Background
Accumulation of neurofibrillary tangles (NFT) is related to cognitive decline and is one of the neuropathological hallmarks of Alzheimer’s disease (AD) [
1]. Furthermore, AD progression has been shown to follow a specific pattern described by the Braak Stages [
2] such that
in-vivo quantification of NFTs may support disease staging and assessing the neurological condition of patients. This can be achieved with positron emission tomography (PET) tracers that bind to hyper-phosphorylated tau proteins, such as [
18F]MK-6240 which has sub-nanomolar affinity for NFTs and has been studied and validated recently in healthy control (HC), mild cognitive impairment (MCI), and AD subjects [
3‐
7].
Previous in-human [
18F]MK-6240 studies performed dynamic PET scans with a duration up to 180 min and used a reference tissue model with the cerebellum as reference brain region. This reference tissue model approach uses the time activity curve (TAC) of a reference tissue assumed to be devoid of specific tracer binding to estimate the ratio of specific to non-displaceable binding in target regions. This way, arterial cannulation for blood collection during the PET scan is obviated, avoiding an invasive and uncomfortable procedure, which often is not feasible in ill-conditioned patients. However, the long PET scanning times increases patient discomfort and the risk of head motion, which could hamper accurate quantification, and long scanning times are impractical for clinical routine scheduling. To address the issues of long dynamic PET scanning, one could consider standardized uptake value ratio (SUVR) relative to the cerebellum of a late static PET scan to estimate specific [
18F]MK-6240 binding. A previous study [
4] proposed static scanning at 90 min post-injection (p.i.), while another study [
5] found that SUVR was still increasing at 135 min p.i. in subjects with high [
18F]MK-6240 binding. In addition to the uncertainty about the optimal time-window for static [
18F]MK-6240 scanning, SUVR can be susceptible to blood–brain barrier (BBB) functionality changes and variations in cerebral blood flow (CBF). It has been shown that such changes can introduce a bias to SUVR-based longitudinal assessment of deposition of β-amyloid plaques [
8‐
10], another AD neuropathological hallmark. It can also be expected that drug intervention may have effects on perfusion. Given the high BBB permeability and fast plasma clearance of [
18F]MK-6240 [
6,
11,
12], one can also expect the influence of those factors and regional CBF functionalities to affect tau quantification. Thus, SUVR-based quantification should be considered carefully, notably in longitudinal settings for cases where the vascular component is expected to play an important role, such as for stroke patients or patients with cognitive decline. Furthermore, tau pathology and changes in relative CBF have been observed to independently contribute to cognitive deficits in AD [
13], emphasizing the need to understand the interplay between tau load, tracer binding, and longitudinal changes in CBF. Nevertheless, SUVR from [
18F]MK-6240 has potential for stratification of different stages of tau accumulation in a cross-sectional setting [
7].
To avoid the limitations of SUVR quantification, two studies implemented a dual-time-window (DTW) protocol consisting of first dynamic PET scan starting at tracer injection, followed by a short resting period of 15 min [
5] or 30 min [
6], and a second dynamic PET scan starting after the short break. However, these studies did not evaluate the potential impact of a DTW protocol on the PET quantification nor did they optimize the scan and break duration of their DTW protocol. For amyloid quantification, there has been an evaluation of the effects of a DTW and CBF changes [
14,
15]; however, no such exploration has been done for tau tracers.
The aim of the present study was to develop and evaluate a DTW protocol for accurate [18F]MK-6240 PET quantification that is comfortable for patients, suitable for longitudinal studies and drug intervention assessment, and appropriate for clinical routine. For this purpose, DTW protocols combining a first dynamic PET scan of variable length, starting at tracer injection, with a second 90- to 120-min PET scan were compared to an uninterrupted 120-min dynamic PET scan using a Logan reference tissue model (Ref Logan) with grey matter cerebellum as reference region. Furthermore, changes in perfusion were simulated to assess the impact of CBF and BBB permeability variations on Ref Logan and SUVR quantification, where SUVR was calculated relative to the grey matter cerebellum for the 90- to 120-min PET scan. Finally, small delays for the 90- to 120-min PET scan were simulated to evaluate the quantitative impact of possible non-compliance with the clinical routine protocol.
Discussion
For this study, a DTW PET protocol for [18F]MK-6240 PET imaging was considered while using a Ref Logan for DVR quantification, as well as the simulation of perfusion changes and scanning protocol non-compliance. As such, it was possible to reduce the overall acquisition time, therefore, potentially increasing patient comfort and clinical route feasibility while approximating a quantitative approach using full dynamic scanning. For the evaluation of different DTW protocols, dynamic PET data based on patient data was simulated up to 120-min post-injection and brain regions related to Braak Stages were considered. This way, realistic and clinically relevant datasets were generated for the regional assessment of a wide range of tau accumulation levels.
