Longitudinal tau and metabolic PET imaging in relation to novel CSF tau measures in Alzheimer’s disease

Purpose Studies comparing CSF and PET tau biomarkers have included only commercial CSF assays examining specific phosphorylation sites (e.g. threonine 181, P-tau181p) and mid-domain tau (i.e. total tau, T-tau). Moreover, these studies did not examine CSF tau levels in relation to cerebral glucose metabolism. We thus aimed to examine CSF tau measures, using both commercial and novel assays, in relation to [18F]THK5317 (tau) and [18F]FDG PET (glucose metabolism). Methods Fourteen Alzheimer’s disease (AD) patients (seven prodromal, seven dementia) underwent [18F]THK5317 and [18F]FDG PET studies, with follow-up performed in ten subjects (six prodromal, four dementia) after 17 months. In addition to commercial assays, novel measures capturing N-terminus+mid-domain (tau N-Mid) and C-terminally truncated (tau-368) fragments were included. Results While the levels of all forms of CSF tau were found to be inversely associated with baseline [18F]FDG uptake, associations with baseline [18F]THK5317 uptake varied in relation to the degree of isocortical hypometabolism ([18F]FDG SUVR). Changes in the levels of the novel CSF markers tracked longitudinal changes in tracer uptake better than changes in P-tau181p and T-tau levels, and improved concordance with dichotomized regional [18F]THK5317 measures. Conclusion Our findings suggest that neurodegeneration may modulate the relationship between CSF and PET tau biomarkers, and that, by comparison to P-tau181p and T-tau, tau-368 and tau N-Mid may better capture tau pathology and synaptic impairment. Electronic supplementary material The online version of this article (10.1007/s00259-018-4242-6) contains supplementary material, which is available to authorized users.


