CSF tau368/total-tau ratio reflects cognitive performance and neocortical tau better compared to p-tau181 and p-tau217 in cognitively impaired individuals

We included individuals from the Translational Biomarkers of Aging and Dementia (TRIAD) cohort, McGill University, Canada. In the TRIAD cohort, all participants had CSF and PET (amyloid and tau) biomarkers and detailed clinical and cognitive assessments, including Mini-Mental State Examination (MMSE), Montreal Cognitive Assessment (MoCA), and the clinical dementia rating (CDR) tests. The TRIAD cohort consisted of cognitively unimpaired (CU) young (median age, years [IQR]) (23.1 [22.7–24.1]) and CU elderly (72.5 [67.7–76.7]) as well as mild cognitive impairment (MCI), AD, and non-AD dementia patients. CU participants had an MMSE score > 24 and a CDR score of 0. MCI participants had a CDR score of 0.5, subjective and objective impairments in cognition, but preserved activities of daily living. AD dementia patients had a CDR score ≥ 0.5 and met the National Institute on Aging and the Alzheimer’s Association criteria for probable Alzheimer’s disease determined by a physician [16] and were Aβ PET positive. The non-AD participants had CDR score ≥ 0.5, were Aβ PET-negative, and had clinical diagnosis of FTD (n = 7), progressive supranuclear palsy (n = 1), MCI (n = 12), or clinically diagnosed AD (n = 2). In the TRIAD cohort, participants were excluded if they had active substance abuse or inadequately treated conditions, recent head trauma or major surgery, or if they presented safety contraindication for the study procedures.

The commercial fully automated LUMIPULSE G1200 (Fujirebio) was used to measure CSF p-tau181, t-tau, as previously described [17]. CSF p-tau217 was measured using an in-house single molecule array (Simoa) assay, as detailed in Karikari et. al. [18] Tau368 was measured using a validated in-house Simoa assay, which has been detailed elsewhere [15]. Briefly, an anti-tau368 antibody was used as capture antibody, whereas K9JA (a rabbit polyclonal antibody against 243-441, Sigma) was used as detector. All biochemical analyses were performed at the Clinical Neurochemistry Laboratory at the Sahlgrenska University hospital, Mölndal, Sweden. Of 125 individuals, CSF p-tau217 was available for 116 of the participants, while the other biomarkers were available for all individuals.

All individuals in the TRIAD cohort were assessed with Siemens 3T MRI as well as Aβ [¹⁸F]-AZD4694 PET and tau [¹⁸F]-MK-6240 PET acquired with a Siemens High Resolution Research Tomograph. [¹⁸F]-MK-6240 images were acquired at 90–110 min after the intravenous bolus injection of the radiotracer [19, 20]. Aβ [¹⁸F]-AZD4694 PET images were acquired at 40–70 min after the intravenous bolus injection of the radiotracer [19, 20]. The PET images were spatially smoothed to achieve a final 8-mm full width at half maximum resolution and were processed using a previously described pipeline [19, 20]. [¹⁸F]-MK6240 images were stripped off the meninges before smoothing, as described elsewhere [20]. [¹⁸F]-MK6240 standard uptake value ratio (SUVR) was measured regionally based on the anatomical brain regions proposed by Braak and Braak as follows: Braak I (transentorhinal), Braak II (entorhinal and hippocampus), Braak III (amygdala, parahippocampal gyrus, fusiform gyrus, lingual gyrus), Braak IV (insula, inferior temporal, lateral temporal, posterior cingulate, and inferior parietal), Braak V (orbitofrontal, superior temporal, inferior frontal, cuneus, anterior cingulate, supramarginal gyrus, lateral occipital, precuneus, superior parietal, superior frontal, rostro medial frontal), and Braak VI (paracentral, postcentral, precentral, and pericalcarine). In this study, the six Braak stages were collapsed to three main stages (as proposed by Braak and Braak) [21]: transentorhinal, limbic, and isocortical regions, approximating Braak stages I–II, III–IV, and V–VI of neurofibrillary tangle (NFT) pathology, respectively. The Desikan-Killiany-Tourville atlas [22] was used to define the regions-of-interest for the PET Braak-like stages. Tau positivity was defined as 2.5 standard deviations (SD) higher than the mean SUVR of cognitively unimpaired young individuals for each respective region of interest. Determination of individual Braak staging was performed using an automatic pipeline in a hierarchical fashion, where later stages can only be achieved if the individual is positive for the previous stages; otherwise, the participant was considered Braak stages discordant. Discordant individuals were not included in the analyses performed in this paper (n = 12) presented in Fig. 3, whereas they were included in all other analyses. Individuals negative for tau PET uptake in all aforementioned regions-of-interest were classified as in vivo Braak stage 0. Further descriptions of the in vivo Braak-like staging can be found elsewhere [20]. Global [¹⁸F]-AZD4694 SUVR was derived from averaging retention in the precuneus, the cingulate, inferior parietal, medial prefrontal, lateral temporal, and orbitofrontal cortices. Aβ [¹⁸F]-AZD4694 SUVR positivity was based on visual assessment [23]. Structural MRI data were acquired using a Siemens 3T scanner using a standard head coil. Hippocampal volume was assessed using FreeSurfer version 6.0 using the Desikian–Killiany–Touriner atlas gray matter segmentation. Hippocampal volume was adjusted for intracranial volume.

