Test–retest performance of [¹⁸F]MK-6240 tau burden and relative delivery indices in cognitively normal older subjects using PET/MRI

Positron emission tomography (PET) imaging has greatly advanced our understanding of the emergence, distribution, and longitudinal accumulation of amyloid-beta (Aβ) plaques and tau-containing neurofibrillary tangles (NFT) in the brains of living individuals, in the context of both ageing and Alzheimer’s disease (AD). However, the PET detection of early deposition of hyperphosphorylated paired-helical filament tau in entorhinal and mesial temporal cortical areas poses a distinct set of challenges compared with those associated with the detection of early deposition of fibrillar Aβ plaques in the cortex (Villemagne et al., 2012). A well-known challenge of tau-PET imaging is the presence of off-target signal, which has been found to increase with age (Baker et al., 2019; Tissot et al., 2022) and exhibit sex-related differences (Scott et al., 2023; Smith et al., 2021). The off-target signal is probably best understood for [¹⁸F]flortaucipir and [¹⁸F]THK-5351. Neuropathological and PET kinetic studies have demonstrated off-target signal of [¹⁸F]flortaucipir in the basal ganglia and choroid plexus (Baker et al., 2019; Flores et al., 2022; López-González et al., 2022; Marquié et al., 2017; Tissot et al., 2022) and have revealed that [¹⁸F]THK-5351 has significant binding to the MAO-B enzyme complex (Ng et al., 2017). A second generation of tau-PET radioligands was developed in the interest of mitigating off-target signal and improving PET signal-to-noise ratios. These radioligands aimed to improve the detection of low tau levels and gain a deeper understanding of its emergence, distribution, and longitudinal accumulation. Comprehensive information about the characteristics of these radioligands has been summarized in various review articles (Lee et al., 2023; Lois et al., 2019; Saint-Aubert et al., 2017) and comparative clinical evaluations (Bischof et al., 2021; Yap et al., 2021).

[¹⁸F]MK-6240 is one of the second-generation tau-selective radioligands with improved in vivo imaging characteristics (Betthauser, Cody, et al., 2019; Guehl et al., 2019; Hostetler et al., 2016; Koole et al., 2020) that is gaining wide use in clinical studies. Numerous [¹⁸F]MK-6240 PET studies have reported strong signal-to-noise ratio in tau-bearing areas, showing utility for the detection of early and late accumulation of tau, as well as for disease staging in individuals ranging from preclinical to symptomatic AD (Bennacef et al., 2017; Betthauser, Koscik, et al., 2019; Pascoal et al., 2020a, 2020b). However, a substantial off-target binding signal is evident in the extracerebral space (Fu et al., 2023; Mertens et al., 2021; Vanderlinden et al., 2022) in about 50% of subjects (Fu et al., 2023). Neuropathological evaluations of human tissue have verified [¹⁸F]MK-6240 binding to melanin-containing cells, such as those found in the meninges within the extracerebral space (Aguero et al., 2019). The extracerebral signal has the potential to spill into the brain, including cortical and reference region areas, and thus confound quantification across cognitively normal (CN) to advanced AD subjects. To minimize this effect, several studies have selected various reference regions to normalize the PET measurement outcomes. Despite no consensus about which is optimal, some commonly used reference regions include grey matter-based regions such as the cerebellar grey matter (with or without erosion of the outer regional voxels) or the inferior cerebellar grey matter; white matter-based regions such as the cerebral white matter (with or without erosion of the outer regional voxels); and mixed regions such as the pons or whole cerebellum (Fu et al., 2023; Gogola et al., 2020; Guehl et al., 2019; Lohith et al., 2019; Pascoal et al., 2018, 2020a, 2021; Salinas et al., 2019; Shuping et al., 2023). Other investigators have proposed less common reference regions, such as the cerebellar white matter (Becker et al., 2023). Due to the spill-in contamination from the extracerebral signal to reference and target regions, the extracerebral signal has, consequently, the potential to impact the short- and long-term reproducibility of PET outcome measures.

Recently, the suitability of [¹⁸F]MK-6240 dynamic imaging to provide surrogate measurements of cerebral perfusion has been investigated. Deficits in cerebral perfusion have been reported in older subjects (Leenders et al., 1990; H. Lu et al., 2011) and in subjects with AD (Austin et al., 2011; Johnson & Albert, 2000) or other tauopathies (Hu et al., 2010), and are important for understanding the pathogenesis of these diseases and ageing. Although cerebral perfusion can be measured using MRI-based methods such as arterial spin labelling (ASL) or with the gold standard [¹⁵O]water PET, these require a PET/MRI scanner or a separate scanning session, which leads to added subject burden and increased logistical difficulties. This has motivated investigations of whether the early phase of amyloid or tau PET imaging can provide surrogate measurements of cerebral perfusion (Bilgel et al., 2019; Chen et al., 2015; Ottoy et al., 2019; Rodriguez-Vieitez et al., 2016; Visser et al., 2020). Two recent studies support the use of the early phase of [¹⁸F]MK-6240 dynamic data to derive robust estimates of relative delivery (R₁) as a quantitative index of relative cerebral perfusion (Fu et al., 2022; Guehl et al., 2023), potentially allowing dual-imaging assessments of tau and cerebral perfusion. However, [¹⁸F]MK-6240 R₁ variability has not yet been reported.

Test–retest (T-RT) variability informs about how different the outcome measurements are when data are obtained repeatedly on the same subject and under similar conditions (i.e., within-subject variability). Understanding the T-RT variability is crucial for the accurate interpretation of quantitative results. This is particularly relevant for longitudinal studies that involve multiple scans over time, such as the investigation of the natural time course of tau accumulation, or the effects of therapy in drug intervention studies where small changes in PET signal need to be detected. Given its relevance, T-RT is one of the properties often evaluated for novel PET tracers (Bullich et al., 2020; Devous et al., 2018; Finnema et al., 2017; Salinas et al., 2019; Timmers et al., 2019; Vanderlinden et al., 2022).

