher in patients with LV PPA than in those with the two other variants, suggesting that this ligand probably has higher affinity for AD tauopathy than for FTLD tauopathy. In patients with SV PPA, binding was mainly situated in the anterior temporal cortex while in patients with NFV PPA, [ 18 F]AV1451 binding was principally seen in the frontal white matter and subcortical nuclei [21]. The discriminative ability of [ 18 F]AV1451 was as high as that of [ 18 F]FDG PET and higher than that of MRI [21].
There have also been a number of smaller case studies investigating the use of PET for determining tau burden in patients with PPA, some of them including neuropathological confirmation [12]. In a patient with NFV PPA and in a patient with PSP with nonfluent aphasia, elevated [ 18 F]THK5351 binding was seen in the frontal cortex [18,19]. In a patient with SV PPA, binding of both [ 18 F]AV1451 [21][22][23] and [ 18 F]THK5351 [19,24,25] in the anteroinferior and lateral temporal cortices has been reported. This is surprising as SV is mainly a TPD-43 proteinopathy. Elevated binding of [ 18 F]AV1451 [19,21,[26][27][28][29] and [ 18 F]THK5351 has been demonstrated in the posterior temporal cortex and inferior parietal lobule bilaterally in single patients with LV PPA [17,19], which mirrors the [ 18 F]FDG binding pattern [17].
The primary objective of this study was to evaluate the ability of [ 18 [26,27]. As yet such a correlation has not been demonstrated for speech and language measures in patients with PPA.

Study participants
A consecutive series of 21 patients who fulfilled the international consensus criteria for PPA [1] were enrolled between August 2016 and October 2017. Of these 21 patients, 18 were recruited through the memory clinic of the University Hospitals Leuven, one (patient 3) was referred to the study from the Free University Amsterdam and two (patients 17 and 20) from the University Hospitals Ghent (Table 1). One patient (patient 15) had to be excluded due to a subarachnoid cyst anterior to the left temporal lobe. The patients were classified on the basis of the clinical evaluation by an experienced neurologist in combination with the results of clinical MRI and [ 18 F]FDG PET scans. In a subset of 13 PPA patients AD biomarkers were measured in the cerebrospinal fluid (CSF) as part of the clinical work-up. This was performed by the Laboratory of Medical Analysis, Medicine Department of UZ Leuven, using Innotest ELISA for amyloid-β 42 (Aβ 1-42 ; cut-off 853 pg/ml [32]), total tau (t-tau; Aβ 1-42 /t-tau cut-off 2.258), and phospho 181 -tau (p 181 -tau; Fujirebio Europe, Ghent, Belgium) ( Table 1). Two patients (patients 9 and 14) received 11 C-labelled Pittsburgh compound B ([ 11 C]PIB) PET as part of the prior clinical work-up [33] (Table 1). In PPA patients in whom amyloid biomarkers had not been measured, [ 11 C]PIB PET was acquired for the current study. Of the 20 PPA patients included, 12 fulfilled the consensus criteria for NFV, 5 for SV, and 3 for LV PPA. Clinical signs and symptoms associated with PSP or CBD were documented on neurological/clinical examination (Supplementary Table 1).
For normative reasons, we recruited 23 cognitively intact older healthy controls matched for age (Kruskall Wallis H(3) = 2.38, P = 0.50), education (H(3) = 1.20, P = 0.75) and gender (χ 2 (3) = 1.97, P = 0.58) with the group of PPA patients through advertisements in newspapers and online. Inclusion criteria were a Mini-Mental State Examination (MMSE) score of ≥27, a Clinical Dementia Rating scale (CDR) global score of zero and neuropsychological test scores within 1.9 of the standard deviations of norms adjusted for age, gender and education [33]. Healthy controls had no history of neurological or psychiatric disease or any brain lesions on structural MRI. White matter lesions were not an exclusion criterion.
The study protocol (EudraCT 2014-002976-10) was approved by the UZ/KU Leuven Ethics Committee for Research. All participants provided written informed consent in accordance with the principles of the Declaration of Helsinki after receiving a complete description of the study protocol.

