KL-VS heterozygosity is associated with lower amyloid-dependent tau accumulation and memory impairment in Alzheimer’s disease

Klotho-VS heterozygosity (KL-VShet) is associated with reduced risk of Alzheimer’s disease (AD). However, whether KL-VShet is associated with lower levels of pathologic tau, i.e., the key AD pathology driving neurodegeneration and cognitive decline, is unknown. Here, we assessed the interaction between KL-VShet and levels of beta-amyloid, a key driver of tau pathology, on the levels of PET-assessed neurofibrillary tau in 551 controls and patients across the AD continuum. KL-VShet showed lower cross-sectional and longitudinal increase in tau-PET per unit increase in amyloid-PET when compared to that of non-carriers. This association of KL-VShet on tau-PET was stronger in Klotho mRNA-expressing brain regions mapped onto a gene expression atlas. KL-VShet was related to better memory functions in amyloid-positive participants and this association was mediated by lower tau-PET. Amyloid-PET levels did not differ between KL-VShet carriers versus non-carriers. Together, our findings provide evidence to suggest a protective role of KL-VShet against amyloid-related tau pathology and tau-related memory impairments in elderly humans at risk of AD dementia.

K lotho is a transmembrane protein that has been associated with enhanced longevity and better brain health in aging 1,2 . Klotho is expressed primarily in the kidney and brain, where it has been implicated in a number of vital cellular functions (for review see 3 ). Loss-of-function mutations in transgenic mice are associated with reduced Klotho protein expression, accelerated aging phenotypes, and dramatically shortened life span 1,4 . In humans, two variants in the Klotho gene (KL, 13q13.1), rs9536314 (F352V) and rs9527025 (C370S), form a functional haplotype. Carrying one copy, but not two copies of the KL-VS haplotype, referred to as KL-VS heterozygosity (KL-VS het ), has been previously linked to increased Klotho levels in the blood 5,6 . KL-VS het occurs in about 20-25% of the population 5 and is associated with higher cognitive performance across the adult life span 5,[7][8][9] , larger frontotemporal gray matter volume in cognitively normal individuals 8 , and lower mortality 6,10 . Together, these results suggest a crucial role of Klotho in the maintenance of cognitive abilities and brain integrity during aging.
Beyond the protective role of Klotho in normal aging, recent studies suggest an association between Klotho and reduced risk of Alzheimer's disease (AD) 11 , the most frequent cause of dementia in the elderly 12 . A recent meta-analysis reported that KL-VS het was associated with reduced AD dementia risk and cognitive decline in elderly individuals carrying the ApoE ε4 allele 13 , i.e., the strongest genetic risk factor for AD dementia possibly through elevated levels of primary AD pathology including cortical beta-amyloid (Aβ) aggregation 14,15 . Importantly, KL-VS het was associated with reduced biomarker levels of Aβ deposition in ApoE ε4 carriers 16 suggesting that KL-VS het may directly alter levels of primary AD pathology.
Yet, an open question is whether Klotho is associated with altered levels of fibrillary tangles containing pathologic tau, i.e., the key driver of disease progression in AD 17 . In the presence of Aβ deposition, i.e., the earliest primary pathology in AD 18,19 , neurofibrillary tangles spread from the medial temporal lobe to higher cortical areas [20][21][22] . The progressive development of neurofibrillary tangles in the presence of Aβ pathology is closely associated with gray matter atrophy [23][24][25] and cognitive worsening 20,[26][27][28] and is more predictive of such alterations than Aβ 29 . Due to the high clinical relevance of tau pathology, it is pivotal to understand whether the KL-VS het variant attenuates the accumulation of neurofibrillary tangles at a given level of Aβ deposition, and thus a cognitive decline in AD. Studies in mouse models of Aβ and accelerated aging reported that enhancing KL expression was associated with reduced Aβ burden and phosphorylated tau 11,30 , although conflicting results were reported as well 31 . However, these mouse models fail to develop neurofibrillary tangles in the presence of Aβ and thus only incompletely recapitulate AD-specific tau pathology in humans.
Here, we examined whether KL-VS het attenuates the association between higher Aβ and higher fibrillar tau assessed via positron emission tomography (PET) in a group of 551 elderly asymptomatic and symptomatic individuals recruited within a large North American multicenter study on AD 32 . We found the KL-VS het variant to be associated with an attenuated increase in regional tau-PET at pathological levels of global amyloid-PET, suggesting that KL-VS het was potentially protective against Aβ-related increase in neurofibrillary tangles. This association was pronounced in ApoE ε4 carriers. The strength of the KL-VS het effect on region-specific tau-PET levels was correlated with the regional expression pattern of KL in the brain 33,34 supporting the notion that the KL-VS het variant modulates the regional accumulation of tau pathology. Importantly, KL-VS het was associated with higher memory performance and this association was mediated by reduced tau-PET levels in KL-VS het carriers with the elevated amyloid-PET burden. For Aβ, we did not find the previously reported association between KL-VS het and lower Aβ pathology in the current sample, but confirmed this link in a larger sample including all individuals with amyloid-PET but not necessarily tau-PET assessment available indicating that the effect size of KL-VS het on Aβ was smaller than that on tau pathology.

