β-Amyloid discordance of cerebrospinal fluid and positron emission tomography imaging shows distinct spatial tau patterns

Abstract Extracellular β-amyloid plaques and intracellular neurofibrillary tau tangles are the primary hallmarks of Alzheimer's disease. β-Amyloid pathology can be directly quantified by positron emission tomography imaging or indirectly by measuring the decrease of cerebrospinal fluid β-amyloid42/β-amyloid40 ratio. Although these two β-amyloid biomarkers may be considered interchangeable, they sometimes show discordance, particularly in early stage of Alzheimer's disease. Individuals with cerebrospinal fluid β-amyloid positive only or β-amyloid positron emission tomography positive only may be at early amyloidosis stage compared to those who are cerebrospinal fluid β-amyloid negative and β-amyloid positron emission tomography negative orcerebrospinal fluid β-amyloid positive and β-amyloid positron emission tomography positive. Besides, β-amyloid pathology may play an initiating role in Alzheimer's disease onset, leading to subsequent tau increases. However, it is still unclear whether individuals with different β-amyloid pathways have distinct spatial patterns of cortical tau tangles in early amyloidosis stage. In this study, we analyzed 238 cognitively unimpaired and 77 mild cognitive impairment individuals with concurrent (interval of acquisition <1 year) 18F-flortaucipir tau positron emission tomography, β-amyloid (18F-florbetapir or 18F-florbetaben) positron emission tomography and cerebrospinal fluid β-amyloid42 and β-amyloid40 and cerebrospinal fluid p-Tau181 and divided them into four different cerebrospinal fluid/positron emission tomography groups based on the abnormal status of cerebrospinal fluid β-amyloid42/β-amyloid40 (cerebrospinal fluid±) and β-amyloid positron emission tomography (±). We determined the cortical regions with significant tau elevations of different cerebrospinal fluid/positron emission tomography groups and investigated the region-wise and voxel-wise associations of tau positron emission tomography images with cerebrospinal fluid β-amyloid42/β-amyloid40, β-amyloid positron emission tomography and cerebrospinal fluid p-Tau/β-amyloid40 in early (cerebrospinal fluid positive/positron emission tomography negative and cerebrospinal fluid negative/positron emission tomography positive) and late (cerebrospinal fluid positive/positron emission tomography positive) amyloidosis stages. By compared to the cerebrospinal fluid negative/positron emission tomography negative individuals (Ref) without evidence of tau increase measured by cerebrospinal fluid or positron emission tomography, cerebrospinal fluid positive/positron emission tomography negative individuals showed higher tau in entorhinal but not in BraakIII/IV and BraakV/VI, whereas cerebrospinal fluid negative/positron emission tomography positive individuals had significant tau elevations in BraakV/VI but not in entorhinal and BraakIII/IV. In contrast, cerebrospinal fluid positive/positron emission tomography positive individuals showed significant tau increases in all the cortical regions than the Ref group. The voxel-wise analyses provided further evidence that lower cerebrospinal fluid β-amyloid42/β-amyloid40 was associated with higher tau in entorhinal, whilst higher β-amyloid positron emission tomography was related to higher tau in BraakV/VI regions in early amyloidosis stage. Both lower cerebrospinal fluid β-amyloid42/β-amyloid40 and higher β-amyloid positron emission tomography were correlated with tau aggregation in all the Braak stages regions in late amyloidosis stage. These findings provide novel insights into the spatial patterns of cortical tau tangles in different amyloidosis stages of Alzheimer's disease, suggesting cerebrospinal fluid β-amyloid and β-amyloid positron emission tomography discordant groups may have distinct characteristics of cortical tau tangles in early amyloidosis stage.