Normality was tested by determining kurtosis and skewness, as well as with the Shapiro-Wilks test. All fluid biomarkers except for tau368/t-tau were non-normally distributed. Thus, non-parametric methods were selected. In analyses of only AD dementia and prodromal AD, tau368/t-tau, tau, and amyloid PET SUVR measures were roughly normally distributed (skewness and kurtosis ± 1), but the other fluid biomarker measures were log-transformed₁₀ due to non-normality in linear regression models. Group-wise comparisons of continuous variables were performed using Kruskal-Wallis test or Mann-Whitney U test where appropriate and were corrected for multiple comparisons using false discovery rate (FDR). Fisher’s exact test was used to compare categorical variables across groups. Correlations between biomarkers and other continuous variables were tested with Spearman rank correlation. Linear regression models were used to evaluate whether amyloid influenced the association between tau PET and tau368/t-tau. These models had CSF tau368/t-tau as the response variable and had tau PET alone or tau PET and amyloid PET as predictors. Akaike information criteria with a correction for small sample sizes (AICc; used when n divided by number of parameters are less than about 40) was used to assess the model fit, with a penalty for a more complex model [24]. Areas under the curve (AUC) was assessed using receiver operating characteristics (ROC) analysis. Differences in AUCs were evaluated with bootstrapping (n = 2000), using the pROC package in R [25]. In the comparisons including p-tau217, only individuals with CSF p-tau217 measured were included. Z-scores of the fluid biomarkers were calculated using the CU elderly individuals in Braak stage 0 as a reference. Statistical analyses were performed using GraphPad Prism v.9.2.0 (La Jolla, CA, USA) and R Statistical Software (https://​www.​r-project.​org/​).

We studied 125 (72 [58%] women) individuals (healthy young individuals [n = 21], CU Aβ− [n = 45] CU Aβ+ [n = 15], MCI Aβ+ [n = 10; henceforth used interchangeably with prodromal AD], AD dementia [n = 14], non-AD cognitive impairment [n = 20]). Participant’s characteristics are summarized in Table 1. There was a significant association between age and t-tau (Spearman rho (ρ) = 0.37, P < 0.001), p-tau217 (ρ = 0.30, P = 0.001), p-tau181 (ρ = 0.39, P < 0.001), and tau368 (ρ = 0.45, P < 0.001), but not tau368/t-tau. After excluding the young individuals, only tau368 remained significantly associated with age (ρ = 0.23, P < 0.05). There were no significant sex-differences in tau368 or in tau368/t-tau. Tau368/t-tau was lower in APOE ε4 carriers as compared to non-carriers (P < 0.05). In addition, tau368 correlated both with both t-tau, p-tau217, and p-tau181 in all individuals, as well as when stratified into participants with symptomatic AD vs all others (all ρ = 0.6–0.9, P < 0.001).

For tau368, higher concentrations were seen in all groups as compared to young healthy individuals, but no other significant group-differences were seen (Supplementary Figure 1). However, when using tau368 in a ratio with t-tau, individuals with AD dementia had a significantly lower ratio than CU elderly individuals, with (P < 0.01) or without (P < 0.001) amyloid pathology, as well as compared to those with non-AD cognitive disorders (P < 0.001; Fig 1A). A decrease could be observed also in prodromal AD, when comparing with CU- elderly (P < 0.01) and non-AD cognitive disorders (P < 0.01). Tau368/t-tau separated AD dementia vs non-AD with an area under the curve (AUC) of 0.84 (95% CI 0.68–0.99) and prodromal AD vs. non-AD with and AUC of 0.82 (95% CI 0.66–0.98). The respective accuracies were numerically higher than t-tau alone (AD dementia vs. non-AD 0.82, 95% CI 0.67–0.97; prodromal AD vs non-AD: 0.80, 95% CI 0.63–0.97) but numerically lower than p-tau181 (AD dementia vs. non-AD 0.92, 95% CI 0.8–1.0; prodromal AD vs non-AD: 0.93, 95% CI 0.83–1.0) and p-tau217 (AD dementia vs. non-AD 0.95, 95% CI 0.85–1.0 and prodromal AD vs non-AD: 0.94, 95% CI 0.84–1.0). No statistical differences between the respective AUCs were seen (all P >0.05).