Results
Detailed sample characteristics are presented in Table 1 KL-VS heterozygosity is associated with lower amyloid-related tau accumulation. In the main analysis, we tested the hypothesis that KL-VS het modifies the association between Aβ and tau pathology (both assessed by continuous measures of PET uptake). In a region-of-interest (ROI)-based analysis, we focused on tau-PET in the inferior temporal cortex (i.e., ROI of early Aβ-related tau pathology 20,26,[35][36][37] ) and whole-brain tau-PET levels. The results of a linear regression analysis showed a significant KL-VS het × amyloid-PET interaction effect on tau-PET levels in both the inferior temporal ROI (standardized beta = −0.12, p = 0.009, N = 551, effect size measured by Cohen's f = 0.112) and the global ROI (beta = −0.13, p = 0.008, N = 551, Cohen's f = 0.114). For both tau-PET ROIs, the increase in tau-PET as a function of rising global amyloid-PET was attenuated in KL-VS het carriers versus non-carriers (Fig. 1a, b). The main effects of amyloid-on tau-PET for KL-VS het carriers and noncarriers are reported in Supplementary Table 1. All analyses were controlled for the main effects of the interaction terms, age, sex, diagnosis, education, and ApoE ε4 carrier status. Next, we performed several secondary analyses. To ensure that our results were not driven by unequally sized KL-VS het groups, we repeated the main analyses in all 144 KL-VS het carriers and 144 out of 407 non-carriers who were selected based on propensity score matching for global amyloid-PET levels and diagnosis. Comparable KL-VS het × amyloid-PET interaction effects were found on tau-PET levels in both ROIs (inferior temporal ROI: beta = −0.20, p = 0.005, N = 288, Cohen's f = 0.168; global ROI: beta = −0.21, p = 0.010, N = 288, Cohen's f = 0.156; Supplementary Fig. 1a, b). In addition, we repeated the interaction analysis in 1000 bootstrapped samples (i.e., random sampling from the participant pool with replacement). As a reference, we also generated a null distribution by randomly reshuffling the KL-VS het labels on each iteration. The bootstrapped mean t-value of the interaction effect differed significantly from that of the null distribution (inferior temporal ROI: t(1901.9) = −47.43, p < 0.001; global ROI: t(1839.2) = −48.99, p < 0.001; Supplementary Fig. 2). Only the distribution of t-values based on the actual, unshuffled KL-VS het labels was significantly greater than zero (inferior temporal ROI: t(999) = −77.76, p < 0.001; global ROI: t(999) = −83.88, p < 0.001) and the 95% confidence intervals did not include zero (inferior temporal ROI: 95% CI = [−4.847, −0.563]; global ROI: 95% CI = [−4.763, −0.794]). Together, these results confirmed a robust association between KL-VS het and lower Aβ-associated tau accumulation.
A few participants showed lower tau-PET levels in the ROIs than in the reference region resulting in a tau-PET standard uptake value ratio (SUVR) < 1 (2 participants for the inferior temporal ROI and 51 participants for the global ROI). Yet, the results of the KL-VS het × amyloid interaction analyses on tau-PET levels remained significant after excluding those participants ( Supplementary Fig. 1g, f). KL-VS heterozygosity is related to lower amyloid-dependent tau accumulation over time. In a subsample of 200 participants in whom longitudinal tau-PET data were available, we investigated whether KL-VS het attenuates the association between baseline amyloid-PET levels and the rate of change in tau-PET assessed over a time interval of 1.63 years on average (range: 1-4 years). We found a significant KL-VS het × amyloid-PET interaction effect on tau-PET annual change rates in the inferior temporal ROI (beta = −0.22, p = 0.039, N = 200, Cohen's f = 0.148), but not in the global ROI (beta = −0.15, p = 0.176, N = 200, Cohen's f = 0.098). KL-VS het carriers showed lower tau-PET increases in inferior temporal cortices over time as a function of rising global amyloid-PET levels (Fig. 1c, d) suggesting that the KL-VS het variant might be protective against Aβassociated increase of tau pathology. The main effects of amyloid-PET on tau-PET change rates for KL-VS het carriers and non-carriers are reported in Supplementary Table 1.