Furthermore, Aβ pathology may play an initiating role in Alzheimer's disease onset, leading to subsequent tau increases 6,33,34 or longitudinal tau changes. 3,[35][36][37] The Braak I-VI stages 38 have been proposed to characterize the spatial patterns of cortical neurofibrillary tau tangles based on the autopsy data. PET imaging studies [39][40][41][42][43] provide further evidence that the cortical tau tangles may initially present in entorhinal cortex, following by spreading to the Braak III/IV and Braak V/VI cortical regions in the presence of substantial Aβ burden. However, it is still unclear whether individuals who are on the Alzheimer's continuum but with different Aβ pathways have distinct spatial patterns of cortical tau aggregation. Exploring the spatial distribution of cortical tau tangles is important for understanding the characteristics of Alzheimer's disease pathophysiology in different stages and may provide novel reference for designing anti-tau clinical trials of Alzheimer's disease.
In this study, we analyzed non-demented Alzheimer's Disease Neuroimaging Initiative (ADNI) participants who had concurrent (within 1 year) CSF Aβ 42 /Aβ 40 , phosphorylated tau (p-Tau), Aβ PET and tau PET data and divided them into four different CSF/PET amyloidosis stages based on the abnormal status of CSF Aβ 42 /Aβ 40 and Aβ PET. In order to explore the spatial patterns of cortical tau tangles in different amyloidosis stages, we determined the cortical regions with significant tau elevation of different CSF/PET Aβ stages and investigated the region-wise and voxel-wise associations of tau PET images with CSF Aβ 42 /Aβ 40 , Aβ PET and CSF p-Tau/Aβ 40 in early (CSF+/PET− and CSF−/ PET+) and late (CSF+/PET+) amyloidosis stages.

Participants
The data were obtained from the ADNI database (ida.loni.usc.edu). The ADNI study was approved by institutional review boards of all participating centres, and written informed consent was obtained from all participants or their authorized representatives. In this study, we identified 238 cognitively unimpaired (CU) and 77 mild cognitive impairment (MCI) ADNI participants with concurrent (interval of acquisition ,1 year) 18 F-flortaucipir (FTP) tau PET, amyloid [ 18 F-florbetapir (FBP) or 18 F-florbetaben (FBB)] PET, CSF Aβ 42 and Aβ 40 and CSF p-Tau. We divided these 315 participants into four CSF/PET groups according to Aβ positivity defined by the thresholds of CSF Aβ 42 /Aβ 40 and Aβ PET as described below: CSF−/PET− (concordant Aβ negative), CSF+/PET− (discordant CSF Aβ positive), CSF−/ PET+ (discordant Aβ PET positive) and CSF+/PET+ (concordant Aβ positive). In order to control for the influence of non-Alzheimer's related tauopathy, 43 CSF−/PET− participants with either abnormal CSF p-Tau/Aβ 40 36 or abnormal FTP tau PET (entorhinal or Temporal-metaROI 44 ) were excluded from the CSF−/PET− group, and the rest of the CSF−/PET− group were defined as the reference (Ref) group.
FTP tau PET scans of 5 minutes × 4 frames were realigned, averaged and registered to the baseline MRI scan that was closest in time to the baseline FTP scan. FTP uptakes in 68 cortical ROIs defined by Desikan-Killiany atlas 45 were extracted in native FTP tau PET space, and one composite Temporal-metaROI 44 (including entorhinal, parahippocampal, fusiform, amygdala, inferior temporal and middle temporal) FTP SUVR was calculated by referring to a mean inferior cerebellar grey matter uptake. 47 In order to evaluate tau deposition in different Braak neurofibrillary tau stages, 38 we also calculated mean FTP SUVRs in Braak III/IV and Braak IV/VI that correspond to anatomical definitions of Braak stage III/IV (temporal/limbic) and V/VI (neocortical). 39 The thresholds of entorhinal FTP SUVR and Temporal-meta ROI FTP SUVR were set as ≥1.21 and ≥1.25 respectively according to an ROC analysis using the Youden index classifying 280 Aβ-ADNI CU participants and 183 Aβ+ ADNI MCI and Alzheimer's disease patients as the endpoint as described previously 36 and also in supplemental material (Supplemental Figs. 1-4).
For the voxel-wise analyses, FTP PET images were spatially normalized to the MNI space, intensity normalized at the voxel-wise level by a mean inferior cerebellar grey matter uptake 47 and smoothed using a Gaussian kernal of 8 mm in SPM12 (Welcome Department of Imaging Neuroscience, London, UK).