Two recent studies reported [¹⁸F]MK-6240 T-RT results for subjects with different characteristics. First, in an AD subject-dominant sample consisting of 12 AD (65 ± 1 years of age) and 3 CN (55 ± 7 years of age) subjects, Salinas et al. (2019) determined T-RT from dynamic [¹⁸F]MK-6240 PET imaging performed within 21 days. Across all subjects, the average T-RT percentage differences in tau-bearing regions were approximately 21%, 14%, and 6% for the total distribution volume (V_T), binding potential (BP_ND), and late-frame standardized-uptake-value ratio (SUVR_90–120) PET outcomes, respectively. However, the T-RT variability of the extracerebral signal was not reported. Second, in a sample consisting of 10 CN subjects (56 ± 12 years of age), Vanderlinden et al. (2022) reported a long-term 6-month [¹⁸F]MK-6240 late-frame SUVR_90–120 T-RT of 2.4 ± 2.8% in whole-brain grey matter. The authors also reported that, at group level, the extracerebral uptake was not significantly different at the 6-month follow-up compared with baseline (extracerebral SUVR_90–120 T-RT: 4.4 ± 20%), nor at a 2-year follow-up for a group of 10 amnestic mild cognitive impairment subjects (extracerebral SUVR_90–120 T-RT: 7.9 ± 19%). The extracerebral signal was found to correlate with age (r’s = −0.48; p < 0.0001) and to be higher in women.

Although the two previous studies included CN subjects, the average age of those participants was 56 years. Estimating the T-RT variability in older CN subjects (>65 years of age) who may be at higher risk of AD may, however, be critical for detecting the emergence of early tau deposition. In addition, these studies did not evaluate the T-RT characteristics of the reference regions, an important step to understanding the variability in normalized target region outcomes. Furthermore, although recent studies have shown that the relative radioligand delivery index R₁ estimated from dynamic [¹⁸F]MK-6240 acquisition may provide reliable estimates of relative perfusion (Fu et al., 2022; Guehl et al., 2023), the evaluation of the T-RT characteristics of [¹⁸F]MK-6240 R₁ is still needed in order to provide further support for dual-biomarker imaging to quantify both tau deposition and relative cerebral perfusion across the AD spectrum. Finally, to reduce the spill-in contamination from the extracerebral signal to the target and reference regions, outer voxel erosion and partial volume correction (PVC) techniques have been used (Fu et al., 2023; Mertens et al., 2021). It is known that the processing methods and applied corrections may affect the quantification of the PET data and, thus, may impact the T-RT variability of PET outcomes. However, the impact of PVC and reference region erosion on the T-RT characteristics of the [¹⁸F]MK-6240 uptake in the target and reference regions is still unclear.

Here, we further evaluate the short-term T-RT characteristics of dynamic [¹⁸F]MK-6240 PET on a sample largely consisting of older CN individuals (oCN, median age 66 years, interquartile range (IQR) [65,71]) studied twice within approximately 3 weeks. First, we evaluate the T-RT characteristics in eight potential reference regions commonly used in [¹⁸F]MK-6240 clinical studies (including grey matter-based, white matter-based, and mixed regions), and in the extracerebral signal. Second, in 11 target regions (including areas of early and late tau accumulation in AD), we assess the impact of using different reference regions on the T-RT of late-frame SUVR. Third, we investigate the differences in regional T-RT values obtained from dynamic imaging outcomes (i.e., distribution volume ratio, DVR) and from late-frame SUVR in target regions. Fourth, we extend the current knowledge on [¹⁸F]MK-6240 dynamic imaging with the investigation of T-RT for R₁. Lastly, we evaluate the impact of applying an iterative voxel-based PVC on the T-RT of all analyzed outcome measurements.

Eight participants (7 oCN, median age 66 years, IQR [65,71]; 1 AD, 54 years) underwent [¹⁸F]MK-6240 baseline PET/MR imaging (Test) and a follow-up PET/MR scan (Retest) within approximately 3 weeks (Test–Retest interval IQR: 2 to 4 weeks). Nine additional subjects underwent baseline PET/MR only: 4 young controls (yCN, median age 26 years, IQR [25,29]), 3 oCN, and 2 AD participants. These additional subjects were included in baseline analyses only, to facilitate comparisons with previously reported results and for completeness.

Subject characteristics are summarized in Table 1. All participants in this study self-identified as non-Hispanic White, except three oCN participants (one African American, one Asian, and one mixed White/Asian). APOE genotype and amyloid status were not available for most subjects. The study was approved by the local Institutional Review Board. The AD participants were recruited from the Massachusetts General Hospital (MGH) Neurology Units and Frontotemporal Disorders Unit after receiving a diagnosis of probable AD of mild severity. The AD diagnosis was performed by an experienced neurologist at MGH specialized in dementia and was based on memory complaints and functional impairment, as determined by a clinical interview with the subject and an informant. The severity of these symptoms met the criteria for probable AD of mild severity, as defined by the National Institute of Neurological and Communicative Disorders and Stroke/Alzheimer’s Disease and Related Disorders Association (McKhann et al., 2011). CN participants were recruited from the local community. All subjects underwent neurological evaluations using the Mini-Mental State Examination (MMSE). For the CN participants, assessment of normal cognition was obtained through MMSE (MMSE > 25 for oCN and MMSE > 27 for yCN). Prior to enrollment, all CN subjects provided written informed consent to participate in the study, and all AD subjects assented to participate in the study with their study partner’s consent.

