Comparing the Clinical Utility and Diagnostic Performance of CSF P-Tau181, P-Tau217, and P-Tau231 Assays

Background and Objectives Phosphorylated tau (p-tau) in CSF is considered an important biomarker in Alzheimer disease (AD) and has been incorporated in recent diagnostic criteria. Several variants exist, including p-tau at threonines 181 (p-tau181), 217 (p-tau217), and 231 (p-tau231). However, no studies have compared their diagnostic performance or association to β-amyloid (Aβ) and tau-PET. Understanding which p-tau variant to use remains an important yet answered question. We aimed to compare the diagnostic accuracy of p-tau181, p-tau217, and p-tau231 in CSF for AD and their association with Aβ and tau-PET. Methods A total of 629 participants in the Swedish BioFINDER-2 study were included (cognitively unimpaired, n = 334; Aβ-positive mild cognitive impairment, n = 84; AD dementia, n = 119; and non-AD disorders, n = 92). In addition to p-tau181 and p-tau217 measured using assays with the same detector antibodies from Eli Lilly (p-tau181Lilly, p-tau217Lilly) and p-tau231, we also included p-tau181 measurements from 2 commonly used assays (Innotest and Elecsys). Results Although all p-tau variants increased across the AD continuum, p-tau217Lilly showed the greatest dynamic range (13-fold increase vs 1.9–5.4-fold increase for other p-tau variants for AD dementia vs non-AD). P-Tau217Lilly showed stronger correlations with Aβ- and tau-PET (p < 0.0001). P-Tau217Lilly exhibited higher accuracy than other p-tau variants for separating AD dementia from non-AD (area under the curve [AUC], 0.98 vs 0.88 [p < 0.0001] - 0.96 [p < 0.05]) and for identifying Aβ-PET (AUC, 0.86 vs 0.74 [p < 0.0001] and 0.83 [p < 0.001]) and tau-PET positivity (AUC, 0.94 vs 0.80—0.92, p < 0.0001). Finally, p-Tau181Lilly generally performed better than the other p-tau181 assays (e.g., AD dementia vs non-AD, AUC, 0.96 vs 0.88 [p-tau181Innotest] and 0.89 [p-tau181Elecsys]; p < 0.0001). Discussion CSF p-tau217Lilly seems to be more useful than other included p-tau assays in the workup of AD. Varied results across p-tau181 assays highlights the importance of anti-tau antibodies for biomarker performance. Classification of Evidence This study provides Class II evidence that p-tau217 provides higher diagnostic accuracy for diagnosis of AD dementia than p-tau181 or p-tau231.

In addition to extracellular deposition of β-amyloid (Aβ) plaques, Alzheimer disease (AD) is defined by the intracellular aggregation of tau in neurofibrillary tangles (NFTs), composed of abnormally hyperphosphorylated tau. 1 Tau pathology is thought to be reflected in CSF levels of phosphorylated tau (ptau). CSF p-tau has shown high prognostic accuracy for AD and for predicting cognitive decline in cognitively unimpaired (CU) individuals and in patients with mild cognitive impairment (MCI) due to AD. 2,3 As CSF p-tau levels are higher in AD compared to other non-AD neurodegenerative disorders, including progressive supranuclear palsy (PSP), corticobasal syndrome (CBS), frontotemporal dementia (FTD), and vascular dementia (VaD), it has also proven of use in the differential diagnosis of AD vs other dementias. 4 Tau in CSF is largely present in the form of different fragments. [5][6][7][8] Of these, N-terminal and midregion variants are the most abundant. In addition, numerous sites exist where tau can undergo abnormal hyperphosphorylation. 9 The most commonly used assays for p-tau, however, use antibodies targeting the midregion of tau as well as an antibody targeting tau phosphorylated at threonine-181 (p-tau181). 10 Besides p-tau181, increased levels of mid tau fragments phosphorylated at threonine-231 (p-tau231) appear to be an early occurrence in AD, preceding the formation of paired helical filaments. 11 Although studies have shown that p-tau231 can accurately discriminate patients with AD from CU individuals and patients with non-AD disorders, similar to p-tau181, a series of postmortem studies that examined both measures reported that CSF p-tau231 was better associated with neocortical fibrillary pathology than CSF p-tau181. 12,13 Recently, p-tau fragments phosphorylated at threonine-217 (p-tau217) were also measured in CSF. 14 Compared to p-tau181, p-tau217 showed stronger correlations with Aβ and tau-PET and more accurately distinguished AD dementia from non-AD neurodegenerative disorders. 