CSF biomarkers
CSF Aβ 40 , Aβ 42 and p-Tau 181 data were analyzed by the University of Pennsylvania ADNI Biomarker core laboratory using the fully automated Roche Elecsys and cobas e 601 immuno-assay analyzer system. CSF data (UPENN-BIOMK10_07_29_19.csv) were downloaded from the ADNI website. Considering that using a CSF p-Tau/Aβ 40 ratio may reduce measurement error likely related to individual differences in CSF production rather than pathology, and improve associations with AD biomarkers compared to using CSF p-Tau alone, 36 we decided to use CSF p-Tau/Aβ 40 ratio to represent CSF tau in this study. The CSF Aβ 42 /Aβ 40 ratio and CSF p-Tau/Aβ 40 ratio 36 were calculated, and their thresholds were defined as ≤0.054 and ≥0.0012 respectively according to the ROC analysis using the Youden index classifying 181 Aβ-CU participants and 163 Aβ+ cognitively impaired participants (MCI and Alzheimer's disease patients) as the endpoint as described in supplemental material (Supplemental Figs. 5-8).

Statistical analysis
Normality of distributions was tested using the Shapiro-Wilk test and visual inspection of data histograms. Data are presented as median (interquartile range [IQR]) or number and percentage. Different CSF/PET groups were compared using a Mann-Whitney U test for continuous characteristics unless otherwise noted. We assessed categorical differences using Fisher's exact test. A false discovery rate (FDR) of 0.05 using Benjamini-Hochberg approach was employed for multi comparisons correction.
In order to investigate tau elevations of different CSF/PET groups, we used generalized linear model (GLM) to compare FTP SUVRs in 68 FreeSurfer-defined ROIs of CSF+/PET−, CSF−/PET+ and CSF+/PET+ groups with the Ref group (CSF−/PET− without evidence of elevated tau measured by either CSF or PET), controlling for age and sex. We also compared entorhinal FTP SUVR, Braak III/IV FTP SUVR, Braak V/VI FTP SUVR of different CSF/PET groups with the Ref group.
In addition, the voxel-wise FTP PET images of the CSF+/ PET−, CSF−/PET+ and CSF+/PET+ groups were compared with the Ref group using two-sample t-test in SPM12, controlling for age and sex. The voxel-wise comparison between the CSF+/PET− group and the Ref group was presented using an uncorrected voxel threshold of P , 0.001, whilst the other comparisons were presented using an uncorrected voxel threshold of P , 0.001 and with family-wise error (FWE) corrected P , 0.05 at the cluster level.
As we described previously, 7 both CSF+/PET− group and CSF−/PET+ group were defined as early amyloidosis stage whilst CSF+/PET+ individuals as late amyloidosis stage of Alzheimer's disease. In order to determine the associations of different Aβ biomarkers as well as CSF p-Tau/Aβ 40 with cortical tau deposition of different Braak stages (entorhinal, Braak III/IV , and Braak V/VI ) cortical regions in different amyloidosis stages, we used GLM model to investigate the association of FTP SUVRs in entorhinal, Braak III/IV and Braak V/VI with CSF Aβ 42 /Aβ 40 , Aβ PET and CSF p-Tau/Aβ 40 in early (CSF+/PET− and CSF−/PET+ groups and including the CSF−/PET− group as the reference) and late (CSF+/PET+ group and including the CSF−/PET− group as the reference) amyloidosis stages separately, controlling for age and sex. We also investigated the voxel-wise association of FTP PET images with CSF Aβ 42 /Aβ 40 , Aβ PET and CSF p-Tau/ Aβ 40 in early and late amyloidosis stages separately, controlling for age and sex. The voxel-wise association between FTP SUVR images and CSF Aβ 42 /Aβ 40 in early amyloidosis stage was presented using an uncorrected voxel threshold of P , 0.005, whilst the other comparisons were presented using an uncorrected voxel threshold of P , 0.001 and with FWE corrected P , 0.05 at the cluster level.
We noticed that one CSF+/PET− individual and one CSF+/PET+ individual had extremely high entorhinal FTP SUVRs; thus, we repeated all the analyses after removing them from the dataset. In addition, in order to control for the influence of those individuals around the thresholds of CSF Aβ 42 /Aβ 40 and Aβ PET SUVR, we also repeated all the analyses after excluding the borderline participants who were within +5% of the CSF Aβ 42 /Aβ 40 and Aβ PET (SUVR) thresholds.
Statistical analyses were performed in the statistical programme R (v3.6.2, The R Foundation for Statistical Computing) unless otherwise noted.