All subjects underwent dynamic Test PET following the intravenous administration of an injection of 185 ± 15 MBq of [¹⁸F]MK-6240 (molar activity: 177 ± 70 GBq/μmol at the end of the synthesis). Scans were performed on a whole-body Biograph-mMR scanner (Siemens Healthineers), which allows the simultaneous acquisition of PET and MR images (Delso et al., 2011). PET data were acquired for 120 minutes divided into 2 segments (0–65 min and 80–120 min) with a 15-minute break during which the subjects were allowed to leave the scanner. The subset of subjects who underwent Retest [¹⁸F]MK-6240 PET/MRI followed the same imaging protocol (injected dose 186 ± 13 MBq). Simultaneous MR imaging was performed during each PET scan. A T1-weighted 3-dimensional magnetization-prepared rapid gradient-echo (MPRAGE) image was acquired at the beginning of each segment of the PET scan (i.e., before and after the break) for the purpose of anatomical localization, exclusion of incidental pathology, and generation of attenuation correction maps.

MPRAGE images were corrected for MR intensity inhomogeneity using N4ITK17 (Tustison et al., 2010) and were denoised using value thresholding and masking using Slicer3d (https://www.slicer.org) (Fedorov et al., 2012). For each subject, a single MPRAGE image was selected for automated volumetric segmentation of cortical and subcortical brain structures, based on the least evidence of subject motion on visual inspection (if visually similar, the baseline MPRAGE was selected). Using FreeSurfer (version 6, http://surfer.nmr.mgh.harvard.edu/), the MPRAGE image was segmented into tissue types and used to classify anatomical brain volumes (Fischl et al., 2002, 2004). Additional processing generated the extracerebral segmentations by dilating the pial surface by 5 mm outward perpendicularly to the cortical surface, as described previously (Fu et al., 2023). The inferior cerebellar grey matter region was obtained using the SUIT cerebellar template (Diedrichsen, 2006), reverse normalized to the subject’s MPRAGE space (Baker et al., 2017), and subtracted from the FreeSurfer-generated cerebellar grey matter.

The MPRAGE image was registered to the Test and Retest PET images, and the corresponding transformation matrix was applied to the regional segmentations using FSL (FMRIB Software Library, https://fsl.fmrib.ox.ac.uk/fsl/) (Woolrich et al., 2009). The segmented regions of interest (ROIs) were used for regional sampling of the PET images and generation of regional time-activity curves (TACs). Identical ROIs were applied to the Test and Retest scans for each subject.

Dynamic PET data were divided into frames (29 frames for the first segment: 6 x 10 s, 6 x 20 s, 2 x 30 s, 2 x 60 s, 2 x 120 s, and 11 x 300 s; and 8 frames for the second segment: 8 x 300 s) and were reconstructed using an ordered-subset expectation maximization algorithm (3 iterations, 21 subsets, 344 × 344 × 127 image matrix, with 2.1 mm in-plane pixel size and 2.0 mm slice thickness; 4 mm Gaussian filter). PET images were corrected for dead time, decay, scatter, and attenuation. For each PET segment, attenuation maps were estimated from the respective MPRAGE image through a combination of intensity- and prior-based tissue segmentation and atlas registration, as described previously (Izquierdo-Garcia et al., 2014). Motion correction was applied through frame-by-frame realignment of the reconstructed data using FSL (Woolrich et al., 2009). Each PET segment was individually motion corrected, and subsequently, the second segment was registered and concatenated to the first.

PVC was performed using the iterative Yang method (Erlandsson et al., 2012; Thomas et al., 2016; Y. Lu et al., 2021) (10 iterations), an adaptation of the Yang method (Yang et al., 1996) that applies a correction factor iteratively for each voxel. Briefly, a regional mean image is created by averaging the PET activity values in each FreeSurfer segmentation for each PET frame. A partial volume correction map is generated iteratively as the ratio of the non-smoothed regional mean image and the regional mean image smoothed by a 5 mm Gaussian kernel (value selected based on our scanner performance; Delso et al., 2011). The final correction map is applied to the dynamic PET data at the voxel level.

Standardized uptake-value (SUV_90–110, g/mL) images were calculated for data acquired 90−110 min post-tracer injection by averaging the corresponding four 5-min reconstructed frames and normalizing by body weight and injected dose.

As described above, ROIs were defined on the individual MPRAGE images and, subsequently, used to sample PET images and obtain regional TACs and SUV_90–110 values. For each subject, left and right hemisphere ROIs were averaged and analyzed together. Twenty a priori selected regions were analyzed:

1. Reference regions. Eight regions (previously used for [¹⁸F]MK-6240 and other tau radioligands such as [¹⁸F]Flortaucipir) were examined: cerebellar grey matter (CerGM), cerebellar grey matter with a 3-mm erosion of the outer regional mask voxels (CerGM_3mm), inferior cerebellum (Inferior CerGM), cerebellar white matter (CerWM), cerebral white matter (WM), cerebral white matter with 4-mm erosion (WM_4mm), whole cerebellum (WholeCer), and pons.

2. Target regions. Eleven regions known to have pathological tau accumulation in various stages of AD were examined: entorhinal cortex, amygdala, hippocampus, fusiform, inferior temporal cortex, precuneus, insula, precentral gyrus, rostral middle frontal, lateral occipital cortex, and a meta-temporal cortical composite (comprising the entorhinal, parahippocampus, amygdala, fusiform, inferior, and middle temporal cortex).

3. Extracerebral region.

The following PET outcome measures were evaluated: (1) For reference regions: SUV_90–110; (2) For target regions: SUV_90–110, SUV-ratios (SUVR_90–110 = SUV_TARGET/SUV_REFERENCE), distribution volume ratio (DVR) computed using Multilinear Reference-Tissue Model 2 (Ichise et al., 2003) (MRTM2; t* = 30 min, k2’ = 0.04; Betthauser, Cody, et al., 2019), and relative delivery (R₁ = K_{1_TARGET}/K_{1_REFERENCE}) estimated using Simplified Reference-Tissue Model (Lammertsma & Hume, 1996) (SRTM); (3) For extracerebral region: SUV_90–110 and SUVR_90–110. All outcome measures were computed for both non-PVC and PVC data.