14 Additional work has shown that p-tau181 and p-tau217 are increased already in preclinical AD (Aβ-positive CU), with these increases preceding tau-PET positivity and even occurring prior to the threshold for Aβ-PET positivity. 15,16 Overall, findings indicate that increases in CSF p-tau occur in response to very early Aβ pathology and precede widespread tau aggregation. Thus far, however, there are no studies comparing p-tau181, p-tau217, and p-tau231 levels in relation to Aβ and tau-PET across the symptomatic stages of AD, nor data directly comparing their diagnostic performance for separating AD dementia from non-AD neurodegenerative disorders and for identifying abnormal Aβ and tau-PET status. Because CSF p-tau is an important biomarker in the workup for AD and is incorporated in its diagnostic criteria, 17 it is of great importance to determine which of these p-tau variants to use, especially as clinical heterogeneity and different stages in AD may be determined by heterogeneity in the post-translational modification (PTM) of tau. 18 We aimed to address these questions using cross-sectional data from a well-characterized cohort, ranging from Aβ-negative CU individuals to Aβ-positive CU individuals and Aβpositive patients with MCI or AD dementia. In addition to comparing p-tau181 and p-tau217 measured using assays with the same detector antibodies from Eli Lilly (p-tau181 Lilly and p-tau217 Lilly ) with p-tau231 measured using an assay with a phospho-specific cis-conformational monoclonal antibody (p-tau231 ADx ), we also compared p-tau181 Lilly with p-tau181 measurements from 2 commonly used assays (Innotest[p-tau181 Innotest ] and Elecsys[p-tau181 Elecsys ]).

Participants
We included participants from the prospective and longitudinal Swedish BioFINDER-2 study (clinical trial NCT03174938), including CU participants and patients with MCI, AD dementia, and non-AD neurodegenerative disorders. CU individuals were aged ≥60 years and did not have MCI or dementia. 17 Patients with MCI fulfilled DSM-5 criteria for mild neurocognitive disorder 19 and patients with AD dementia fulfilled the DSM-5 criteria for major neurocognitive disorder due to AD. 19 Patients with non-AD disorders fulfilled diagnostic criteria for PSP or CBS, 20,21 Parkinson disease (PD) with or without cognitive impairment, 22 FTD, 23 or VaD. 24 Further details pertaining to inclusion and exclusion criteria are described in the Supplement (eAppendix 1, available from Dryad, doi.org/10.5061/dryad.4f4qrfjc7). Groups were established without the use of biomarkers, but CU and MCI participants were subdivided based on Aβ status, determined using CSF Aβ42/Aβ40 (Innotest; Fujirebio) and a cutoff of <0.089. 25 We included only Aβ-positive AD dementia cases, in keeping with the research framework by the National Institute on Aging-Alzheimer's Association. 17 As Aβ-PET is by design performed only in CU individuals and in patients with MCI, CSF Aβ42/ Aβ40 was thus chosen to have a common measure of Aβ pathology across all participants.
Standard Protocol Approvals, Registrations, and Patient Consents All participants gave written informed consent. Ethical approval was given by the Regional Ethical Committee in Lund, Sweden. Approval for PET imaging was obtained from the Swedish Medical Products Agency and the local Radiation Safety Committee at Skåne University Hospital, Sweden.
CSF P-Tau181 and P-Tau217 Measurements (Eli Lilly) Analysis of CSF mid-domain p-tau181 Lilly and p-tau217 Lilly was performed at Eli Lilly and Company using the Meso Scale Discovery (MSD) platform, as previously described. 14 The anti-p-tau217 antibody IBA413 and anti-p-tau181 antibody AT270 were used as capture antibodies in the p-tau181 and p-tau217 assays, respectively. Capture antibodies were conjugated with biotin (Thermo Scientific). Sulfo-tag (MSD) conjugated LRL antibody was used as a detector in both assays. The assays were calibrated using a recombinant tau (4R2N) protein that was phosphorylated in vitro using a reaction with glycogen synthase kinase-3 and characterized by mass spectrometry. Samples were analyzed in duplicate and the mean of duplicates was used in statistical analysis. Ten samples from Aβnegative CU individuals (1.59%) were below the limit of detection and were excluded, as were 5 patients with AD with very high p-tau217 Lilly levels (>3 SD above the mean).