Introduction
Although many details surrounding the pathogenesis of Alzheimer's disease (AD) remain unclear, a dominant viewpoint is that the dysmetabolism of amyloid-β (Aβ) initiates tau pathology that results in neurodegeneration and cognitive decline. Cerebrospinal fluid (CSF) measures of tau include phosphorylated tau (P-tau), an indicator of an abnormal state associated with the formation of neurofibrillary tangle (NFT) pathology, and total tau (T-tau), that reflects the intensity of axonal and neuronal damage [1]. Recently, positron emission tomography (PET) ligands selective for AD-related paired helical filament (PHF) tau present in NFTs have been introduced into the field.
One such ligand is [ 18 F]flortaucipir (formerly known as [ 18 F]T-807 and [ 18 F]AV1451) [2,3], and studies examining the relationship between [ 18 F]flortaucipir uptake and CSF tau levels have shown modest positive correlations in cognitively unimpaired older adults and in patients with AD and non-AD neurodegenerative disorders [4][5][6][7]. These studies, however, used commercial assays that use monoclonal antibodies directed against specific phosphorylation sites (e.g. threonine-181, P-tau 181p ) and the mid-region of tau (i.e. T-tau) [8,9]. Increasing evidence, however, indicates the presence of tau fragments spanning both the mid-domain and various terminal regions [10], suggesting that assays targeting specific variants of tau may be required to more fully characterize the relationship between CSF and PET tau biomarkers. Moreover, these studies did not examine the link between CSF tau levels and [ 18 F]FDG PET, an important measure given evidence indicating that synaptic integrity mediates the impact of tau on cognition [11].
The aims of this study were thus to investigate the association between CSF tau levels, derived from both commercial assays (P-tau 181p and T-tau) [8,9] and novel assays, the latter including N-terminus plus mid-domain regions (tau N-Mid) and C-terminally truncated tau (tau-368), and longitudinal findings using the tau PET tracer [ 1 8 F]THK5317 [(S)-[ 18 F]THK5117] [12][13][14] and [ 18 F]FDG in a sample of patients with AD. Given findings suggesting that disease progression may be accompanied by a reduction in CSF tau [15,16], it was hypothesized that the degree of glucose hypometabolism may modulate associations between CSF tau levels and [ 18 F]THK5317 uptake. Further, in light of the likely future clinical use of tau imaging, we also aimed to examine the agreement between dichotomized CSF tau levels and [ 18 F]THK5317 uptake.
All patients had originally undergone clinical assessment at Them e A ging, K arolinska U niversity Hospital, Stockholm, Sweden, and underwent follow-up, as detailed elsewhere [17]. Patients diagnosed with AD dementia fulfilled the criteria for probable AD and showed in vivo evidence of Aβ pathology with an abnormal [ 11 C]Pittsburgh compound B [ 11 C]PIB) scan [19]. Patients diagnosed with prodromal AD met the criteria for mild cognitive impairment [20] and showed an abnormal [ 11 C]PIB scan [21]. THK-controls were recruited through Clinical Trial Consultants AB (Uppsala University Hospital, Uppsala, Sweden) or from among patients' relatives; all were nonsmokers and not taking medication, completed extensive clinical evaluations, and at inclusion had no cognitive complaints, prior head injury or known neurological/psychiatric disorders. Elderly THK-controls were [ 11 C]PIB-negative, and showed no differences in regional [ 18 F]THK5317 binding as compared with young THKcontrols.
Informed consent was obtained from all individual participants included in the study. Ethical approval was obtained from the Regional Human Ethics Committee of Stockholm and the Faculty of Medicine and Radiation Hazard Ethics Committee of Uppsala University Hospital, Uppsala, Sweden. A further 15 individuals were included as CSF biomarker-negative controls (CSF-controls); these were patients with minor cognitive complaints but with basic (cell count, albumin ratio, IgG index) and core (Aβ 1-42 /Aβ 1-40 , P-tau 181p , and T-tau) CSF biomarkers within normal ranges. The CSF samples used were deidentified leftover aliquots from routine clinical analyses, following a procedure approved by the Ethics Committee of the University of Gothenburg (EPN 140811).
INNOTEST enzyme-linked immunosorbent assay (ELISA) kits (Fujirebio) for P-tau 181P and T-tau, an inhouse ELISA targeting tau species from the N-terminus to the mid-domain of tau (tau N-Mid), and a single molecule array (Simoa) method targeting a C-terminally truncated form of tau ending at amino acid 368 (tau-368) [22]. Aβ 1-42 and Aβ 1-40 were determined using the Meso Scale D i s c o v e r y A b e t a Tr i p l e x i m m u n o a s s a y ( M S D , Gaithersburg, MD) according to the kit insert. Additional methodological details, including those pertaining to the validation of the tau N-Mid and tau-368 assays, are included in Online Resource 1.

PET data acquisition and analysis
Dynamic [ 18 F]THK5317 PET studies were performed over 60 min following intravenous bolus injection of 217 ± 42 MBq. A static 15 min [ 18 F]FDG scan was performed 30 min after injection of 3 MBq/kg. Baseline and follow-up images were coregistered to their respective T1weighted MR images using PMOD v. 3.5 (PMOD Technologies., Zurich, Switzerland). In order to account for possible spillover effects from the white matter, MRIbased partial volume correction was applied to [ 18 F]THK5317 PET images, as implemented in PMOD [23].
Distribution volume ratio (DVR) images were calculated for [ 18 F]THK5317 using the cerebellar cortex as the reference tissue [12,24]. For [ 18 F]FDG [18], standardized uptake value ratio (SUVR) was used as the outcome measure using the pons as the reference region.
Following segmentation of T1 images using SPM8, the inverse transformation parameters generated were used to spatially warp a probabilistic atlas into native space in order to sample the parametric images [25]. Bilateral grey matter regions of interest (ROIs) were based on previously published findings with [ 18 F]THK5317 [17], and included the frontal, temporal (medial, lateral aspects), parietal, posterior cingulate and occipital cortices, as well as an isocortical composite ROI comprising these regions minus medial temporal structures. Two additional composite regions approximating the Braak staging scheme for tau pathology [26] were also included: a limbic ROI ( B r a a k I I I / I V ) c o m p r i s i n g t h e h i p p o c a m p u s , parahippocampus, amygdala, fusiform and middle inferior temporal gyri, orbital and straight frontal gyri, parietaltemporal-occipital junctions and the temporal poles, and an isocortical ROI (Braak V/VI) comprising all isocortical regions except the precentral and postcentral gyri. Annual rates of change in [ 18 F]THK5317 DVR and [ 18 F]FDG SUVR were calculated as follows: (follow-up value − baseline value)/between-scan interval (years) [18]. Determination of regional [ 18 F]THK5317 and CSF tau cut-off values CSF P-tau 181p and T-tau values were classified as abnormal using cut-off values of >80 pg/mL and > 400 pg/mL, respectively [27]. Unpublished internal cut-off values of >272 pg/mL and <0.038 were used for tau N-Mid and tau-368/T-tau, respectively. These were determined using receiver operating characteristic (ROC) analysis (highest combined sensitivity and specificity) among 50 biomarker-positive AD patients (low Aβ 1-42 , high Ptau 181p and T-tau) and 50 cognitively unimpaired biomarker-negative individuals. Regional [ 18 F]THK5317 DVR cut-off values were defined using the mean plus two standard deviations of the DVR values in the THKcontrol group (elderly and young combined), in agreement with previous approaches for both tau [6] and Aβ PET [28]. Given the lack of cut-off values for tau-368 and tau-368/tau N-Mid, as well as the absence of controls with both CSF and [ 18 F]THK5317 data, concordance using these measures was not investigated.