The variability of the outcome measures of interest was evaluated using absolute percentage T-RT differences, calculated as $T−RT(%)=200*|Testoutcome−Retestoutcome|Testoutcome+Retestoutcome$⁠, and Bland–Altman plots in subjects with Retest PET (n = 8: 7 oCN, 1 AD). To further evaluate the within-subject variability, intraclass correlation coefficients (ICC, two-way model) were calculated for each ROI in the T-RT oCN group only (n = 7, the AD subject was excluded to avoid an increase in the between-subjects variability explained by disease). ICC values less than 0.5 were considered poor reliability, between 0.5 and 0.75 moderate, between 0.75 and 0.9 good, and greater than 0.90 were considered excellent reliability (Koo & Li, 2016).

Continuous demographic variables were reported as medians with interquartile ranges (IQR) to address the small sample size. However, group-level PET outcome measures were expressed as mean ± standard deviation (SD) to facilitate comparisons with existing literature. Baseline reference region outcome measures were compared across regions using the Kruskal–Wallis test. Differences between Test and Retest data in each region were evaluated using a Wilcoxon signed-rank test. Associations between regional SUVR_90–110 and DVR values were assessed using Spearman’s correlation. Significance level was assessed at p = 0.05. No multiple comparison corrections were applied.

Statistical analyses were performed using Python version 3.9 (https://www.python.org/), including the SciPy (Virtanen et al., 2020), Seaborn (Waskom, 2021), Matplotlib (Hunter, 2007), Pandas (McKinney, 2010), and Pingouin (Vallat, 2018) packages. PET image analyses were performed using Miakat (Gunn et al., 2016).

The pattern of brain [¹⁸F]MK-6240 uptake distribution was consistent with previous studies (Betthauser, Cody, et al., 2019; Fu et al., 2023; Guehl et al., 2019; Pascoal et al., 2018) and with the expected localization of tau across the AD spectrum (Braak & Braak, 1999; Kreisl et al., 2022; Pascoal et al., 2020a), and was similar in the Test and Retest scans (Supplementary Fig. 1).

All subjects exhibited some degree of meningeal or extracerebral signal, with seven of them presenting visually high extracerebral signal. We observed high inter-subject variability in extracerebral uptake, both in location and magnitude. In the meningeal space, CN participants showed varying signal levels, from high (e.g., Subjects 6 and 7; Supplementary Fig. 1) to relatively low (e.g., Subjects 2 and 3; Supplementary Fig. 1). The meningeal signal distribution was non-uniform. All participants displayed elevated signal in the meningeal space superior to the cerebellum. In all AD participants, the superior portion of the cerebellum showed elevated signal due to spill-in from the high tau binding nearby cortical regions. For the CN participants, the extracerebral signal was generally more pronounced inferior to the cerebellum compared with superior to the cerebellum. Extracerebral uptake was also observed in the sinuses, with high inter-subject variability. Two participants (Subjects 1 and 7; Supplementary Fig. 1) exhibited notable differences between the Test and Retest scans in this region.

Figure 1A shows an example oCN subject with high extracerebral signal in the PET SUVR_90–110 image (CerGM as the reference region) and the extracerebral segmentation. Illustrative examples of Test and Retest [¹⁸F]MK-6240 SUVR_90–110 PET images are shown for one oCN and one AD subject in Figure 1B and C, respectively. The oCN subject shows low cerebral uptake and a mild but noticeable extracerebral signal. The AD subject shows high uptake in numerous cortical areas and low extracerebral signal.

Table 2 lists the SUV_90–110 values for the evaluated reference regions. The mean regional SUV_90–110 values ranged from 0.35 (pons) to 0.62 g/mL (inferior CerGM) in the CN groups, and from 0.38 (pons) to 1.11 g/mL (WM) in the AD group. SUV_90–110 values were similar across diagnostic groups for all evaluated reference regions except for the cerebral WM and WM_4mm, which were twofold higher in the AD group than in the CN groups. The pons showed the lowest SUV_90–110 among all evaluated reference regions in both the CN and AD groups, although there were no significant differences between any of these reference regions. Figure 2A shows violin plots visualizing baseline SUV_90–110 for illustrative reference regions, including the CerGM, WholeCer, WM, and pons.

For the evaluated reference regions, the mean regional SUV_90–110 T-RT variability ranged from 6 to 11% in the oCN group, and from 8 to 13% in the AD group (Table 2). Figure 2B shows violin plots representing SUV_90–110 T-RT for some illustrative reference regions. The ICC values were excellent for all reference regions (≥0.89; Supplementary Fig. 2). In oCN, among the evaluated reference regions, the pons showed the lowest SUV_90–110 T-RT variability (6.5%) and highest ICC (0.97), while the CerWM showed the highest SUV_90–110 T-RT variability (10.9%) and lowest ICC (0.89). The Wilcoxon rank-sum test revealed no statistical differences between the Test and Retest SUV_90–110 for any of the evaluated regions. Figure 2C shows individual Test and Retest SUV_90–110 values in those selected regions.

The iterative Yang PVC SUV_90–110 and corresponding T-RT values are provided in Supplementary Table 1. PVC SUV_90–110 values ranged from 0.31 (pons) to 0.59 g/mL (CerGM and inferior CerGM) in the CN groups, and from 0.37 (pons) to 0.92 g/mL (WM) in the AD group. The application of PVC had a minor impact on the range of SUV_90–110 for all evaluated reference regions, except for the cerebral WM in the AD group, where PVC SUV_90–110 decreased by 17% compared with the non-PVC SUV_90–110. The corresponding SUV_90–110 T-RT and ICC values were similar between PVC and non-PVC SUV_90–110. The impact of applying PVC to SUV_90–110 in the selected reference regions and on the corresponding SUV_90–110 T-RT is illustrated in Figure 2A and B, respectively, and the Test and Retest PVC SUV_90–110 values for each oCN subject are shown in Figure 2D.