CSF P-Tau231 Measurements (ADx NeuroSciences) CSF p-tau231 ADx was measured at ADx NeuroSciences with a research sandwich ELISA (version 1) according to the kit instructions. Phospho-specific cis-conformational monoclonal antibody ADx253 (T1H11) was used as a capture antibody and biotinylated pan-tau monoclonal antibody ADx205 (epitope region aa224-238) as a detector. The assay was calibrated using an in-house designed synthetic peptide combining both antibody epitopes and having the corresponding threonine231 phosphorylated and proline232 replaced by a homoproline, Pip, to reflect cis selectivity of ADx253. In all prior analyses, we observed a consistent low interplate variability well below 15%. Because p-tau231 quantifications require 80 mL per single measurements-requiring at least 160 mL per result-we opted to run the p-tau231 measurement in singlicate as the study was designed to explore the difference between the phospho-tau assays. Quality control samples run on each plate, which were leftovers of CSF, confirmed high precision of these runs with coefficients of variation below 15%.

Image Acquisition and Processing
Aβ and tau-PET were performed using [ 18

Regions of Interest and Cutoffs
Target regions of interest (ROIs) were chosen on the basis of previously published findings: a neocortical meta-ROI for Aβ-PET (prefrontal, lateral temporal, parietal, anterior cingulate, and posterior cingulate/precuneus) 28,29 and, for tau-PET, the entorhinal cortex (Braak I/II), a temporal meta-ROI (amygdala, inferior/middle temporal gyri, fusiform gyrus, and parahippocampal gyrus, approximating Braak III/IV), 30 and a neocortical meta-ROI capturing late stage tau pathology (Braak V/VI). 31 A priori cutoffs based on Gaussian mixture modeling (Aβ-PET) 14 and the mean SUVR within a given ROI plus 2.5 SDs among young Aβ-negative CU individuals (tau-PET) 25 were used to define positivity within these ROIs.

Statistical Analyses
Group differences in age-adjusted CSF p-tau levels were assessed using pairwise analysis of variance-based comparisons of linear regression models. Associations between CSF p-tau isoforms and between p-tau isoforms and ROI-based Aβ and tau-PET SUVR values were assessed using correlation analysis; differences between correlation coefficients were tested using a confidence interval (CI)-based approach with bootstrapping. 32 Log-transformed biomarker and PET measures were used in regression analyses. Generalized additive models with cubic regression splines were used to compare the slopes of CSF p-tau isoforms (mean change from Aβ-negative CU) across different tau and Aβ-PET SUVR values. Differences between the estimated functions were assessed by means of bootstrapped CIs. These were computed by repeatedly (n = 10,000) resampling the dataset (with replacement) and calculating the differences between spline fits. The discriminative performance of CSF p-tau measures was assessed using the area under the receiver operating characteristic (ROC) curve (AUC), adjusted for age. Significant differences in AUC values were tested using DeLong statistics 33 and Bonferroni correction was applied to account for multiple comparisons. In addition to AUC, sensitivity and specificity at the cutoff that resulted in the highest Youden index (sensitivity + specificity -1) are reported. Analyses were performed in R, v.4.0.2, with significance set at p < 0.05, 2-tailed. Voxelwise analyses examining the association between CSF p-tau levels and Aβ and tau-PET were performed using multilinear models, as implemented in SPM12, adjusted for age and the interval between lumbar puncture and PET scan.

Data Availability
Anonymized study data for the primary analyses presented in this report are available on request from any qualified investigator for purposes of replicating the results.  Table 1. For a flow diagram of participants included in the study, see eFigure 1 (available from Dryad, doi.org/10.5061/dryad.4f4qrfjc7).