Data availability
All data used in the current study were obtained from the ADNI database (available at https://adni.loni.usc.edu).

Demographics
The characteristics of the participants analyzed in this study can be found in Table 1 Cortical tau elevations of different CSF/PET groups   In the voxel-wise analysis, the CSF+/PET− group showed higher tau PET in the left entorhinal cortex and parahippocampal ( Fig. 2D, P , 0.001 without cluster correction) than the Ref group, whereas the CSF−/PET+ group showed significant increases of FTP SUVRs in Braak IV stage ROIs (bilateral caudal anterior cingulate, rostral anterior cingulate, insula, posterior cingulate and isthmus cingulate) and most of the Braak V/VI stage ROIs (bilateral medial orbitofrontal, caudal middle frontal, rostral middle frontal, superior temporal, precuneus, postcentral, superior frontal, precentral and paracentral) (Fig. 2E, P , 0.001 with FWE P , 0.05 cluster correction). In addition, the CSF+/PET+ group showed significant (Fig. 2F, P , 0.001 with FWE cluster correction) elevated tau in most of the cortical regions except for pericalcarine and cuneus.
After removing two individuals with extremely high entorhinal FTP SUVRs, the CSF+/PET− group still had significantly higher tau deposition in the left entorhinal than the Ref group, and the results of other comparisons were substantially the same (Supplemental Fig. 9). Besides, the two discordant groups also showed distinct spatial cortical tau deposition, whilst we removed the individuals who were within +5% of the CSF Aβ 42 /Aβ 40 and Aβ PET thresholds (Supplemental Fig. 14), although more cortical regions (left entorhinal, parahippocampal, fusiform, inferior temporal, middle temporal, superior temporal and BANKSSTS) showed significant tau increases in CSF+/PET− group compared to the Ref group (Supplemental Fig. 15).
In late amyloidosis stage, lower CSF Aβ 42 /Aβ 40 was negatively related to higher FTP SUVR in entorhinal  After removing two individuals with extremely high entorhinal FTP SUVR, the significant association between tau PET and CSF p-Tau/Aβ 40 in early amyloidosis stage disappeared, whilst the other associations retained (Supplemental Figs. 11 and 12). Besides, the results were substantially the same whilst we removed the borderline individuals (Supplemental Figs. 17 and 18).

Voxel-wise analysis of cortical tau with CSF biomarkers and Aβ PET in early and late amyloidosis stages
In early amyloidosis stage, lower CSF Aβ 42 /Aβ 40 was significantly associated with higher FTP SUVR in the left entorhinal cortex (Fig. 5A), higher Aβ PET Centiloid was significantly related to higher FTP SUVRs in right insula and bilateral cingulate cortex, paracentral, frontal and parietal cortices (Fig. 5C), and CSF p-Tau/Aβ 40 showed significant positive association with FTP SUVRs in left entorhinal, parahippocampal, fusiform, inferior temporal, middle temporal and BANKSSTS (Fig. 5E). In contrast, CSF Aβ 42 / Aβ 40 , Aβ PET and CSF p-Tau/Aβ 40 all showed significant relation with FTP SUVRs in all the Braak ROIs, and the strongest association was found between FTP SUVR and CSF p-Tau/Aβ 40 in late amyloidosis stage (Fig. 5B, D and F). Besides, the strongest associations with tau PET for all the biomarkers were found in early Braak ROIs (entorhinal and Braak III/IV ) in late amyloidosis stage. After removing two individuals with extremely high entorhinal FTP SUVRs, the significant association between tau PET and CSF p-Tau/Aβ 40 in early amyloidosis stage disappeared, whilst the other associations retained (Supplemental Fig. 13). Besides, the results were substantially the same whilst we removed the borderline individuals (Supplemental Fig. 19).