The impact of normalizing by different reference regions on SUVR_90–110 and SUVR_90–110 T-RT is summarized in Supplementary Table 2. In the CN groups, normalization by CerGM and Inferior CerGM resulted in the lowest SUVR_90–110 values, ranging from 0.58 (amygdala) to 1.14 (lateral occipital), while pons resulted in the highest, ranging from 1.08 (amygdala) to 2.26 (lateral occipital). In the AD group, normalization by WM and WM_4mm resulted in lower SUVR_90–110 values than other reference regions, due to spill-in signal from grey matter to the white matter-based reference regions. Excluding WM and WM_4mm, in the AD group, CerGM and pons resulted in the lowest, respectively, highest SUVR_90–110 values, consistent with the CN groups. In the CN groups, target region SUVR_90–110 T-RT variability was similar when using for normalization CerGM-based reference regions (CerGM: 4–9%, CerGM_3mm: 5–10%, Inferior CerGM: 4–9%,), WholeCer (4–9%) or WM (3–11%), and slightly higher in WM_4mm (2–14%), pons (3–14%), and CerWM (6–15%). In addition, regardless of the reference region used, the rank orders of the target regions remained consistent in the AD subjects.

For the remaining analyses, we selected CerGM as the reference region, as it is one of the most used reference regions in the current literature (Bourgeat et al., 2023; Krishnadas et al., 2023; Vanderlinden et al., 2022), and all evaluated reference regions resulted in similar T-RT SUVR_90–110 variability.

Table 3 summarizes the SUVR_90–110, DVR, and R₁ values obtained using CerGM as the reference region for the examined target regions. In all evaluated regions, mean regional SUVR_90–110 and corresponding regional DVR were similar and showed Spearman’s correlation ranging from 0.84 (insula) to 0.99 (entorhinal). In the CN groups, mean regional SUVR_90–110 and DVR ranged from 0.59 (amygdala) to 1.20 (lateral occipital) across target regions. In the AD group, the regional rank orders for SUVR_90–110 and DVR were identical: Precuneus (SUVR_90–110 = 2.75) > Inferior Temporal > Rostral Middle Frontal > Fusiform > Lateral Occipital > Entorhinal > Precentral > Insula, Amygdala > Hippocampus (SUVR_90–110 = 1.04). Figure 3A shows violin plots representing baseline SUVR_90–110 for illustrative target regions (entorhinal, amygdala, inferior temporal cortex). Individual Test and Retest SUVR_90–110 values for those same regions are shown in Figure 3C. In all evaluated target regions, R₁ values were generally lower for AD (0.66 (entorhinal) - 0.93 (insula)) than for oCN subjects (0.69 (entorhinal) - 1.39 (insula)). The regional rank orders of R₁ values were consistent between the CN and AD groups: Insula > Precentral, Precuneus, Rostral Middle Frontal, Fusiform > Amygdala, Hippocampus > Inferior Temporal, Lateral Occipital > Entorhinal. The meta-temporal composite region showed similar tau uptake and R₁ values as the inferior temporal region.

Table 3 summarizes SUVR_90–110, DVR, and R₁ T-RT values obtained using CerGM as the reference region for the examined target regions. In the oCN group, target region SUVR_90–110 T-RT ranged from 3.9% (entorhinal) to 8.7% (lateral occipital); DVR T-RT ranged from 2.9% (precuneus) to 10.3% (lateral occipital); and R₁ T-RT ranged from 2.8% (precuneus) to 13.8% (lateral occipital). In the AD subject, target region SUVR_90–110 T-RT ranged from 1.0% (insula) to 8.8% (hippocampus); DVR T-RT ranged from 2.3% (amygdala) to 9.2% (hippocampus); and R₁ T-RT ranged from 1.5% (entorhinal) to 16.1% (lateral occipital). The meta-temporal composite region showed T-RT for tau uptake and R₁ values similar (slightly lower) to the inferior temporal region. Violin plots representing SUVR_90–110 T-RT for illustrative target regions (entorhinal, amygdala, inferior temporal cortex) are shown in Figure 3B. The ICC values of SUVR_90–110 and DVR were moderate to excellent in all target regions (0.62 (fusiform)— 0.97 (entorhinal)), and ICC of R₁ was moderate in the lateral occipital (0.52) and good to excellent in all other regions, as shown in Figure 4. The Wilcoxon matched-pairs signed-rank test revealed no statistical differences between the Test and Retest scans for SUVR_90–110, DVR, or R₁ in any of the evaluated target regions.

Supplementary Table 3 lists PVC SUVR_90–110, PVC DVR, and PVC R₁, and the corresponding T-RT values for all examined target regions. The application of PVC resulted in a slight change in average regional SUVR_90–110 and DVR (range: 0.29 (amygdala)–1.33 (lateral occipital)) in the CN groups. In the AD group, PVC resulted in an increase in SUVR_90–110 and DVR in some cortical target regions (e.g.: fusiform and inferior temporal ~50%, precuneus ~70%), and the rank order of target region uptake was consistent with that obtained with non-PVC. PVC R₁ values ranged from 0.70 (entorhinal) to 1.70 (insula) and were, on average, ~20% higher than the non-PVC R₁ values in both the CN and AD groups.