Neuropsychological and neurolinguistic protocol
All study participants underwent a standard neuropsychological and neurolinguistic examination. General cognitive functioning was assessed using the CDR and MMSE. Nonverbal fluid intelligence was assessed using Coloured Progressive Matrices (CPM). Confrontation naming was assessed using the Boston Naming Test (BNT) and category verbal fluency using the 1-min Animal Verbal Fluency (AVF) test. Singleword comprehension (spoken and written input modality) was assessed using the Dutch versions of the Aachen Aphasia Test (Akense Afasie Test, AAT) and associative-semantic ability was assessed using the Psycholinguistic Assessment of Language Processing in Aphasia (PALPA; subtest 45). PALPA subtest 49 and the picture version of the Pyramids and Palm Trees Test (PPT). Object identification was assessed using the Birmingham Object Recognition Battery (BORB), easy (B) and hard (A). Sentence comprehension and grammaticality were assessed using the Werkwoorden En Zinnen Test (WEZT) [34] (verbs and sentences test). Repetition was assessed using the AAT repetition test and real words and pseudowords using the PALPA subtest 9. Speech apraxia was assessed using the Diagnostisch Instrument voor Apraxie van de Spraak (DIAS) [35] (Diagnostic Instrument for Apraxia of Speech). The DIAS allows the examiner to assess consonant and vowel repetition (DIAS severity score). The DIAS also includes a diadochokinesis task, which assesses the ability to make antagonistic movements using different parts of the mouth, tongue and soft palate in quick succession.
The scores of individual PPA patients were compared against those of a larger group of 67 cognitively intact older controls, including 23 control subjects who participated in the PET study, to increase the statistical power for comparisons with PPA patients. These 67 controls were matched for age    9-192.3 MBq, in patients). A low-dose CT scan was acquired for attenuation correction prior to the PET scan. PET emission images were acquired in 3D list mode and subsequently reconstructed as six 5-min frames using the ordered subsets expectation maximization algorithm (four iterations, 16 subsets) [32]. [ 18 F]THK5351 PET emission frames were realigned to correct for head motion, summed and rigidly coregistered to the subject's T 1 -weighted MRI scan using statistical parametric mapping software (SPM12; Wellcome Trust Centre for Neuroimaging, London, UK) implemented in Matlab R2014b Table 1 Demographics, neurolinguistic and neuropsychological assessment of the 20 included PPA patients   Values in bold are significantly different from those in healthy controls based on Crawford and Garthwaite [30] or Crawford and Howell [31] statistical procedures AAT Akense Afasie Test (Aachen Aphasia Test), Aβ (Mathworks, Natick, MA). The MRI scan was segmented using SPM12 into grey matter, white matter and CSF. The summed PET images were subsequently corrected for partial volume effects (partial volume correction, PVC) using the modified method of Müller-Gärtner et al. [36]. All images were then warped to MNI template space using the deformation field obtained during the segmentation step. [  An approach similar to that used by Josephs et al. [21] was used to compare the regional effect sizes between variants.
Regions of interest (ROIs) were selected from the Automated Anatomical Labeling (AAL) atlas [37] based on their typical involvement in the different variants of PPA [21]. In addition to the regions used by Josephs et al. [21] a medial temporal region (hippocampus, parahippocampus and amygdala) and a medial parietal region (posterior cingulum and precuneus) were also included. The binary ROIs were made subject-specific by intersecting them with the individual's grey matter map, thresholded at 0.3 [32].

Volumetric MRI acquisition and analysis
A high resolution T 1 -weighted structural MRI scan was acquired on the same day as the neuropsychological testing and used for processing the PET data. As a secondary analysis voxel-based morphometry (VBM8) [38] was performed on the T 1 -weighted images as previously described [33]. For voxel-based statistical analyses, modulated grey matter maps were smoothed with an 8-mm FWHM gaussian 3D kernel.

Statistical analysis
Standard statistical analyses were performed using SPSS, version 24 (IBM Statistics, Armonk, NY). The significance threshold was set at α < 0.05 for all standard statistical analyses. Demographic and neuropsychological data were statistically compared between groups using the two-tailed  [31] developed using in-house software. The threshold was set at voxel-level uncorrected P < 0.001.