Spatial match between KL mRNA expression and the effect of KL-VS heterozygosity on tau-PET. In order to estimate the spatial overlap between the strength of KL gene expression and the test statistic of the KL-VS het × amyloid-PET interaction on tau-PET, we obtained whole-brain mRNA expression levels of KL generated by post-mortem microarray assessments of six healthy brain donors and subsequently mapped to the Allen Brain Atlas 33,34 . We computed median scores of log2 mRNA expression of KL across the six donors within 34 left-hemispheric regions of the Freesurfer-based Desikan-Killiany brain atlas 38 . We focused on the left hemisphere since all donors had microarray assessment available for the left hemisphere and only two donors had an assessment for the right hemisphere. Furthermore, we estimated the KL-VS het × amyloid-PET interaction effect on tau-PET levels within the same 34 brain atlas regions using the aforementioned regression model. Surface mapping of both the KL-VS het × amyloid-PET interaction effect (which were all in the same direction) and KL mRNA expression is displayed in Fig. 2a-d. Spatial correlation analysis revealed a significant association (r = 0.46, p = 0.007; Fig. 2e). This result suggests that regions with higher KL mRNA expression levels were more likely to display lower Aβ-related tau-PET levels in KL-VS het carriers versus non-carriers. Visual inspection of the thresholded spatial maps indicated that those areas showing both a significant KL-VS het × amyloid-PET interaction effect ( Fig. 2b; see Supplementary Table 2 for detailed statistical results) and high KL mRNA expression (log2 > 75th percentile) ( Fig. 2d) were specifically located within the mesiotemporal and inferior and middle temporal brain regions and the posterior cingulum.  Table 2 for details statistical results). b Thresholded spatial map color-code only regions with a significant (p < 0.01, unadjusted for multiple comparisons) KL-VS het × amyloid-PET interaction effect. c Surface mapping of median KL mRNA expression (i.e., log2 derived from the Allen Brain Atlas) within the identical 34 atlas regions. Yellow colors indicate higher KL mRNA expression. d Thresholded spatial maps restricted to regions falling above the 75th percentile of KL mRNA expression. e Scatterplot depicting the association between ROI-based KL mRNA expression and KL-VS het × amyloid-PET interaction effect on regional tau-PET uptake. Statistical results are derived from the Pearson correlation (two-sided). Linear model fits are indicated together with 95% confidence intervals. Source data are provided as a Source Data file. ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-021-23755-z Tau mediates the association between KL-VS heterozygosity and less memory impairment. In the main analysis, we assessed whether KL-VS het is beneficial for memory functions via lowering tau pathology. Because the interaction effect of KL-VS het × amyloid-PET on tau-PET levels showed that KL-VS het is associated with lower tau accumulation at higher levels of amyloid-PET, we restricted our analysis to amyloid-positive participants. We used mediation analysis with 10,000 bootstrapping iterations in order to test whether KL-VS het is associated with better memory in individuals with elevated Aβ burden and whether this effect is mediated via reduced global tau-PET levels. Memory performance was measured by an established composite score based on participant's results across multiple different memory tests (ADNI-MEM) 39 . The mediation analysis was controlled for age, sex, education, diagnosis, ApoE ε4 status, and global amyloid-PET levels. Supporting our hypothesis, we found KL-VS het to be associated with higher ADNI-MEM scores (beta = 0.13, p = 0.040, Cohen's f = 0.104, N = 229, Fig. 3) and that this relationship was significantly mediated by lower tau-PET levels (bootstrapped average causal mediation effect: beta = 0.06, 95% CI = 0.01-0.12, p = 0.013, N = 229). This result indicates that, in individuals with an elevated Aβ burden, KL-VS het carriers showed less impaired episodic-memory abilities when compared to KL-VS het non-carriers due to lower tau-PET levels in KL-VS het carriers. A path model of the mediation analysis is shown in Fig. 4.
In the secondary analysis, we addressed the question of whether KL-VS het exerts a beneficial influence on memory via lowered neurofibrillary tau in the whole sample without stratification based on amyloid-PET levels. To this end, we estimated the interaction effect between KL-VS het and amyloid-PET on ADNI-MEM scores, with and without controlling for global tau-PET levels as a covariate. We reasoned that if the association between KL-VS het on cognition is mediated by tau-PET, the interaction between KL-VS het and amyloid-PET on cognition should be diminished when controlling for tau-PET. Without controlling tau-PET levels, we found a significant KL-VS het × amyloid-PET interaction effect on memory functions (beta = 0.08, p = 0.037, N = 549, Cohen's f = 0.090). Specifically, we observed that individuals with a high amyloid-PET burden showed better cognitive performance when being KL-VS het carriers compared to non-carriers ( Supplementary Fig. 4). As hypothesized, the KL-VS het × amyloid-PET interaction on memory no longer reached significance level when global tau-PET levels were controlled for (beta = 0.06, p = 0.125, N = 549).
