Alzheimer’s disease (AD), the most common form of progressive degenerative dementia, is characterized by cognitive deterioration and behavioral impairment. One of the neuropathological hallmarks of AD is the aggregation of hyperphosphorylated tau protein into intracellular neurofibrillary tangles (NFTs) (Braak and Braak, 1991). Tau pathology correlates with cognitive functions better than amyloid-β (Nelson et al., 2012; Murray et al., 2015). The recent advent of tau-specific positron emission tomography (PET) has enabled in vivo assessment of tau pathology in AD, which will aid in early detection, disease staging, and treatment development (Holtzman et al., 2016).

Because PET scanners have limited spatial resolution, the measured signals of small brain structures such as cortical gray matter are underestimated by partial volume effects (PVEs) (Matsuda et al., 2002; Yanase et al., 2005; Gonzalez-Escamilla et al., 2017). AD is characterized by progressive brain atrophy and thus PVEs are estimated to be even more severe in AD patients (Scheltens et al., 1992; Schroeter et al., 2009; Rullmann et al., 2016). Although some previous studies reported that PVC was effective for evaluating tau load in AD (Villemagne et al., 2014; Shidahara et al., 2017), other studies found that PVC produced only small effects (Ossenkoppele et al., 2016; Schöll et al., 2016; Maass et al., 2017). The aim of the present study was to determine whether PVC improves the precision of ¹⁸F-THK5351 PET measures of tau deposits in early AD.

The early AD patients showed significantly increased tau deposits on ¹⁸F-THK5351 PET data before and after PVC compared with the healthy controls (Figures 1, 2, 3 and Tables 2, 3). We found widespread neocortical tracer retention predominantly in the posterior cingulate and precuneus areas, as well as in the inferior temporal lobes, inferior parietal lobes, frontal lobes, and occipital lobes; the primary sensory and motor regions were spared. No significant difference of tracer retention was found in the white matter or subcortical structures between the two groups. The pattern of tracer distribution was similar between before and after PVC. PVC marginally reduced the size of clusters throughout the neocortex and no new clusters were produced, suggesting that PVC had little effect.

We also observed gray matter atrophy in the medial temporal lobes including hippocampal head extending into amygdala and lateral temporal lobes, as well as basal frontal lobes in AD patients (Table 4 and Figures 3, 4). No significant difference of atrophy was found in the white matter or subcortical structures between the two groups.

Visual evaluation of the atrophied areas and tau deposits measured by ¹⁸F-THK5351 PET after PVC in AD patients as compared to healthy controls revealed only slight overlap in the precuneus, posterior cingulate gyrus, lateral parietal cortex, inferior/lateral temporal cortex, and amygdala (Figures 5, 6). Among them, only few areas showed significant correlations in direct comparison of gray matter volume and PVE-corrected ¹⁸F-THK5351 retention (Table 5). An FDR correction showed significant negative correlations in the right fusiform gyrus (r = -0.78, p = 0.00012), left fusiform gyrus (r = -0.65, p = 0.0036), left inferior temporal gyrus (r = -0.73, p = 0.00059), right lingual gyrus (r = -0.66, p = 0.0032), left lingual gyrus (r = -0.76, p = 0.00023), right middle temporal gyrus (r = -0.72, p = 0.00075), left middle temporal gyrus (r = -0.75, p = 0.00038), and left parahippocampal gyrus (r = -0.71, p = 0.00089) in AD patients. The right fusiform gyrus, left lingual gyrus, and left middle temporal gyrus correlations survived Bonferroni correction for multiple comparisons. No significant negative correlations were found in healthy controls in this direct comparison.

A significant correlation with ¹⁸F-THK5351 retention and cognitive test scores (MMSE and CDR sum of boxes) within AD patients was not detected.

In this study, we compared tau deposits as measured by ¹⁸F-THK5351 PET before and after PVC along with regional brain atrophy using VBM. PVC was applied to correct for atrophy but had little effect, even in atrophied areas such as the medial/lateral temporal lobes and basal frontal lobes. Somewhat surprisingly, only a few associations were found between atrophy and tau deposits. The finding that tau load seemed to show very few relationships with gray matter atrophy has never been reported in previous tau PET studies.

A pathological hallmark of AD is that NFTs first form in the transentorhinal cortex, followed by the hippocampus, spread from the nearby cortex (predominantly the limbic, insular, and basal frontal areas) to the posterior cingulate cortex, and then spread throughout the neocortex (Braak and Braak, 1991; Braak et al., 2006, 2011). Previous tau PET studies reported tracer retention in the neocortical regions, most prominently throughout the temporal lobes, posterior cingulate/precuneus, and parieto-occipital and frontal lobes, whereas the primary sensory, motor, and visual areas were spared (Villemagne et al., 2014; Johnson et al., 2016; Ossenkoppele et al., 2016; Schöll et al., 2016; Maass et al., 2017). The widespread distribution of tau deposits in neocortical areas detected in our study was consistent with the results of previous studies (Serrano-Pozo et al., 2011; Villemagne et al., 2014; Johnson et al., 2016; Ossenkoppele et al., 2016; Schöll et al., 2016; Maass et al., 2017). Our results also corresponded to neuropathological findings in AD patients (Braak stages V/VI) (Braak et al., 2006, 2011).