Secondary outcome analyses
[ 18 F]THK5351 binding and clinical impairment To reveal the underlying structure and to reduce the dimensionality of the neurolinguistic dataset, a factor analysis was performed in SPSS. PPA patient 18 was excluded because only limited neuropsychological data were available as a result of disease severity. Two healthy control subjects were also excluded because data were missing. Hence, the data from 40 subjects (19 patients and 21 healthy controls) were available for factor analysis (Kaiser-Meyer-Olkin's test score of 0.77 and Bartlett's test P < 0.001). The principal axis factoring method was applied to extract factors with an eigenvalue of >1.0 (Kaiser's criterion). The eigenvalue of each factor corresponds to the proportion of variance explained by that factor. The factors were rotated with a VARIMAX orthogonal rotation to obtain interpretable and uncorrelated factor loadings. Only factor loadings greater than 0.7 were retained [40]. In a next step, individual weighted factor scores were calculated for each extracted factor using a regression-based approach. Weighting refers here to the factor loading of that variable, i.e. the higher the factor loading of a variable, the more that variable is associated with the underlying factor. Missing values were replaced by the mean score from the entire group for that variable. In a next step, a whole-brain voxel-wise linear regression analysis across PPA patients was performed for each extracted factor separately, with the individual weighted factor scores as the independent variable and the [ 18 F]THK5351 PVC SUVR images as dependent variable, corrected for education, age and gender. Only individual weighted factor scores that were not a priori classified by Grubb's test as significant outliers (α < 0.05) were included in the whole-brain voxel-wise regression analysis. As Grubb's test classified the weighted factor 2 score of patient 17 as a significant outlier, this patient was excluded from the regression analysis. The significance threshold was a voxel level uncorrected P < 0.001 combined with a cluster level family-wise error correction at P < 0.05 [39].
For the sake of completeness, whole-brain voxel-wise linear regression analyses across PPA patients were also performed for the individual tests used in the factor analysis (after outlier exclusion). For each significant cluster, the mean voxel value in that cluster was extracted for each individual PPA patient to visualize the slope of the correlation between the variables solely for illustrative purposes.

Topography of [ 18 F]THK5351 binding and atrophy
To compare the pattern of [ 18 F]THK5351 binding with the pattern of atrophy, a voxel-wise ANOVA was performed with modulated grey matter images as within-subjects factor and diagnostic groups as between-subjects factor, similar to the procedure described in section Primary outcome analysis.

Primary outcome analysis
Voxel-level statistical analysis of [ 18 F]THK5351 binding in PPA variants NFV patients showed increased [ 18 F]THK5351 binding in the supplementary motor area, precentral gyrus and premotor cortex bilaterally, in the left cingulum, pars triangularis, pars opercularis, insula, basal ganglia, thalamus, and in the midbrain nuclei (subthalamic nucleus, red nucleus and substantia nigra) compared with controls in the PVC-based analysis ( Fig. 1a; Table 2). SV patients showed significantly increased [ 18 F]THK5351 binding in the temporal lobes bilaterally (right > left), including the temporal pole, inferior and middle temporal gyri, amygdala, anterior hippocampus, parahippocampal gyrus and entorhinal cortex, and in the fusiform gyri and the right ventromedial frontal cortex compared with controls ( Fig. 1b; Table 2). LV patients did not show elevated binding compared with controls, most probably because of the small sample size and the heterogeneity among the individual patterns. Similar results were obtained without PVC for NFV and SV patients compared with controls, although the clusters of elevated [ 18 F]THK5351 binding in the left hemisphere of NFV patients appeared larger than with PVC ( Supplementary Fig. 1a) and an additional cluster was obtained in the left ventromedial frontal cortex in SV patients (Supplementary Fig. 1b).
The pattern of increased binding of [ 18 F]THK5351 in the left-hemispheric regions in NFV patients compared with SV patients was similar to that seen for the comparison between NFV patients and controls, with the highest binding in the supplementary motor area (Fig. 1c; Table 2). The pattern of increased [ 18 F]THK5351 binding in SV patients was similar to that seen for the comparison between SV patients and controls, with the highest binding in the right fusiform gyrus and the right inferior and middle temporal gyri (Fig. 1d). Binding in the right ventromedial frontal cortex was also increased in SV patients compared with NFV patients (Fig. 1d; Table 2). The results of between-group comparisons without PVC showing increased binding in NFV and SV patients were similar to those with PVC ( Supplementary Fig. 1c, d).
At the individual patient level, of 12 NFV patients, 8 showed significantly increased binding in the left premotor cortex, 9 in the supplementary motor area, 8 in the midbrain and 5 in the basal ganglia. One NFV patient also showed increased binding in the left frontal and temporal cortical regions. The amyloidpositive NFV patient (patient 13) showed increased binding in the supplementary motor area and the dorsal premotor cortex (Fig. 2a). NFV patient 17 also showed elevated binding in these regions together with more widespread binding in the temporal and parietal neocortex (Fig. 2b).
All five SV patients showed binding in the anterior temporal lobe, four showed binding in the ventromedial frontal cortex, and two also showed increased binding in the anterior hippocampus and amygdala. One of the three LV patients (patient 7) showed widespread [ 18 F]THK5351 binding in the temporal, parietal and frontal cortices and in the precuneus bilaterally (Fig. 2c). The two other LV patients showed only restricted binding foci: patient 11 showed binding in the middle occipital gyrus bilaterally (Fig. 2d), and patient 9 showed minimally increased binding in the right middle temporal lobe and left premotor cortex.  (Fig. 3a). The representative SV patient (patient 1) showed highly increased [ 18 F]THK5351 signal in the temporal lobes bilaterally and to a lesser degree in the ventromedial frontal cortex bilaterally (Fig. 3b). The representative LV patient (patient 7) showed increased [ 18 F]THK5351 signal in the temporal lobes, temporo-occipital cortex, and the angular gyrus with additional involvement of the precuneus and frontal lobes (left > right; Fig. 3c). These representative PPA patients all showed binding in the basal ganglia, a pattern that was also seen in the control(s) (Fig. 3d).