Statistical analysis
Patient characteristics and CSF measures were compared using the Mann-Whitney U test. To examine the relationship between regional PET ([ 18 F]THK5317 DVR and [ 18 F]FDG SUVR) and CSF measures, nonparametric rank-based estimation multilinear regression models were implemented [29]. Here, PET measures were regressed onto CSF values, adjusting for age, ventricular volume and the time between the CSF and PET investigations. For models examining CSF tau values in relation to baseline [ 18 F]THK5317 DVR, isocortical composite [ 18 F]FDG SUVR, included as an interaction term with CSF tau, was binarized using a median value of 1.4. Simple slope analyses were used to assess multiple linear regression twoway interactions [30]. All analyses were performed using R, v.3.3.2, with significance set at p < 0.05, two-tailed. ROC-derived areas under the curve (AUC) for CSF tau measures were compared between AD patients and CSFcontrols using the pROC package, v.1.10 [31].

Results
Demographic, clinical and CSF data are presented in Table 1. As expected, baseline Mini-Mental State Examination (MMSE) scores were significantly lower in AD dementia patients than in prodromal AD patients (p < 0.01). MMSE scores at follow-up were significantly lower than scores at baseline, in prodromal AD patients (p < 0.01), and across prodromal AD and AD dementia patients combined (p < 0.01). Tau-368/T-tau ratios were lower in AD dementia patients than in prodromal AD patients (p < 0.05).
According to the selection criteria for CSF-controls, the levels of P-tau 181p and T-tau were significantly higher in AD patients. A clear group separation was also seen when using novel tau fragments, with AUC values for these close to 1. There were no significant differences in AUC values between groups (Online Resources 2, 3 and 4).