We performed PVC to correct for brain atrophy. The pattern of tracer retention was similar between uncorrected and corrected data, suggesting that PVC had little effect, in line with previous studies with ¹⁸F-AV-1451 (Ossenkoppele et al., 2016; Schöll et al., 2016; Maass et al., 2017). A potential explanation for ineffectiveness of PVC was that we evaluated early AD patients with mild brain atrophy. However, VBM analysis revealed significant atrophy in the medial/lateral temporal lobes and basal frontal lobes in the AD patients. We could not detect significant tau deposits in these atrophied areas, even after PVC. This significant discrepancy between tau deposits and atrophy revealed in the present study might help to explain the mechanisms of ¹⁸F-THK5351 retention.

¹⁸F-THK5351 retention was detected in less atrophied areas in the neocortex but not in the most severely atrophied areas; these findings have not previously been reported. According to pathological studies, the spread of NFTs in the neocortex leads to synaptic dysfunction, glial activation, and neuronal loss (Sperling et al., 2014). At the neuronal cytoskeleton level, NFTs begin with the accumulation of soluble phosphorylated tau in cell bodies. As the phosphorylation progresses, this pretangle gradually aggregates and forms insoluble intracellular mature NFTs. After neuronal death, the abnormal hyperphosphorylated tau becomes exposed as extracellular ghost tangles (Braak et al., 1994, 2011). According to previous studies (Harada et al., 2016b; Ono et al., 2017; Saint-Aubert et al., 2017), all three families of radiotracers (¹¹C-PBB3, ¹⁸F-THK5117, ¹⁸F-THK5351, and ¹⁸F-AV-1451) are insensitive to pretangles. However, ¹¹C-PBB3, ¹⁸F-THK5117, and ¹⁸F-AV-1451 are sensitive to both intracellular mature tangles and extracellular ghost tangles. ¹⁸F-THK5351 is sensitive to tangles but discrimination between mature tangles and ghost tangles has not been demonstrated (Harada et al., 2016a). Lowe et al. (2016) recently reported that ¹⁸F-AV-1451 has a preference for mature tangles over pretangles and ghost tangles, suggesting that morphological differences might affect the binding intensity. If ¹⁸F-THK5351 has a similar preference, this could explain our results because areas with more severe neuronal loss have more ghost tangles (Fukutani et al., 1995), which might be less sensitive to tracer binding.

Interestingly, ¹⁸F-THK5351 retention was found in more advanced Braak stages compared with the Braak stage of atrophied areas revealed by VBM. A recent report on retrospective cortical atrophy and tau PET in clinically normal elderly adults revealed a correlation between retrospective parahippocampal atrophy and inferior temporal AV-1451 retention instead of parahippocampal retention, suggesting that tau pathology precedes atrophy (LaPoint et al., 2017). While our results support this hypothesis, longitudinal analysis of the association between tau pathology and atrophy are needed.

Although hippocampal atrophy is a key structural finding in AD (Matsuda, 2016), we did not find significant differences in hippocampal ¹⁸F-THK5351 retention between the AD patients and controls even after PVC. Johnson et al. (2016) reported that AV-1451 retention in the hippocampus was relatively low, which is consistent with our findings. They considered that off-target binding in the choroid plexus might affect retention in the hippocampus. However, there must be another reason because the THK series does not have off-target binding in the choroid plexus (Ishiki et al., 2015; Harada et al., 2016a; Saint-Aubert et al., 2017). Schöll et al. (2016) reported that PVC produced only a small increase in tracer retention in the hippocampus, which is in line with our results. Recent imaging and pathological studies revealed that medial temporal atrophy and NFTs that are restricted to the medial temporal lobe in the absence of amyloid plaques are generally observed in cognitively normal healthy elderly adults (primary age-related tauopathy) (Crary et al., 2014; Jack et al., 2016; Shimada et al., 2016). Cognitively impaired individuals have a large proportion of ghost tangles in the medial temporal lobe compared with healthy elderly adults (Santa-Maria et al., 2012; Crary et al., 2014). Thus, we believe that individuals with mature tau deposits and patients with mild cognitive impairment or AD with increased tau deposits that are insensitive to tracer binding show small differences in tracer uptake in the hippocampus.

This study has several limitations. First, this study had small sample sizes of both early AD patients and healthy controls. Second, we did not collect genetic data (apolipoprotein E). Therefore, further longitudinal studies are needed to investigate whether tau deposits predict atrophy. Third, a recent human blocking study using monoamine oxidase B (MAO-B) inhibitor demonstrated significant reduction of ¹⁸F-THK5351 SUVs in cortical regions (Ng et al., 2017). Even the reference region such as cerebellar cortex, which is considered the least-affected region, also reduced ¹⁸F-THK5351 SUVs by 41.6%. Because ¹⁸F-THK5351 retention seems to reflect reactive astrocytes in addition to tau pathology, it is hard to reveal the exact correlation of atrophy and tau pathology. Further studies are needed to clarify ¹⁸F-THK5351 binding sensitivity to tau pathology and MAO-B.

We applied PVC to correct for atrophy in early AD. However, the pattern of ¹⁸F-THK5351 tracer retention was similar between before and after PVC, suggesting that PVC is not required to estimate tau deposits. Moreover, ¹⁸F-THK5351 retention was detected in less atrophied areas but not in the most severely atrophied areas even after PVC. Although this discrepancy between tau deposits and atrophy could be ¹⁸F-THK5351 specific findings, our findings suggest that tau load precedes atrophy. Further prospective longitudinal studies of the association between tau pathology and atrophy are needed to confirm our findings.

YS, DS, and HM designed the experiments. HT, YY, MS, and TT collected and sorted the data. YS, DS, EI, NM, MO, and KK carried out the experiments, data analysis, and statistical analysis. YS, DS, and HM wrote the manuscript. SF, YK, NS, and NO revised the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.