Secondary outcome analysis [ 18 F]THK5351 binding and degree of clinical deficit
The factor analysis yielded two factors ( Table 3). The first factor grouped measures of language comprehension and object recognition. The second factor grouped measures of agrammatism and motor speech impairment. The whole-brain voxel-wise regression analysis revealed a correlation between lower  Table 2, Supplementary Fig. 2).

Topography of [ 18 F]THK5351 binding and atrophy
As shown in Fig. 5, [ 18 F]THK5351 binding and MRI atrophy patterns were similar. The LV patients showed no significant atrophy at the pre-set significance threshold, most likely because of the small sample size and the heterogeneity in LV.

Differences in effect size between variants
Evaluation of the regional effect sizes on an individual basis showed the highest regional effect sizes (regional SUVR >4) in two NFV patients (patients 17 and 18; Fig. 6). The significance threshold was set at voxel-level uncorrected P < 0.001 combined with cluster-level family-wise error-corrected threshold P < 0.05 NFV nonfluent variant, SV semantic variant, T t value, Z z value T-contrasts are overlaid on MNI template brain renderings and on coronal slices (brighter colour means higher t value). The significance threshold was set at voxel-level uncorrected P < 0.001 for the Crawford t-contrasts. Patient numbers refer to Table 1 Discussion [ 18 F]THK5351 imaging revealed characteristic patterns of binding in the three variants of PPA, with a topography that closely matched the anatomical distribution of the predicted underlying pathology for each variant [3,4]. Elevated [ 18 F]THK5351 binding in the left supplementary motor area and in the left postcentral gyrus correlated with clinical measures of agrammatism and speech production deficits. In SV patients, [ 18 F]THK5351 binding was consistently increased in the anterior temporal cortex despite this being most commonly a TDP-43 proteinopathy. NFV PPA is most frequently associated with a tauopathy (3R or 4R tau) [2][3][4]. Increased [ 18 F]THK5351 binding was observed in all but one of the NFV patients with a topography that matched the known anatomical distribution of the lesions: elevated [ 18 F]THK5351 binding in NFV patients compared with controls included the medial frontal, premotor and inferior frontal cortical regions as well as subcortical regions, namely the midbrain, thalamus and basal ganglia. There was also increased binding in the frontal subcortical white matter. These regions have increased vulnerability to FTLD tau [7,41]. While the initial and most salient feature in NFV patients was language and speech impairment, a subset of NFV patients in this study showed mild clinical signs and symptoms that may be indicative of underlying PSP or CBD pathology, including right-sided extrapyramidal signs and vertical eye movement abnormalities (Supplementary Table 1). These patients may develop a PSP-like or CBD-like syndrome over time, similar to that described in patients with primary progressive apraxia of speech [42].
Two of the NFV patients were amyloid-positive. One had a focal pattern of uptake in the supplementary motor area and in the dorsal premotor cortex, and the other had widely distributed binding in the neocortical association zones of the left hemisphere, according to a pattern identical to that typically seen in AD, predominantly in the left hemisphere (Fig. 2). The THK5351 pattern in the former patient is in line with the  Table 1 notion that the distribution of tau PET abnormalities corresponds more closely to the clinical phenotype than amyloid PET [26], which was more isocortically elevated.
Elevated [ 18 F]THK5351 binding in the left supplementary motor area was correlated with the degree of agrammatism and motor speech impairment. [ 18 F]AV1451 binding in the supplementary motor area, dorsal premotor cortex and inferior frontal gyrus has been demonstrated previously in primary progressive apraxia of speech [43], a syndrome that most often shows underlying FTLD tauopathy on postmortem examination [5], and in patients with corticobasal syndrome who have apraxia of speech [44]. The current study is the first to show a quantitative relationship between speech apraxia measures and [ 18 F]THK5351 binding. The supplementary motor area plays a crucial role in speech motor control [45]. Premotor cortical involvement has been linked to the severity of speech apraxia in MRI and [ 18 F]FDG PET studies [46]. Damage to the white matter tract connecting the supplementary motor area to the inferior frontal gyrus (i.e. the aslant tract) affects the amount of distortion errors that NFV PPA patients make during spontaneous speech [45].