Correlations Between P-Tau Isoforms
A schematic overview of the included p-tau assays is provided in Figure 1. P-Tau isoforms were strongly correlated across all participants (range 0.853-0.977, all p < 0.0001) (eFigure 2, available from Dryad, doi.org/10.5061/dryad.4f4qrfjc7). These associations were significant in Aβ-positive CU, Aβpositive MCI, and AD dementia, but not in Aβ-negative individuals. As moderate but significant correlations were observed between age and CSF p-tau levels (eTables 1 and 2, available from Dryad, doi.org/10.5061/dryad.4f4qrfjc7), age was accounted for when comparing CSF p-tau levels across tau-PET-based Braak stages and diagnostic groups.
CSF P-Tau Slopes as a Function of Aβ and Tau-PET Spline models examining CSF p-tau concentrations across Aβ and tau-PET are shown in Figure 2; CIs for differences in p-tau biomarkers at specified SUVR values are detailed in the Supplement (eTable 5, available from Dryad, doi.org/10. 5061/dryad.4f4qrfjc7). Using Aβ-PET, the slope of p-tau217 Lilly was significantly different from those of p-tau181 Innotest , p-tau181 Lilly , p-tau181 Elecsys , and p-tau231 ADx in Aβ-positive CU. The slope of p-tau181 Lilly differed significantly from those of p-tau181 Innotest and p-tau181 Elecsys ; no significant difference was seen between CIs for p-tau181 Lilly and p-tau231 ADx , however. The same pattern was seen for p-tau217 Lilly when looking at Aβ-PET in Aβ-positive MCI. When analyses were performed separately in Aβ-positive and Aβ-negative CU individuals (eFigure 5, eTable 6, available from Dryad, doi.org/10.5061/dryad.4f4qrfjc7), increasing fold change with increasing SUVR values and separation of p-tau trajectories was largely confined to the Aβ-positive CU group.
Using tau-PET SUVR in the Braak I/II ROI in CU individuals, the slope of p-tau217 Lilly differed significantly from those  of p-tau181 Innotest , p-tau181 Lilly , p-tau181 Elecsys , and p-tau231 ADx at SUVR values of 1.5 or greater. The slopes of p-tau181 Lilly and p-tau231 ADx also differed significantly from those of p-tau181 Innotest and p-tau181 Elecsys ; no significant difference was seen between p-tau181 Lilly and p-tau231 ADx . The same pattern was seen when using the Braak III/IV ROI in Aβ-positive cognitively impaired participants and using the Braak V/VI ROI (data not shown). Similar to the analyses with Aβ-PET, greater fold change at higher SUVR levels and separation of p-tau trajectories was largely confined to the Aβpositive CU group (eFigure 5 and eTable 6, available from Dryad, doi.org/10.5061/dryad.4f4qrfjc7).

CSF P-Tau Levels by Tau-PET-Based Braak Stages
When dividing participants on the basis of their tau-PET status in Braak ROIs (Figure 3)

CSF P-Tau Levels by Diagnostic Group
By comparison to all Aβ-negative participants, CSF p-tau levels were increased in Aβ-positive CU, MCI, and AD ( Figure 4). In Aβ-positive MCI and AD, the mean fold increases (compared to Aβ-negative CU) were between 7.36 (95% CI, 6. Diagnostic Accuracy of CSF P-Tau Isoforms ROC curves and associated AUC values are shown in Figure 5. AUC values-along with sensitivity and specificity estimates at cutoffs that resulted in the highest Youden index-are reported in Table 2. The diagnostic performance of CSF p-tau for AD dementia vs Aβ-negative CU ( Figure 5A) and non-AD neurodegenerative disorders ( Figure 5B) was highest using p-tau217 Lilly . For both contrasts, AUC values for p-tau217 Lilly were significantly higher than those for p-tau181 Innotest and p-tau181 Elecsys (p < 0.0001). For the separation of AD dementia from non-AD neurodegenerative disorders, the AUC value for p-tau217 Lilly was significantly higher than that for p-tau181 Lilly and p-tau231 ADx (p < 0.05). For both contrasts, AUC values for p-tau181 Lilly were significantly higher than those for p-tau181 Innotest and p-tau181 Elecsys (p < 0.0001).