The T-RT of PVC SUVR_90–110 and DVR increased by over twofold compared with the respective non-PVC measures. In the oCN group, T-RT of PVC SUVR_90–110 ranged from 9.9% (entorhinal) to 26.2% (precentral); T-RT of PVC DVR ranged from 6.3% (fusiform) to 21.8% (lateral occipital); and T-RT of PVC R₁ ranged from 3.5% (precuneus) to 19.9% (lateral occipital). For the AD subject, T-RT of PVC SUVR_90–110 ranged from 0 (insula) to 21.8% (hippocampus), T-RT of PVC DVR ranged from 1.5% (precentral) to 21.3% (hippocampus), and the T-RT of PVC R₁ ranged from 0.6% (entorhinal) to 22.6% (rostral middle frontal). SUVR_90–110, DVR, and R₁ ICC values decreased notably with PVC in some target regions (e.g., SUVR_90–110 Rostral Middle Frontal: 0.86 (non-PVC) to 0.21 (PVC), Fig. 4). The effect of PVC on SUVR_90–110 in selected target regions and on the corresponding SUVR_90–110 T-RT is illustrated in Figure 3A and B, respectively, and individual Test and Retest PVC SUVR_90–110 values are shown in Figure 3D. Figure 5 shows Bland–Altman plots summarizing values across multiple target regions for the examined outcome measures without and with the application of PVC (see Bland–Altman bias and limit of agreement estimates by anatomical region in Supplementary Table 4). Differences between Test and Retest in SUVR_90–110, DVR, and R₁ values are similar and consistently small in all evaluated regions (generally < 0.1). PVC increases the differences between Test and Retest in all outcome metrics, and thus the measurement variability.

Tables 2 and 3 summarize the SUV_90–110, SUVR_90–110, and the corresponding T-RT values for the extracerebral region. Extracerebral SUV_90–110 values in the CN groups (yCN: 0.80; oCN: 0.77 g/mL) were higher than reference and target region SUV_90–110 values (0.34–0.66 g/mL); AD group values (0.91 g/mL) were in the range of target region SUV_90–110 values (0.94–1.77 g/mL). Extracerebral SUVR_90–110 values were similar in the CN and AD groups (1.33–1.39) and higher than the target region SUVR_90–110 values in the CN groups (0.64–1.12). In the oCN group, the SUVR_90–110 T-RT was twofold higher in the extracerebral region (~12%) than the average of examined target regions (~6%), while in the AD subject, the SUVR_90–110 T-RT was similar to the average of examined target regions (~4%). The extracerebral Test and Retest SUVR_90–110 displayed high inter- and intra-subject variability (Fig. 3C). The extracerebral T-RT variability increased slightly with the application of PVC (~15% and 6% in the oCN and AD groups, respectively). ICC values were poor for extracerebral SUV_90–110 (~0.4) but good for SUVR_90–110 (> 0.8), as summarized in Figure 4 and Supplementary Figure 2.

In this study, we investigated the short-term (~ 3 weeks) T-RT characteristics of dynamic [¹⁸F]MK-6240 outcomes in a group largely consisting of cognitively normal older individuals using simultaneous PET/MRI. We reported T-RT results for uptake in potential reference regions, for extracerebral off-target signal, as well as for estimates of tau burden and relative delivery indices in tau-bearing target regions.

For the evaluated reference regions, we found acceptable T-RT variability in late-frame SUV_90–110 measures, ranging from 6 to 13% across all subjects. Currently, there is no consensus on the optimal reference region for [¹⁸F]MK-6240, and various studies have chosen different regions for normalization (Fu et al., 2023; Gogola et al., 2020; Guehl et al., 2019; Lohith et al., 2019; Pascoal et al., 2018, 2020b, 2021; Salinas et al., 2019; Shuping et al., 2023). In a recent study, Fu et al. (2023) evaluated longitudinal changes in reference region SUV_90–110, as well as the impact of the reference region on target region outcomes. The authors suggested that, for cross-sectional studies, eroded CerGM may be preferred for differentiation between CN and AD groups, while eroded WM or pons may be preferred for detecting longitudinal [¹⁸F]MK-6240 changes. Although the SUV_90–110 T-RT could influence the selection of the optimal reference region, comparative evaluations have not been reported. For oCN, our results showed the lowest SUV_90–110 T-RT variability in pons and CerGM, but no significant differences were observed between the evaluated regions. Since SUV_90–110 T-RT did not strongly favor a specific region, it should be used as complementary but not as a main factor in reference region selection. An important factor to consider is whether the chosen reference region is prone to spill-in contamination signal from high-uptake cortical regions and extracerebral regions.

In this work, we also assessed how the potential reference regions impact SUVR_90–110 T-RT in target regions. As expected, various reference regions resulted in quantitative differences in SUVR_90–110 outcomes (lowest SUVR_90–110 values with CerGM or Inferior CerGM; highest with pons). Note that, as previously reported (Fu et al., 2023), WM and WM_4mm are prone to spill-in signal from the cortex in AD participants with high tau uptake, resulting in higher SUV_90–110 in the AD than in CN subjects. Therefore, WM and WM_4mm resulted in lower SUVR_90–110 in the AD than in the CN group, appearing not to be good reference regions in this study. For the set of evaluated reference regions, target region SUVR_90–110 T-RT ranged from 2 to 15%, with no significant differences due to the choice of reference regions. For the remaining analysis, we selected CerGM as the reference region, as it is one of the most used in the current literature (Bourgeat et al., 2023; Krishnadas et al., 2023; Vanderlinden et al., 2022).

For the evaluated target regions, we found low SUVR_90–110 T-RT variability, with regional values ranging between 1 and 9% across all subjects, and an average SUVR_90–110 T-RT across all target regions and subjects of 5%. These results are consistent with two previous reports, despite differences in the study design and methodology. Using a standalone PET system, Salinas et al. (2019) determined SUVR using 90–120 min of [¹⁸F]MK-6240 data (SUVR_90–120) and the inferior CerGM as reference region in a group consisting of 12 AD (65 ± 1 years) and 3 CN (55 ± 7 years) subjects. With a similar T-RT interval (21 days), they reported an average 6% target region T-RT across all subjects (ranging between 2 and 9% in tau-rich and 2 and 21% in tau-poor regions). Using a PET/MRI system, Vanderlinden et al. (2022) reported long-term 6-month [¹⁸F]MK-6240 SUVR_90–120 T-RT in a sample consisting of 10 CN (56 ± 12 years) using the CerGM as reference region. For mean whole grey matter, SUVR_90–120 T-RT was 2%, and regional SUVR_90–120 T-RT ranged from 3 to 8%.