No correlation was observed between [ 18 F]THK5351 binding and the first factor related to language comprehension. However, when the scores on the individual tests were correlated with [ 18 F]THK5351 binding, correlations were seen between temporal cortical binding and the BNT, PPT and BORB scores. Based on MRI or [ 18 F]FDG PET findings in SV patients [1], one would predict that in particular the left temporal lobe would be involved in comprehension deficits. SV patients indeed showed elevated [ 18 F]THK5351 binding in this region compared with healthy controls and NFV PPA patients (Fig. 2). This finding is in accordance with the findings of previous studies of [ 18 F]THK5351 binding [19,24,25] and [ 18 F]AV1451 binding in SV patients [21][22][23]. The current cohort included only three LV patients. The pattern in LV was heterogeneous, in line with the heterogeneous composition of LV in general (for review see [6]). In one patient the pattern resembled that seen in typical AD (Fig. 2c) [17].
Overall, the outcome of the current study is similar to that in another recent study of the binding of [ 18 F]AV1451 in PPA patients. The two tracers show topographical patterns that differentiate the three different PPA variants [21] and both studies showed focal binding in regions of predilection even in SV. There are also some differences between the two studies, which might be related to the tracer used or to the sample studied. At the individual level, the strongest regional effect sizes for [ 18 F]AV1451 were obtained in LV patients [21], while in the current study the maximal regional effect sizes were highest in NFV patients and lowest in LV patients (Fig. 6). This could suggest differences between the tracers in their affinity for different types of tauopathy. Second, [ 18 F]AV1451 binding in the frontal cortex was mostly restricted to the subcortical white matter, while [ 18 F]THK5351 in the current study also showed neocortical binding in the regions typically affected in NFV, such as the supplementary motor area, precentral gyrus and inferior frontal gyrus (Fig. 1).
Two important issues regarding the first generation of tau PET tracers remain. First, high binding is systematically observed in the basal ganglia even in healthy controls for both [ 18 F]THK5351 [19] and for [ 18 F]AV1451 [10-12, 14, 47]. The [ 18 F]AV1451 binding pattern might be attributed to neuromelanin in the basal ganglia and substantia nigra [11,12,14] or to affinity for monoamine oxidase-B (MAO-B) [48,53]. Second, high binding is observed in SV, a disease that is most often associated with TDP-43 type C [2][3][4]7]. As a first and obvious possibility, in vivo [ 18 F]THK5351 may bind to the β-sheet conformation of TDP-43 type C aggregates. An [ 18 F]THK5351 autoradiography study on human brain sections demonstrated higher affinity for the β-sheet conformation of tau than for fibrillary amyloid and no affinity for TDP-43 and α-synuclein [9]. An autoradiography study with a different tracer, PBB3, showed binding to α-synuclein if the aggregates were available in sufficiently high amounts [49], but this cannot yet be extrapolated to [ 18 F]THK5351. Nevertheless, it remains possible that [ 18 F]THK5351 binding in the anterior temporal lobe reflects binding to the β-sheet conformation of non-tau aggregates. As an alternative explanation, binding to MAO-B may play a role given the reactive astrogliosis that accompanies neurodegeneration in a stage-dependent manner [50] and the  [48]. In the AD dementia stage, neocortical binding of [ 11 C]deuterium-L-deprenyl is not increased compared with that in controls [51]. Provided that the degree of reactive astrogliosis in neocortical regions of predilection is comparable between AD and PPA, reactive astrogliosis cannot account for the strong signal in SV in the temporal cortex. [ 18 F]AV1451 also binds to MAO-A [53], and this could also be the case for [ 18 F]THK5351. As a final possibility, these first-generation tau PET tracers may bind to a molecular target that is focally present in the regions of highest neurodegeneration, such as haem by-products or mineralized structures or a yet-undefined molecular target [14,48]. Further work is needed to evaluate these  The significance threshold was set at voxel-level uncorrected P < 0.001 combined with cluster-level family-wise error-corrected threshold P < 0.05, T t value, Z z value different possibilities before the first-generation tau PET tracers can be considered valid biomarkers of the pathological processes underlying PPA. The currently available CSF tau biomarkers are unable to positively detect underlying FTLD pathology [54] and are mainly useful for exclusion, i.e. to rule out AD as a cause [32]. [ [19] have demonstrated that [ 18 F]THK5351, despite its offtarget binding, might be a valid option for measuring pathology associated with PPA. However, it is important that extensive preclinical characterization is performed including assessment of binding to tau maturation stages and to cell types affected before moving to clinical phase trials.