Presently, Ref Logan was assumed to be independent of the reference tissue clearance rate (
k2’), which requires a late equilibrium time. To assess if the mismatch between 2TCM and Ref Logan for high binding subjects comes from the violation of this assumption, the 2TCM cerebellar
k2 was used as
k2’ on the Ref Logan model. This was done with a population-based
k2’ as an average of the cerebellar
k2 from all subjects, and with each individual’s cerebellar
k2 as Ref Logan’s
k2’. As there was no improvement on the agreement between the kinetic models (Additional file
1: Fig. S1), it is reasonable to trust the applicability of the Ref Logan model as independent of reference tissue clearance rate and indicates that the mismatch for high binding subjects is not a violation of this assumption.
Changes in perfusion had, in general, a significant impact on Ref Logan DVR estimations of MCI/AD subjects. It was observed that halving
K1 lead to a regional DVR underestimation larger than 5% for most target regions, either when
R1 was constant or variable (Fig.
1). Such extreme change in perfusion is not traditionally expected on longitudinal assessment of patients; however, these simulations were included to better understand general trends in quantification as function of perfusion changes and include possible extreme cases that could come from an interplay between disease progression and a possible drug effect. For less drastic perfusion changes, the average DVR bias was less than 3.5% for any target region, perfusion change (constant
R1 or not), and subject diagnosis (Fig.
1). Therefore, such results indicate that a Ref Logan modelling using the cerebellar cortex as reference region can be a reasonable approach for longitudinal tau assessment of MCI and AD patients, or perhaps even for subjects undergoing treatment which may affect CBF and BBB functionalities. Furthermore, the simulation with variable
R1 violates the assumption of reference tissue models that the ratio
K1/
k2 is the same for both target and reference tissues. As such, this simulation tested the robustness of the Ref Logan model and showed reliable quantification even when using a reference region that may have some PET signal (e.g. from spill-over, partial volume effects, or non-specific binding).
Regarding the introduction of a gap in the time-activity curves, it was demonstrated that a DTW protocol with a 60-min break between two 30-min dynamic acquisitions provided, on average, a bias of less than 2.5% when compared to a full dynamic scanning protocol (Fig.
2). This bias level is on the range of the expected test–retest variability [
23‐
25] for AD studies and shorter breaks resulted in less quantification bias, which is in line with previous literature [
14,
26,
27]. However, compared to shorter breaks, a 60-min break DTW protocol allows for efficient use of hospital facilities without compromising quantification accuracy since it still provides enough flexibility to schedule an interleaved clinical study while having low bias when compared against an uninterrupted 120-min scan. Additionally, by halving the total acquisition time, this DTW protocol (TAC
DTW60) has the potential to decrease patient discomfort and, therefore, we considered this design as the DTW protocol of choice. Presently, a simple linear interpolation was performed to fill the gap in the TAC. Even though a previous study found that filling the DTW gap with more advanced techniques could improve quantification, linear interpolation still resulted in comparable results to those obtained from uninterrupted scanning [
27]. Another scanning protocol option previously explored in the literature would be one uninterrupted dynamic PET scanning session of 60 min using the reference Logan plot to estimate the DVR values, as this approach provided low quantitative bias when compared to a more lengthy scanning session [
5]. However, a scanning session of 60 min without a break could already be difficult to tolerate for patients suffering from significant cognitive decline. Moreover, the proposed DTW protocol is identical to other DTW protocols recommended for [
18F]flutemetamol and [
18F]florbetaben PET studies [
14], which further facilitates its implementation as having the same protocol for different studies can simplify clinical workflow.
Next, the impact of changes in perfusion on the Ref Logan DVR quantification using the DTW 60-min break protocol was compared to the DVR results from the full dynamic scan analysis (TAC
120). Furthermore, late time SUVR
90 of TACs with perfusion changes were compared with the SUVR
90 of the unaltered acquisition (fitted
K1 and
k2 parameters). It was observed that perfusion changes lead up to a regional DVR bias of 10% or less (Fig.
1), while for SUVR
90 it was up to − 19%, showing that quantification of static images is more vulnerable to perfusion changes than the Ref Logan model. Ref Logan DVR from TAC
DTW60 had a smaller range on the 95% confidence interval than SUVR90 (Additional file
2: Tables S8, S10, S14 and S16), showing that it can be more reliable against perfusion changes than SUVR. Furthermore, DVR dependence on perfusion with either constant or variable
R1 was similar: lower perfusion (lower
K1 values) decreased DVR of MCI/AD subjects and increased perfusion overestimated their DVR, while HC subjects were essentially unaffected (Fig.
1c, d). SUVR
90 dependence on perfusion was different with constant
R1 and variable
R1. With constant
R1, SUVR
90 of MCI/AD subjects was directly related to perfusion (similar to what was observed on DVR estimates), while for HC subjects the SUVR
90 had an opposite trend: increasing with lower perfusion and decreasing when perfusion increased (Fig.