Discussion
In this study, we investigated the cortical tau deposition measured by tau PET imaging of different amyloidosis stages defined by CSF Aβ 42 /Aβ 40 and Aβ PET in non-demented elderly adults. Compared to CSF Aβ 42 /Aβ 40 negative and Aβ PET negative (CSF−/PET−) individuals without tau increase in CSF or cortex, we found individuals with abnormal CSF Aβ 42 /Aβ 40 only (CSF+/PET−) showed higher tau in entorhinal but not in Braak III/IV and Braak V/VI , whereas individuals with abnormal Aβ PET only (CSF−/PET+) had significant tau elevations in Braak V/VI but not in entorhinal and Braak III/IV . The voxel-wise analyses provided further evidence that lower CSF Aβ 42 /Aβ 40 was associated with higher tau in entorhinal, whilst higher Aβ PET was related to higher tau in Braak V/VI ROIs in early amyloidosis stage (CSF+/ PET− and CSF−/PET+). In contrast, individuals with abnormal CSF Aβ 42 /Aβ 40 and abnormal Aβ PET (CSF+/ PET+) had significant tau elevations in all the Brank ROIs, and both lower CSF Aβ 42 /Aβ 40 and higher Aβ PET were correlated with higher tau in entorhinal, Braak III/ IV and Braak V/VI in late amyloidosis stage. These findings provide novel insights into understanding the cortical tau aggregation in different amyloidosis stages of Alzheimer's disease, suggesting CSF Aβ and Aβ PET discordant individuals may have initial tau tangles in distinct cortical regions in early amyloidosis stage of Alzheimer's disease.
Whilst CSF measurement of p-Tau provides complementary early tau increase, 36,49,50 PET imaging offers spatial information on where tau deposits. Notably, our group 36 and other laboratories 49,50 very recently observed evidence that CSF p-Tau may detect early tau increase than the tau PET imaging, which was also supported by postmortem studies. [51][52][53] Importantly, both dichotomous and continuous analyses showed that the CSF-first Aβ pathway had Aβ-related tau increase in entorhinal cortex, which has been regarded as the earliest cortical region of tau aggregation. [38][39][40][41][42][43] In contrast, the PET-first Aβ pathway showed significant Aβ-related tau increase in Braak V/VI but not in entorhinal and Braak III/IV . In concordance with our findings, one recent important ADNI study 32 also found that the CSF+/PET− and CSF−/ PET+ individuals have numerically (not significant) higher and lower entorhinal tau measured around 5 year postbaseline CSF Aβ and Aβ PET than the CSF−/PET− individuals respectively, although they used CSF Aβ 42 to define CSF Aβ status, which may be less reliable than CSF Aβ 42 /Aβ 40 used in this study. They also found CSF−/PET+ but not CSF+/PET− individuals had significant tau increase in Braak V/VI but not in entorhinal and Braak III/IV , which was consistent with our findings. Together with our findings and previous study, 32 it is probably that individuals with PET-first Aβ pathway may not have tau increases in entorhinal and Braak III/IV cortical regions due to their lower CSF p-Tau, which plays an important role in tau spreading in cortical regions of early Braak stages (entorhinal and Braak III/ IV ). 36,49,50 The voxel-wise results provide further evidence to support the notion that CSF Aβ 42 /Aβ 40 may be related to tau aggregation in entorhinal cortex whereas cortical Aβ burden correlates with elevated tau in cortical regions of Braak V/ VI stage in early amyloidosis stages. Notably, the CSF−/ PET+ individuals had smaller CSF p-Tau/Aβ 40 ratio than the CSF+/PET− individuals, suggesting we may not be able to use CSF p-Tau biomarker to represent complementary Braak V/VI tau increase in CSF−/PET+ individuals.
Our group 7 and other laboratory 11,54 previously observed that CSF+/PET− individuals were accumulating cortical Aβ burden with a similar rate to the CSF+/PET+ individuals, and will become CSF+/PET+ in future. Besides, the CSF-first Aβ pathway has been observed more frequent than the PET-first Aβ pathway according to the previous reports. 9,[13][14][15]17 In this study, we found similar proportion of CSF and PET Aβ discordant groups, but our previous longitudinal analyses 7 also support that CSF Aβ may become abnormal earlier than Aβ PET. Consequently, it is probably that the CSF-first Aβ pathway may represent the typical evolution of Alzheimer's disease which shows Alzheimer's typical tau spreading pattern, [38][39][40][41][42][43] whilst the PET-first Aβ pathway may have cortical Aβ-burden related hippocampalsparing elevated tau in early amyloidosis stage of Alzheimer's disease. The Temporal-metaROI (entorhinal, amygdala, parahippocampal, fusiform, inferior temporal and middle temporal) 44 composite regions have been commonly used to detect Alzheimer's-related tau deposition in human brain. 3,55-60 However, our findings suggested that we may not be able to use Temporal-metaROI regions to capture the cortical tau increase in early amyloidosis stage among these individuals who have a PET-first Aβ pathway, implying different cortical regions should be selected to detect early tau increase in early Alzheimer's disease.
In contrast, we found CSF Aβ positive and Aβ PET positive (CSF+/PET+, late amyloidosis stage) individuals showed significant tau increases in all the cortical regions than the control group, and the highest tau elevations were found in the Temporal-metaROI 44 composite regions. Furthermore, the voxel-wise analyses revealed that lower CSF Aβ 42 /Aβ 40 , higher Aβ PET, and larger CSF p-Tau/Aβ 40 ratio were related to significant tau elevations in all the cortical regions and with Temporal-metaROI 44 composite regions showing the strongest association. These findings suggest that it is reasonable to use Temporal-metaROI regions to detect cortical tau increase in CSF and PET Aβ concordant individuals.