Using tau-PET status in the Braak III/IV ( Figure 5D) and V/VI ( Figure 5E) ROIs, AUCs for p-tau217 Lilly were significantly higher than those for p-tau181 and p-tau231 ADx (p < 0.0001). Using both Aβ and tau-PET, the diagnostic performance of p-tau181 Lilly was superior to that of p-tau181 Innotest and p-tau181 Elecsys (p < 0.0001). Using tau-PET, AUC values for p-tau231 ADx were significantly higher than those for p-tau181 Innotest (Braak III/IV and Braak V/VI, p < 0.0001), p-tau181 Elecsys (Braak III/IV, p < 0.0001; Braak V/VI, p < 0.01), and p-tau181 Lilly (Braak III/IV and Braak V/VI, p < 0.01). The AUC value of p-tau181 Elecsys was significantly higher than that for p-tau181 Innotest using the Braak III/IV ROI (p < 0.05) but not when using the Braak V/VI ROI.

Discussion
Consistent with previous work using these assays, levels of CSF p-tau181 Lilly and p-tau217 Lilly were progressively higher across both the AD continuum (i.e., moving from Aβ-positive CU through Aβ-positive AD dementia) 14,15 and tau-PET Braak stages. 14 Furthermore, in agreement with a previous study, we found that p-tau217 Lilly had significantly higher correlations with Aβ and tau-PET as compared to p-tau181 Lilly 14 and extended this finding to show that the correlation was also significantly higher than for p-tau181 Innotest , p-tau181 Elecsys , and p-tau231 ADx . Previously, using CSF samples taken prior to baseline tau-PET in Aβ-positive CU, 56% of patients showed positive p-tau217-Lilly levels, compared with only 25% for p-tau181 Lilly . 15 Combined with mass spectrometry findings in AD showing an increased degree of phosphorylation at threonine 217 compared with position 181, 16,34 these results were interpreted as suggesting that phosphorylation at position 217 may be more pronounced by comparison to other sites. Although the differences were modest, stronger correlations observed with PET would also prove consistent with findings showing that threonine 217 phosphorylation was considerably increased in AD as compared to threonine 181 8,35 and with the preferential phosphorylation of tau at specific sites across the different stages of AD. 16,36 In addition, by comparison to studies using p-tau181 measurements from commercial assays such as p-tau181 Innotest and p-tau181 Elecsys , 37 larger effect sizes were seen when using p-tau181 Lilly , p-tau217 Lilly , and p-tau231 ADx .
Using spline-based analyses, we compared the slopes of p-tau isoforms in relation to continuous Aβ and tau-PET SUVR values. These analyses were performed in CU individuals and in cognitively impaired Aβ-positive patients. Although greater PET SUVR values were associated with higher CSF p-tau concentrations for all isoforms, no significant differences were seen in the courses of p-tau181 Lilly and p-tau231 ADx . By contrast, comparison of CIs showed that the slope of p-tau217 Lilly diverged from those of p-tau181 Lilly and p-tau231 ADx across a range of SUVR values, particularly in the CU group when using tau-PET in the Braak I/II ROI. These findings are consistent with increases in the active production of soluble tau in the presence of aggregated Aβ 8 and, possibly, with the idea that the relative phosphorylation of tau at specific sites varies across the course of AD. 16,36 Although findings with tau-PET in the cognitively impaired group suggest a plateau in the course of all 3 isoforms-possibly due to a process though which phosphorylation rates are reduced due to sequestration by hyperphosphorylated aggregates 38,39 -phosphorylation of threonine 217 may continue to increase later into the disease course, similar, for example, to what has been reported for p-tau205. 16 This, combined with p-tau217 Lilly possibly showing a higher specificity for AD, 14 may explain the higher AUC values seen for p-tau217 Lilly . Although a tau-centric hypothesis ascribing a primary role to tau 40 has been proposed as an alternative to the view that AD is caused by the accumulation of Aβ in the brain, 41 both spline-and correlation-based sensitivity analyses in CU individuals by Aβ status showed there to be little association with Aβ and tau-PET in Aβ-negative CU individuals. Braak stages III/VI was established using a priori cutoffs based on the mean SUVR within a given region of interest plus 2.5 SD among β-amyloid (Aβ)-negative young controls. Solid gray horizontal lines indicate age-adjusted group comparisons: Alzheimer disease (AD) dementia higher than all groups (p < 0.001); Aβpositive mild cognitive impairment (MCI) higher than cognitively unimpaired (CU) and non-AD (p < 0.001); Aβ-positive CU higher than Aβ-negative CU and non-AD (p < 0.001). In order to facilitate comparison between p-tau measures, y-axes were scaled to the maximum fold change seen across biomarkers. AD dem. = Alzheimer disease dementia; non-AD = non-Alzheimer disease neurodegenerative disorders.