Study limitations
No firm conclusions regarding the underlying type of tauopathy can be drawn as this study was limited by the lack of postmortem confirmation. Furthermore, the total number of PPA patients included was rather small, which is partly inherent in the relatively low prevalence of the syndrome [3]. The negative findings in LV patients are most likely due to the limited statistical power. Individual LV patients did show elevated signal, suggestive of binding to AD tauopathy. A Significantly increased [ 18 F]THK5351 binding in SUVR images with partial volume correction and significant atrophy, compared using voxelwise ANOVA, and depicted as a t-contrast overlaid on an MNI template brain renderings (brighter colour means higher t value). Nonfluent variant (NFV) PPA patient (a) and semantic variant (SV) PPA patient (b) compared with healthy controls (HC). c Higher binding and atrophy are apparent in the NFV patient compared with the SV patient. d Higher binding and atrophy are apparent in the SV patient compared with the NFV patient. The significance threshold was set at voxel-level uncorrected P < 0.001 combined with cluster-level family-wise error-corrected threshold P < 0.05 drawback of using [ 18 F]THK5351 PET is the nonspecific binding. Nonetheless, elevated [ 18 F]THK5351 binding seen in the current study was highly focalized and colocalized with regions known to be associated with conditions in which a tauopathy is the underlying neuropathological cause [4,7,41].

Conclusion
The primary outcome analysis showed that the [ 18 F]THK5351 patterns were highly subtype-specific and in accordance with the predictions regarding the topography of the hallmark lesions in the different PPA variants. We were able to observe increases in signal colocalizing with areas of predilection in vivo. Elevated [ 18 F]THK5351 binding in the left supplementary motor area correlated with clinical features, i.e. the degree of agrammatism and speech production deficit. The tracer may therefore serve as a topographically specific marker of neurodegeneration in patients with the different subtypes. Each row corresponds to a patient; the colour indicates the variant, and the individual data points correspond to the regional SUVR values of each of the ten ROIs per patient. The presentation of this figure is similar to that used by Josephs et al. [21]