Relationship between baseline PET measures and CSF tau
Regression models for baseline [ 18 F]THK5317 DVR (Fig. 1a, c; Online Resource 5) showed that tracer retention in the lateral temporal and parietal lobes, as well as in the isocortical composite and Braak V/VI ROIs, was associated with Ptau 181p levels, while T-tau levels were associated only with isocortical composite uptake. Tau-368 levels were associated with [ 18 F]THK5317 DVR in the lateral temporal and frontal regions, as well as in the isocortical and Braak III/IV ROIs, and tau-368/tau N-Mid ratios were significantly associated with [ 18 F]THK5317 DVR in the temporal ROIs, as well as in the parietal and occipital cortices. In contrast, tau-368/T-tau ratios were associated with [ 18 F]THK5317 DVR only in the parietal lobe. Although the direction of these associations was negative, with the exception of tau ratios, as hypothesized, the relationship between CSF tau levels and [ 18 F]THK5317 DVR varied according to the degree of glucose hypometabolism. Specifically, a significant interaction was observed between CSF tau measures and isocortical metabolism ([ 18 F]FDG SUVR) whereby those with higher metabolism (SUVR >1.4; n = 7, five prodromal AD, two AD dementia) showed positive slopes for the above-described associations, while those with more impaired metabolism (SUVR <1.4; n = 6, four prodromal AD, two AD dementia) showed negative slopes, similar to the findings observed at the group level (in analyses using ratios, the additional AD dementia subject had an SUVR >1.4; Online Resource 6).
A significant association was observed between baseline parietal [ 18 F]FDG SUVR and P-tau 181p levels (Fig. 1b,   Concordance between tau biomarkers Only isolated CSF positivity was found when using tau-368/T-tau. This pattern held when subgroups were examined separately, although concordance was higher overall in AD dementia patients than in those with prodromal AD (Online Resources 11 and 12).

Illustrative cases
The

Prodromal AD
The PET and CSF biomarker findings in two patients with prodromal AD are presented in Fig. 4a

AD dementia
The PET and CSF biomarker findings in two patients with AD dementia are presented in Fig. 4b Fig. 4 Imaging and CSF profiles in patients with a baseline diagnosis of prodromal AD (a, c), and AD dementia (b, c). Given the widespread use of P-tau 181p and T-tau in the clinical work-up of patients with dementia disorder, these measures were selected to define concordance (a, b).
[ 18 F]THK5317 positivity was considered present when a high percentage of ROIs showed DVR values above the defined cut-off values for THK-controls