1e). With variable
R1, on the other hand, the SUVR
90 variability for MCI/AD subjects was dependent of target region: on NFT-poor regions the behavior was similar to HC subjects, while on NFT-rich regions (high uptake VOIs: 1, 3, 3+, and 4) SUVR
90 was essentially not affected (small effect size and
p > 0.05) with exception of
K1-50% changes, which showed larger differences when compared to SUVR
90 from TAC with fitted
K1. Furthermore, SUVR
90 of HC subjects was increased with lower perfusion and decreased with higher perfusion, a similar trend when compared to perfusion changes with constant
R1. Since reliable quantification requires a method that is unaffected by
K1 and
k2 changes, a DTW approach is preferred to SUVR
90 quantification, since otherwise perfusion, plasma clearance, and BBB functionality changes can introduce a confounding factor for assessment of drug effects, follow-up studies, and disease staging, especially in diseases such as Alzheimer’s and stroke where vascular changes or changes in BBB permeability are to be expected.
Non-compliance with the scanning protocol led to an overestimation of the SUVR
90 of MCI/AD subjects with longer delays, reaching an overall average overestimation of 6.0% [1.2%:10.9%] (
p = 0.015) for 20 min of delay (Additional file
2: Table S17). This indicates that [
18F]MK-6240 uptake is not in an equilibrium state after 140 min of uptake, which was also observed for high-binding subjects on a previous study [
5] but contradicts the equilibrium found at 90 min p.i. by another group [
4]. Without reaching an equilibrium, it is not possible to compare SUVR values estimated from different time windows and then deviations from the clinical protocol will lead to a bias on patient assessment. Since published studies have used different time windows for SUVR estimation, such as 60–90 min p.i. [
6], 70–90 min p.i. [
3,
5], 90–110 min [
4,
7], and 90–120 min p.i. [
5,
24], the SUVR values from these studies cannot be compared. However, SUVR at the 90–120-min window shows promising test–retest results for NFT-rich regions of AD subjects (6 ± 2%) and could possibly be a suitable metric for cross-sectional patient assessment [
24] as long as there are no deviations from the scanning protocol. That study also found that NFT-poor regions presented higher test–retest variability (14 ± 6%). Contrary to SUVR, overall Ref Logan DVR was not impacted by scan delays (effect size < 1% regardless of diagnosis and delay length; Additional file
2: Table S11) and its use is recommended over SUVR for quantification on a time window in which tracer equilibrium has not been reached. It was observed, however, that the DVR of VOI 5 and VOI 6 were less impacted than other target regions of MCI/AD subjects (Fig.
3). That is a consequence of the lower binding found in these late-stage regions (Additional file
3: Fig. S2), showing that the DVR variability with protocol non-compliance in NFT-poor regions of patients is similar to that observed on healthy brains.
A potential limitation of SUV-based quantification of [
18F]MK-6240 is the signal spill-over from the meninges into the superior section of the cerebellum, affecting the measurements of the reference region [
3,
24]. Some previous studies addressed this issue by removing the dorsal cerebellum from the reference region [
3,
7,
24]; however, this decreases the signal-to-noise ratio due to its reduced size and may introduce bias on the quantification. Presently, we observed possible influence of such spill-over signal when assessing protocol delays on SUVR
90 quantification, as longer delays in the protocol lowered the SUVR
90 of HC subjects, showing an increased PET signal in the reference region as the delay increases.
The present work was a pilot study exploring the feasibility of the DTW protocol for [
18F]MK-6240 studies including consideration of perfusion changes, and therefore, additional validation with larger cohorts is required. One of the current main limitations is the analysis of noiseless TACs, which does not completely represent a real scenario of PET data. Nevertheless, a similar study for amyloid tracers did not find strong effects of noise in their quantification from DTW simulations [
15]. Furthermore, since the simulated TACs were extrapolated from real data of shorter acquisition time, there could be a bias on the late time point of the TAC. Another limitation is the small sample size, which makes it difficult to draw definitive conclusions on the best clinical protocol to be used for future [
18F]MK-6240 studies; nevertheless, subjects corresponding to different levels of specific binding were evaluated. Additionally, arterial sampling failed for the two AD subjects with highest tracer binding (S7 and S8) and, therefore, their simulated TACs were generated using a population-based input curve of the other MCI/AD subjects. Their bias was already expected to be higher than lower binding subjects [
4,
5], and the population-based input curve could introduce further quantitative bias. Finally, an actual dual-time-window acquisition would require alignment between the two PET acquisitions, making the analysis more laborious and potentially introduce a small bias due to image registration misalignments [
28], however, this was not an issue for previously published studies using dual-time-window protocols.