A recent study focused on characterizing the patterns of change in p-tau231 ADx and p-tau181 Elecsys in preclinical AD. 42 In ROC analyses, the authors found that p-tau231 ADx had statistically significant higher predictive accuracies than p-tau181 Elecsys for discriminating Aβ-positive from Aβ-negative CU individuals. Moreover, p-tau231 ADx showed an AUC that Levels of CSF p-tau181 Innotest (A), p-tau181 Lilly (B), p-tau181 Elecsys (C), p-tau217 Lilly (D), and p-tau231 ADx (E) are expressed relative to the mean of β-amyloid (Aβ)-negative participants (n = 253). Solid gray horizontal lines indicate age-adjusted group comparisons: Alzheimer disease (AD) dementia higher than all groups (p < 0.001); Aβ-positive mild cognitive impairment higher than cognitively unimpaired (CU) and non-AD neurodegenerative disorders (p < 0.001); Aβpositive CU higher than Aβ-negative CU and non-AD (p < 0.001). In order to facilitate comparison between p-tau measures, y-axes were scaled to the maximum fold change seen across biomarkers. In order to facilitate comparison between p-tau measures, y-axes were scaled to the maximum fold change seen across biomarkers.
was higher than that of p-tau181 Elecsys . Although our findings showing that p-tau231 ADx had higher AUCs compared to p-tau181 Elecsys and p-tau181 Innotest are consistent with this, our results also suggest that p-tau231 ADx is similar to p-tau181 Lilly . In a related study by Karikari et al., 43 N-p-tau217 showed higher diagnostic performance for identifying Aβ pathology and AD at the MCI stage compared to established p-tau181 assays (p-tau181 Innotest and p-tau181 Lumipulse ), but not compared to N-p-tau181. Possibly complicating this comparison, however, is the comparatively small number of prodromal AD cases. Although our results cannot be directly compared because of differences in the assays used for p-tau217, studies thus far suggest, overall, that p-tau217 assays are generally more sensitive. Further studies directly comparing these assays are required, however, as well as whether p-tau231 ADx and p-tau181 Lilly begin to increase at the same time point or if p-tau231 ADx starts to increase earlier in order to help establish the temporal dynamics of these different measures. Here,   longitudinal studies comparing p-tau181, p-tau217, and p-tau231 will prove crucial.
Clinical utility in terms of fold change with respect to levels in Aβ-negative CU individuals varied across the investigated CSF p-tau measures. Although sharing the same p-tau181specific antibody (AT270), the Innotest and Lilly p-tau181 assays had different total tau (i.e., not binding to the phosphorylation site) antibodies and showed large differences in fold change. This indicates the importance of tau isoforms or fragmentation with respect to clinical utility: should fragmentation occur in the region of the protein where the 2 totaltau antibodies bind, this could lead to the measurement of different pools of tau present in CSF. This hypothesis is reinforced by the fact that p-tau181 Innotest and p-tau181 Elecsys assays showed similar performance despite differing in both p-tau181-specific and total tau antibodies; presumably, this reflects these assays measuring the same tau isoform/fragment. The influence of the total tau antibody on the clinical utility of p-tau181 is thus significant, as p-tau181 Lilly showed about double the fold change as p-tau181 Innotest and p-tau181 Elecsys in AD.