Discussion
In this retrospective longitudinal study, the relationship between PET and CSF tau biomarkers was found to vary in relation to the degree of isocortical glucose metabolism. While a negative association was observed in those with more globally decreased [ 18 F]FDG uptake, positive slopes, in line with the findings of previous studies, were found in those with more preserved glucose metabolism. This interaction with [ 18 F]FDG uptake was observed for both established and novel CSF tau fragments. The levels of all forms of CSF tau investigated were found to be negatively associated with [ 18 F]FDG SUVR. Given the unclear relationship between established CSF tau biomarkers and NFT burden [32,33], increases in P-tau 181p and T-tau levels may in fact reflect increased tau secretion resulting from the spread of AD-affected cells at risk of tangle pathology [34], with increases in tau-368 levels reflecting an expansion in the pool of C-terminal fragments, possibly due to increased cleavage activity. In this way, higher levels of these biomarkers would precede the appearance of NFT pathology and neurodegenerative changes on [ 18 F]FDG PET. This scenario may explain the positive associations with [ 18 F]THK5317 uptake at baseline observed in this study and in previous studies using [ 18 F]flortaucipir [4][5][6][7]. The inverse association between tau biomarker levels seen in patients with more impaired metabolism, by contrast, may reflect a decrease in the release of neuronal tau to the CSF. Indeed, several studies, including recent work using stable isotope labelling kinetics (SILK) to monitor the half-life and turnover rate of tau [35], have shown that, following an initial increase, CSF tau levels decrease in patients with symptomatic AD [15,16,[36][37][38]. This may reflect a deceleration in neurodegenerative processes and/or a reduction in the number of viable neurons releasing tau. The inverse association with tau-368 levels may also reflect a slowing in the production of tau. Another intriguing hypothesis, however, is that this truncated tau form may be sequestered in NFTs and thus represent a more direct tangle marker than, for example, P-tau 181p . Possibly, ratios incorporating tau-368 (particularly the tau-368/N-Mid ratio) might better reflect this process, in much the same that way Aβ 1-42 /Aβ 1-40 better reflects Aβ pathology. The strong associations between tau-368 levels and baseline [ 18 F]THK5317 uptake, and between tau-368 ratios and the change in [ 18 F]THK5317 uptake, lend support to this hypothesis.
Previous studies addressing the relationship between crosssectional [ 18 F]FDG uptake and CSF tau levels (P-tau 181p and T-tau) in patients with AD have shown mixed results: although all have shown an inverse association, some have shown no significant association [39], significant associations only for P-tau 181p [40] and T-tau [41] levels, or significant associations for both [42]. In our cohort, a significant negative association was found only between P-tau 181p levels and parietal metabolism at baseline. Interestingly, P-tau 181p levels were also associated with the rate of change in [ 18 F]THK5317 DVR in the parietal lobe, with this pattern (i.e. CSF tau levels associated with baseline metabolism and the rate of change in [ 18 F]THK5317 uptake in the same ROI) also observed for the tau-368/T-tau ratio (lateral temporal), tau-368/tau N-Mid ratio (isocortical composite) and, most strikingly, tau N-Mid levels (posterior cingulate and Braak composite regions). Together, these findings suggest a nonlinear relationship between the accumulation of AD-type PHF tau and neurodegeneration [18,43], similar to that observed for Aβ [44]. Further, our results suggest that tau N-Mid might prove suitable for capturing increases in tangle pathology and synaptic impairment, acting as both a state marker (reflecting the intensity of the disease process) and a stage marker (reflecting the cumulative extent of the degenerative process) [45].
The level of agreement between dichotomized [ 18 F]THK5317 uptake and CSF tau levels was somewhat lower, overall, than that found in the only other study so far to have examined concordance between tau biomarkers ([ 18 F]flortaucipir, and P-tau 181p and T-tau) [6]. This may relate to cohort-specific differences (age and CSF tau levels, as well as MMSE scores were higher and lower, respectively, in the study by Mattsson et al. [6], which also had a shorter time between CSF sampling and PET), ROI definition, tracer characteristics, and, possibly, the control subjects used to define tau PET cut-off levels. Similar to the study by Mattsson et al. [6], however, we also found higher concordance rates among AD dementia subjects than in subjects with prodromal AD. However, using single biomarkers concordance was markedly improved when using tau-368/T-tau. Although the findings for tau-368/T-tau and baseline [ 18 F]THK5317 DVR were limited, this improved concordance, combined with the strong association between this measure and longitudinal increases in [ 18 F]THK5317 retention, suggests that the tau-368/T-tau ratio may better reflect the shift between soluble and insoluble pools of tau than the single analytes (P-tau 181p , T-tau, tau N-Mid) investigated. This would also probably hold true for tau-368 alone and as a ratio with tau N-Mid, although we were unable to directly test this in this study due the lack of a priori cut-off values or controls with both CSF and [ 18 F]THK5317 PET data available.
With regard to the patterns of discordance, isolated [ 18 F]THK5317 positivity was the predominant finding for Ptau 181p , T-tau and tau N-Mid. By contrast, only isolated CSF positivity was found using tau-368/T-tau. Possibly, the isolated [ 18 F]THK5317 positivity may reflect a deceleration in neurodegeneration and/or the spread of NFT pathology, as previously described [16,37]. The pattern of discordance seen with tau-368/T-tau lends partial support to this idea, in that the spread of tangle pathology would be accompanied by a decrease in soluble C-terminal fragments, a process that would precede suprathreshold [ 18  showing values below the thresholds for P-tau 181p , T-tau and tau N-Mid. However, both these patients showed low baseline isocortical glucose metabolism. These observations, combined with our regression findings, suggest that tau biomarkers may initially follow an offset parallel upward course, followed by divergence, as CSF tau levels decline. Although this may indeed reflect the sequestration of tau fragments in NFTs, given the strong coupling between tangle pathology and neurodegeneration in AD [46], this decrease in soluble tau may relate less to a slowing of degenerative processes and more to a shift in the way tau is released from tanglebearing cells [35].
Certain methodological limitations complicate the interpretation of the present findings. Firstly, in addition to the modest sample size, which precluded comparisons between subgroups (prodromal AD, AD dementia), not all participants completed follow-up PET. Also, due to the limited number of subjects, and the exploratory nature of this study, we did not correct for multiple comparisons. Moreover, the AD dementia subgroup represented a limited cross section of disease severity; with the exception of one AD dementia patient who had an MMSE score of 17, all had MMSE scores that placed them in the mildly impaired range. Ideally, Aβ-positive cognitively unimpaired elderly would also have been included so as to cover the entire AD continuum. In addition, although recent evidenc e i ndicates off-target binding of [ 18 F]THK5351 to monoamine oxidase B (MAO-B) [47][48][49], it remains unclear how well these findings extrapolate to [ 18 F]THK5317, given results showing that despite a similar cortical laminar distribution, the MAO-B-specific tracer [ 3 H]deprenyl and [ 3 H]THK5117 do not compete at the concentration ranges used in PET studies [50]. This, combined with the low levels of MAO-B in cortical regions [51], suggests that the DVR values reported here largely reflect specific binding of [ 18 F]THK5317 to PHF-tau. Further, longitudinal CSF sampling performed concurrently with PET studies would have allowed a better understanding of the trajectories of tau biomarkers, in particular whether a decline in CSF tau levels precedes increases in [ 18 F]THK5317 retention. Lastly, partial volume correction of longitudinal data would ideally have been performed using MRI at follow-up. However, given the low rates of atrophy reported within isocortical regions over comparable time intervals [52], this would not be expected to have had an important effect on our results.