Our SUVR_90–110 T-RT results complement previous reports in two main aspects: (1) The focus of our study was on older CN subjects (median age 69 years, IQR [66,71]), not well represented in previous studies that included younger CN subjects (average age ~56 years). Our results thus fill an important gap in understanding the T-RT characteristics in oCN, a population especially important for longitudinal observational studies involving largely these subjects (such as the Harvard Aging Brain Study (Dagley et al., 2017)). Second, the previous studies focused on the evaluation of T-RT in larger (or aggregated) target regions but did not include important smaller regions, such as the entorhinal cortex or amygdala, that are known to be pathologically affected in the early stages of AD. While using large ROIs has advantages in terms of lower sensitivity to image noise and contamination due to partial volume effects, the superiority of global or aggregate measures versus regional measurements has not been rigorously established for tau imaging. The anatomic distribution of [¹⁸F]MK-6240 retention, particularly in the early stages of AD-related tauopathy, is variable within vulnerable areas and may be masked when employing large ROIs. Our data showed excellent SUVR_90–110 T-RT characteristics in important regions (e.g., 4% in the entorhinal cortex), suggesting [¹⁸F]MK-6240 is suitable for detecting early tau deposition as well as for measuring longitudinal changes over time.

We also investigated differences between T-RT variability obtained from DVR with dynamic imaging and with late frame SUVR_90–110 in target regions. [¹⁸F]MK-6240 SUVR has been shown to lack a plateau in high tau binding areas (Guehl et al., 2019; Salinas et al., 2019), even 150 min after injection, which could introduce bias unless a precise acquisition time across longitudinal assessments is maintained. This potential pitfall has also been described for other tau tracers (Barret et al., 2017; Bullich et al., 2020; Kuwabara et al., 2018). Although our sample consists primarily of oCN with low tau accumulation, investigating differences between DVR and SUVR T-RT may provide valuable information for the design of prospective studies. Salinas et al. (2019) analyzed dynamic data and reported that the T-RT variability of V_T and BP_ND outcome measures was higher than the T-RT variability of SUVR_90–120 across subjects. For the subset of CN subjects (n = 3), they reported an average T-RT of 22% for V_T and 5% for SUVR_90–120 in tau-rich regions, whereas the T-RT of BP_ND exceeded 100% (likely due to the small number of CN subjects in the study). Contrary to their results, we found low DVR variability, with target region DVR T-RT between 1 and 10% and similar to SUVR_90–110 T-RT (1–9%) across all subjects. When considering oCN only, DVR T-RT variability was marginally but consistently lower than SUVR_90–110 T-RT variability in all examined target regions except the entorhinal cortex and lateral occipital. Given the potential bias of SUVR, good DVR reproducibility could favor the use of this outcome measure in longitudinal studies aimed at detecting small changes in [¹⁸F]MK-6240 uptake.

In the present work, we also evaluated R₁ T-RT using dynamic data. A recent study supports using the early phase of [¹⁸F]MK-6240 dynamic data to derive robust estimates of relative delivery R₁ as a quantitative index of relative cerebral perfusion (Guehl et al., 2023), but [¹⁸F]MK-6240 R₁ T-RT was still lacking. Our study demonstrated acceptable short-term R₁ T-RT variability in all evaluated target regions, ranging from 1 to 16% across all subjects, further supporting the viability of [¹⁸F]MK-6240 R₁ to robustly evaluate longitudinal changes in perfusion.

We investigated the effects of an iterative voxel-based PVC on the evaluated outcome measurements. For oCN, PVC had a small effect on SUV_90–110, SUVR_90–110, and DVR, as these subjects present an overall low and uniform uptake. For AD, PVC resulted in the expected patterns, with an increase of signal in high tau-bearing regions (e.g., inferior temporal cortex), a decrease of signal in areas prone to contamination (e.g., white matter), and no effect in regions not prone to contamination and with minimal tau accumulation (e.g., pons). The application of PVC increased T-RT variability for SUVR_90–110, DVR, and R₁ in target regions, up to over twofold compared with the T-RT of non-PVC measures. These results are consistent with the increase in T-RT variability in SUVR due to PVC observed by Vanderlinden et al. (2022). Although they report smaller changes than in our study, this may be explained by the use of larger regions in their analyses, which are less affected by spill-in and spill-out effects. Due to the resulting increase in inter-subject and T-RT variability in reference region and target region uptake, caution should be taken when applying PVC to [¹⁸F]MK-6240 PET analyses.

A large meta-temporal composite region was evaluated in this study. It showed similar magnitudes of tau uptake (SUVR_90–110 and DVR) and R₁ values as the inferior temporal region, with slightly lower inter-subject variability. These observations provide additional support for the feasibility of using a meta-temporal composite as a reliable target to measure tau burden and relative perfusion in clinical studies.

Contrary to common assumptions, our analysis revealed no clear relationship between target region T-RT and ROI size. This is well illustrated by the entorhinal cortex, which, despite being the smallest region in this study (~650 voxels in the PET space), exhibited one of the lowest T-RT values (3.9% TRT for SUVR_90–110, 5.1% TRT for DVR, and 4.8% TRT for R₁ in oCN participants). This suggests that factors such as the anatomical location of the ROI (e.g., proximity to ventricles, extracerebral off-target signal, and high-tau ROIs) may have a larger impact on T-RT for [¹⁸F]MK-6240 outcomes than ROI size alone, which has important implications for optimizing ROI selection in clinical research.