One explanation for the differences in fold change is also the possibility of a different binding affinity of the total tau antibody for tau, which could lead to differences in the measured signal with increasing protein concentrations. However, the Lilly assays showed that p-tau217 showed greater fold change compared to p-tau181 when using the same total tau antibody in combination with different phosphorylation-specific antibodies. The variability observed in the fold change of the measurements with p-tau181 assay cannot therefore be fully explained by technical differences of the assay or antibody affinity. One could speculate that the binding of antibodies to different phospho-epitopes could lead to conformational changes in the protein and therefore different affinity of the total tau antibody, but further studies are needed to demonstrate this. Similarly, we do not yet know whether p-tau231 will be better or worse in a head-to-head comparison to p-tau217 as the p-tau231 ADx assay uses a different total tau antibody.
Recent mass spectrometry-based work 44 addressing tau PTMs has shown that p-tau181, p-tau217, and p-tau231 appear to be indicators of early AD pathology based on Braak NFT staging of postmortem brain tissue. 18 In a related study exploring the biochemical link between measures of Aβ and tau phosphorylation, however, a somewhat different conclusion was reached: while soluble p-tau181, p-tau217, and p-tau231 were highly correlated to Aβ levels, 45 the highest degree of tau phosphorylation was observed in the insoluble fractions of AD brain tissue, suggesting that correlations with Aβ and tau aggregates may be more complex than simple linear relations. Despite this recent progress in understanding the link between abnormal PTMs and the aggregation of tau in AD, additional studies are required to understand how such abnormal PTMs are reflected in predominantly C-terminally truncated tau. 8,46 Current findings nevertheless highlight the importance of mapping PTMs in order to better understand the pathophysiology of AD; moreover, increased insight into the role of PTMs will facilitate the identification of novel therapeutic targets and improve AD diagnostics.
Strengths of our study include the large number of patients spanning the AD continuum, within-subject measurements of multiple CSF p-tau isoforms and their comparison to the widely used p-tau181 from Innotest and Elecsys, and the availability of Aβ and tau-PET imaging. Moreover, the use of mid-fragments across all p-tau measures allowed for a more direct comparison of p-tau biomarkers. This study has limitations, however. First, our inferences as to the ordering of changes in p-tau isoforms over the course of AD are based on cross-sectional data whereas longitudinal studies are needed to accurately address this question. Second, we did not have Aβ-PET in the AD dementia group. Although earlier work indicated that Aβ pathology may reach a plateau during the dementia phase of AD, 47 recent findings suggest that this may not be the case. 48 As such, we were not able to examine the effect of higher Aβ-PET SUVR values on p-tau isoforms but were nonetheless able to identify the significantly higher dynamic range of p-tau217 Lilly using the available Aβ-PET from participants with out dementia. Although we acknowledge the lack of Aβ-PET across all groups as a limitation, very high concordance is seen between CSF Aβ42/Aβ40 and Aβ-PET. 49 As such, the 2 measures provide similar information with respect to defining Aβ status. As we were interested in the relationship between p-tau and the amount of fibrillary brain Aβ, however, we chose Aβ-PET as this measure reflects the cumulative burden of accumulated Aβ pathology while CSF Aβ42/40 reflects the production and clearance of Aβ42 and Aβ40 at a given time point. 50 Lastly, though our study used the same assay for p-tau231 ADx and p-tau181 Elecsys as used in the study by Suárez-Calvet et al., 42 assays for p-tau181 and p-tau217 differed. In order to more definitively address the ordering of p-tau biomarkers, future work comparing phosphorylation epitopes will require the use of assays that are as similar as possible using head-to-head designs and validation in independent datasets.
We found that CSF p-tau217 Lilly more strongly correlated with Aβ and tau-PET and showed greater increases as compared to p-tau181 Innotest , p-tau181 Lilly , p-tau181 Elecsys , and p-tau231 ADx in AD dementia and across tau-PET Braak stages. Moreover, CSF p-tau217 Lilly showed greater discriminative accuracy for AD dementia, as compared to CSF p-tau181 Innotest , p-tau181 Lilly , p-tau181 Elecsys , and p-tau231 ADx . These results suggest that CSF p-tau217 Lilly should be the preferred p-tau variant to use for AD diagnostics and for tracking disease progression (e.g., as an outcome in clinical AD trials).