As it has been hypothesized that tau368/t-tau well reflects tau accumulation, we evaluated its relationship with tau PET and then compared it with the associations between tau PET and t-tau, p-tau181, and p-tau217. When investigating the continuous relationship between the SUVR of [¹⁸F]-MK-6240 PET in in vivo Braak-like regions and fluid biomarkers, as seen in previous studies [4], all biomarkers correlated moderately to strongly with tau PET uptake in the whole group. The strongest associations were seen for CSF p-tau181 and p-tau217 in all in vivo Braak-like regions (Supplementary Table 1). However, when stratifying the groups into symptomatic AD, CU, and other CI, we found that tau368/t-tau was increasingly associated with tau PET SUVR in brain regions commonly being affected later by fibrillar tau deposition in participants with symptomatic AD (limbic regions: ρ = − 0.58, P < 0.01; isocortical regions: ρ = − 0.67, P < 0.001), with no strong associations observed between tau368/t-tau and tau PET SUVR in any of the brain regions in the groups encompassing other participants (Fig. 2A–C). Regarding the associations of other CSF tau biomarkers with tau PET in the symptomatic AD group, p-tau181 and t-tau only associated with SUVR in the transentorhinal (corresponding to in vivo Braak I–II) composite region (p-tau181: ρ = 0.48, P < 0.05, t-tau: ρ = 0.47, P < 0.05), while p-tau217 was not associated with SUVR in any of the tau PET brain regions (Fig. 2D–L). Further, tau368 when instead used in a ratio with p-tau181 presented very similar findings as compared with tau368/t-tau, whereas no significant associations were seen for tau368/p-tau217 (Supplementary Table 2). Interestingly, among these participants, there were strong correlations between lower tau368/t-tau and younger age. Similar associations were seen between limbic and isocortical tau as indexed with tau PET (Supplementary Figure 2). Further, when evaluating the associations of CSF tau biomarkers with tau PET SUVR in the groups encompassing the other participants (all CU, other CI), no strong significant associations were observed for tau368/t-tau, while several significant associations were observed for p-tau181, p-tau217, and t-tau.

When using linear models to investigate whether these associations observed in the symptomatic AD group could be driven by Aβ pathology, we found that the relationship between tau368/t-tau and tau PET signal in limbic regions was not affected when amyloid PET SUVr was included in linear regression models (Table 2; corresponding tables for p-tau181, p-tau217, and t-tau are available in Supplementary Tables 3-5).

Furthermore, we investigated the trajectories of tau368/t-tau across in vivo Braak-like brain stages, by comparing the Z-scored CSF tau biomarkers of individuals in that were considered tau PET positive in transentorhinal, limbic, and isocortical regions, respectively. There were pronounced increases between the transentorhinal and limbic stages especially for p-tau181 and p-tau217 (Fig. 3C, D). Although there was no statistically significant change between individuals in the limbic vs. the isocortical stages, tau368/t-tau seemingly continues to become more abnormal as the tau pathology progresses, unlike p-tau217 and 181 and t-tau, which plateau as the tau pathology becomes widespread across the neocortex (Fig. 3A–D).

Regarding the discriminative ability of tau368/t-tau for tangle pathology, it identified tau positivity in transentorhinal regions (Braak I-II) with an AUC of 0.85 (95% CI 0.77-0.91), as compared to 0.93 (95% CI 0.88–0.98), 0.94 (95% CI 0.89–0.99), and 0.87 (95% CI 0.80–0.93) for p-tau181, p-tau217, and t-tau, respectively (Supplementary Figure 3). This was significantly lower as compared to p-tau181 (P < 0.05; p-tau217, P = 0.06; t-tau, P = 0.48). Individuals with tau PET positivity at least in limbic regions (Braak III–IV) were identified with an accuracy of 0.92 (0.86–0.97) using tau368/t-tau, with AUCs being 0.98 (95% CI 0.95–1.0; P < 0.05 vs. tau368/t-tau), 0.99 (95% CI 0.97–1.0; P < 0.05 vs. tau368/t-tau), and 0.93 (95% CI 0.88–0.98; P = 0.47 vs. tau368/t-tau) for p-tau181, p-tau217, and t-tau. In the isocortical stages (Braak V-VI), however, tau368/t-tau (95% CI 0.93, 0.87–0.98) performed similarly (all P > 0.05 vs tau368/t-tau) as compared to p-tau181 (0.93, 95% CI 0.88–0.98), p-tau217 (0.94, 95% CI 0.90–0.99), and t-tau (0.89, 95% CI 0.83–0.95). Of note, there was no association between tau368/t-tau and hippocampal volume (Supplementary Figure 4).

As tau368/t-tau was significantly associated with tau PET tracer uptake in brain regions associated with more advanced tau pathology, we hypothesized that it may have a closer relationship with cognitive performance than core CSF biomarkers. Using MoCA as a measure of global cognition, there were expected correlations in the whole group with all CSF biomarkers, being the strongest for p-tau181 and p-tau217 (ρ = − 0.52 and ρ = − 0.55, respectively) (Supplementary Table 6). However, the degree of cognitive impairment was more accurately reflected by tau368/t-tau (ρ = 0.53, P < 0.01) than p-tau181 (ρ = 0.23, P = 0.27), p-tau217 (ρ = 0.23, P = 0.29), t-tau (ρ = 0.18, P = 0.39), and ratios of tau368/p-tau181 or 217 (Supplementary Table 2), in individuals with symptomatic AD (Fig. 4A–D).