A key part of this study is the evaluation of T-RT of [¹⁸F]MK-6240 uptake in the extracerebral space, which has been shown to present high signal in about 50% of subjects (Fu et al., 2023). Our findings reveal notable heterogeneity in extracerebral uptake, consistent with recent literature (Fu et al., 2023; McVea et al., 2024). We observed high inter-subject variability in both the location and magnitude of the extracerebral signal, particularly in the meningeal and sinus spaces. Some CN participants exhibited high meningeal signal, while others showed relatively low signal. Even in subjects with high meningeal signal, the distribution was non-uniform. In CN participants, the meningeal uptake was generally more pronounced inferior to the cerebellum compared with the superior portion.

We also found considerable intra-subject T-RT variability in the extracerebral signal. In the oCN group, T-RT variability for SUV_90–110 and SUVR_90–110 values was 10.9 ± 10.1% and 12.1 ± 14.6%, respectively, with 57% of participants showing greater than 10% T-RT variability in extracerebral SUV_90–110. This aligns with recent results from McVea et al. (2024), who reported 44% of subjects showing >10% longitudinal variability in meningeal signal over an average 2.4-year follow-up.

Overall, our results indicate that extracerebral Test and Retest SUVR_90–110 present higher inter-subject and intra-subject variability than any of the evaluated target and reference regions, which could affect longitudinal target region quantification if not taken into account. Although PVC is expected to reduce the effect of the extracerebral signal contamination in target and reference regions, it is likely that some spill-in effects will remain after its application. At the group level, the SUVR_90–110 T-RT variability was slightly higher in the extracerebral region (8%) than in the average of target regions (5%) and, when analyzing the oCN group separately, the SUVR_90–110 T-RT variability was two times higher in the extracerebral region (~12%) than in the average of examined target regions (~6%). In oCN, PVC increased the extracerebral SUVR_90–110 T-RT to 15%. In a long-term T-RT evaluation of the extracerebral signal uptake, Vanderlinden et al. (2022) reported no significant mean differences over a 6-month period. However, their data also display high individual variability over time (15–35% for 4 of the 10 CN subjects). Due to the potential of extracerebral signal to hamper accurate quantification, it is important to examine the individual variability in addition to the group average. Our observations emphasize the importance of considering and potentially correcting for the effects of extracerebral signal on [¹⁸F]MK-6240 PET analyses, especially in oCN.

One of the strengths of our study is the use of dynamic [¹⁸F]MK-6240 data. This allowed us to evaluate the T-RT characteristics of DVR and R₁ and compare them with those of late-frame SUVR imaging. However, our study has some limitations. The main limitation is the small sample size, and larger sample sizes may be needed for robust validation of these results. In addition, our sample may not be representative of older adults at high risk for AD, and future studies should include a larger proportion of amyloid-positive individuals. Furthermore, the lack of racial and ethnic diversity in the sample may limit the generalizability of the findings to other populations. In addition, we did not have arterial data to measure the input functions, and, therefore, our analyses were constrained to the evaluation of normalized outcome measures. Another limitation of our study is that we assumed a uniform distribution of uptake inside the extracerebral mask, and the mask did not include the signal in the sinus space.

In low-signal oCN subjects, the T-RT variability of [¹⁸F]MK-6240 was acceptable for SUV_90–110 in potential reference regions (6–11%), as well as in SUVR_90–110 (4–9%), DVR (3–10%), and R₁ (3–14%) in the evaluated target regions. A voxel-based PVC resulted in increased T-RT variability for SUVR_90–110 (10–26%) and DVR (6–22%), but was similar for PVC R₁ (3–20%). Therefore, caution should be taken when applying PVC to [18F]MK-6240 tau and R₁ outcome measures. Extracerebral SUVR_90–110 exhibited higher T-RT variability (~12%) than other evaluated regions (~6%) and revealed higher intra-subject variability that may affect the quantification in target and reference regions, particularly in longitudinal studies. Our observations emphasize the importance of considering and potentially correcting for the effects of extracerebral signal on [¹⁸F]MK-6240 PET analyses, especially in oCN. Our findings build upon results from previous studies, further suggesting [¹⁸F]MK-6240 is suitable for detecting early tau deposition and measuring longitudinal changes over time, as well as further supporting the viability of [¹⁸F]MK-6240 R₁ to evaluate longitudinal changes in perfusion.

The data supporting this study’s findings are not openly available due to reasons of sensitivity and are available from the corresponding author upon reasonable request. Data are in controlled access data storage at Massachusetts General Hospital.

C.L.: data analysis, figure generation, results interpretation, manuscript preparation (draft and edit). J.F.F.: data collection, data processing, results interpretation, manuscript preparation (edit). A.N.S.: data collection, data processing, results interpretation, provides critical feedback on manuscript. A.H.G. and D.H.: data processing, provide critical feedback on manuscript. H.S.: data collection, data processing, provides critical feedback on manuscript. D.I.G.: results interpretation, provides critical feedback on manuscript. N.D.M.: results interpretation, manuscript preparation (edit). B.D.: subject recruitment, provides critical feedback on manuscript. K.A.J.: subject recruitment, results interpretation, provides critical feedback on manuscript. C.C.: results interpretation, manuscript preparation (edit). J.C.P.: study design, results interpretation, manuscript preparation (edit).

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 1964 Helsinki Declaration and its later amendments or comparable ethical standards. Written consents were obtained from all participants. The study was approved by the Massachusetts General Hospital Institutional Review Board.

Not applicable.

The authors declared no potential competing interest with respect to the research, authorship, and/or publication of this article.

This research is supported by the following funding sources from the National Institutes of Health (NIH)/National Institute on Aging (NIA): R01AG050436, 5R01AG052414-03, and K99AG081457.

The authors thank all participants for their invaluable time and commitment. We also acknowledge the Martinos Center PET/MRI research technologists, G.E. Arabasz and S. Hsu, and the radiochemistry team for their essential assistance in conducting the studies. We thank the members of the Dickerson team, including A. Touroutoglou, M. Eldaief, S. McGinnis, and R. Eckbo, for their support in recruitment.

Supplementary material for this article is available with the online version here: https://doi.org/10.1162/imag_a_00402.