Conclusion
Although our findings are preliminary and subject to the above-mentioned caveats, they suggest that both positive and negative associations may exist between CSF and PET tau biomarkers, and that the directionality of their relationship may depend on the type of tau form measured, and/or an asyet-unclear interplay between tau pathology and neurodegeneration. By comparison with P-tau 181p and T-tau, the novel fragments investigated in this study showed greater sensitivity toward findings with [ 18 F]THK5317 and [ 18 F]FDG PET, and improved concordance with binarized [ 18 F]THK5317 measurements. CSF and PET tau biomarkers, however, probably capture different aspects of tau pathology, similar to that proposed for CSF Aβ 1-42 and amyloid PET [53]. Further studies incorporating longitudinal CSF analysis and tau PET imaging in larger cohorts are required to clarify the findings reported here and to address their potential implications in the context of the recently proposed dichotomous classification scheme that seeks to combine tau biomarkers with those for Aβ and neurodegeneration [54].
Funding This study was funded by the Swedish Research Council (projects 05817 and 06086), the Swedish Foundation for Strategic Research (SSF), the Regional Agreement on Medical Training and Clinical Research (ALF) for Stockholm County Council, the Foundation for Old Servants, the Axel Linders Foundation, Gun and Bertil Stohne's Foundation, the KI Foundations, the Swedish Brain Foundation, the Swedish Alzheimer's Foundation, Demensfonden, the Wenner-Gren Foundations, KTH-SLL grant, the European Research Council, the Knut and Alice Wallenberg Foundation, the Torsten Söderberg Foundation, and the EU FW7 large-scale integrating project INMiND (http://www.unimuenster.de/InMind).

Compliance with ethical standards
Conflicts of interest A.N. has served as a consultant or on advisory boards for GE Healthcare, Merck AB, Lundbeck AB, Neuroscios, Pharmatrophix, TauRX Therapeutics and Eli Lilly. H.Z. has served on advisory boards for Eli Lilly and Roche Diagnostics, and has received travel support from Teva. K.B. has served as a consultant or on advisory boards for Alzheon, BioArctic, Biogen, Eli Lilly, Fujirebio Europe, IBL International, Merck, Novartis, Pfizer and Roche Diagnostics. H.Z. and K.B. are cofounders of Brain Biomarker Solutions in Gothenburg AB, a GU Ventures-based platform company at the University of Gothenburg. All other authors declare no conflicts of interest.
Ethical approval All procedures performed in studies involving human participants were in accordance with the ethical standards of the institutional and/or national research committee and with the principles of the 1964 Declaration of Helsinki and its later amendments or comparable ethical standards.
Informed consent Informed consent was obtained from all individual participants included in the study.
Open Access This article is distributed under the terms of the Creative Comm ons Attribution 4.0 International License (http:// creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.