Limitations
This study has several limitations. First, as CSF Aβ and Aβ PET have a very high agreement, the sample sizes of the discordant CSF/PET Aβ groups with concurrent tau PET imaging were relatively small and the results need to be replicated in a larger cohort. To the best of our knowledge, there is currently no larger cohort available with the measurements needed for this analysis. Second, the ADNI participants overall are a highly selected sample, recruited to reflect the exclusionary criteria and types of individuals likely to participate in clinical trials; thus, it would be extremely useful to validate these findings in other aging cohorts. Third, our analyses were limited to cross-sectional PET measured with FTP, which may need to be replicated using longitudinal data and other PET ligands. Forth, the threshold of CSF Aβ 42 /Aβ 40 was defined on the basis of the ADNI database, which requires validation with other databases.

Conclusion
In conclusions, we found that CSF and PET Aβ discordant individuals have distinct cortical tau deposition patterns in non-demented elderly adults. Recent studies [61][62][63][64] suggest that Alzheimer's disease may have distinct biological features of tau spreading patterns, which are important for explaining the heterogeneity of tau-related neurodegeneration and cognitive decline and the design of anti-tau clinical trials. Our findings are useful for understanding the subtypes of tau spreading patterns in Alzheimer's disease and provide novel reference for cortical tau detection in individuals who are at early amyloidosis stage.
Technologies; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Piramal Imaging; Servier; Takeda Pharmaceutical Company; and Transition Therapeutics. The Canadian Institutes of Health Research is providing funds to support ADNI clinical sites in Canada. Private sector contributions are facilitated by the Foundation for the National Institutes of Health (www.fnih.org). The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimer's Disease Cooperative Study at the University of California, San Diego. ADNI data are disseminated by the Laboratory for Neuro Imaging at the University of Southern California.