Is KL-VS heterozygosity associated with lower Aβ accumulation? A recent study found a protective influence of KL-VS het on longitudinal amyloid-PET in cognitively unimpaired ApoE ε4 carriers aged between 60 and 80 years, but not in ApoE ε4 non-carriers or older participants 13 . In contrast to this earlier report, we did not find an age-dependent KL-VS het effect on cross-sectional amyloid-PET levels acquired at the time of tau-PET assessment in the current sample (KL-VS het × age interaction: beta = 0.21, p = 0.597, N = 551; Supplementary Fig. 6a) Fig. 6b). However, more subtle effects may have been overlooked in the current more restricted sample of individuals undergoing both amyloid-and tau-PET. Therefore, in a supplementary analysis, we included all participants with amyloid-PET (N = 1067) from ADNI, regardless of whether or not they underwent tau-PET assessment. We found a trend-level significant KL-VS het × age interaction effect one amyloid-PET that demonstrated that KL-VS het carriers in the lower age range (<80 years) displayed lower amyloid-PET levels than non-carriers (beta = 0.53, p = 0.046, Cohen's f = 0.061, N = 1067; Supplementary Fig. 7a). Consistent with the earlier report, we found a significant KL-VS het × ApoE ε4 interaction effect on global amyloid-PET levels in CN participants aged between 60 and 80 years (beta = −0.121, p = 0.043, Cohen's f = 0.095, N = 464; Supplementary  Fig. 7b). The same analysis in MCI participants within the same age range showed no significant interaction effect (beta = 0.02, p = 0.780, N = 463, Cohen's f = 0.013), suggesting that the association between KL-VS het and lower amyloid-PET uptake is restricted to a younger age and non-symptomatic cognitive status. See Supplementary  Table 3 for detailed sample characteristics of the larger ADNI amyloid-PET sample compared to the current ADNI tau-PET sample. Thus, our results in the larger sample are consistent with those from Belloy et al.'s analysis on the effect of KL-VS het on amyloid-PET stratified by age and ApoE genotype while also showing that the effect size of KL-VS het on tau-PET is stronger than that on amyloid-PET.

Discussion
The heterozygous KL gene variant KL-VS het has been previously associated with higher longevity and cognition performance in adulthood and reduced AD dementia risk 13 . We demonstrate that elderly KL-VS het carriers with elevated Aβ burden, i.e., the earliest primary AD pathology, exhibited lower tau-PET levels and tau-PET annual change rates when compared to those in KL-VS het noncarriers. In amyloid-positive participants, the KL-VS het variant was associated with better memory performance, and this relationship was mediated by lower tau-PET levels, suggesting that lower levels of pathologic tau in the KL-VS het carriers explained the association between KL-VS het and better memory performance. Although our findings do not implicate a causative mechanism of Klotho in AD, we provide evidence for a potential protective role of KL-VS het against Aβ-dependent tau pathology that is the key AD brain alteration linked to cognitive impairment.
To our knowledge, the current study is the first to date that evaluated the interaction between KL-VS het and Aβ on tau accumulation and cognitive decline in humans. There is a growing literature on protective genetic variants in AD 40-42 , but only a few studies have reported genetic variants to be associated with lower tau pathology in AD 43 . For the KL-VS het variant, previous studies reported an association with reduced Aβ accumulation in elderly ApoE ε4 risk-carriers 13,16 . We extend these previous findings by showing that the relationship between Aβ accumulation and fibrillar tau is modulated by KL-VS het , such that lower local and global tau-PET levels were observed per unit increase of global amyloid-PET burden in KL-VS het carriers when compared to those in non-carriers. This is important because Aβ deposition precedes the development of dementia symptoms by up to 20 years 14 , and as confirmed by a very recent longitudinal amyloid/tau-PET study, high baseline Aβ is associated with subsequent tau accumulation, while Aβ and tau in synergy lead to most pronounced subsequent cognitive decline 21 . The region showing one of the strongest interaction effects between KL-VS het and amyloid-PET on tau-PET was the inferior temporal gyrus (Fig. 2a), a brain area that typically shows an early Aβ-related increase in tau-PET 26 before elevated tau-PET levels extend to other higher cortical brain areas 20 . The protective association between KL-VS het and tau-PET was present selectively in participants with abnormally elevated levels of amyloid-PET and more pronounced in ApoE ε4 carriers. Stratified analyses further revealed a significant KL-VS het effect in the MCI but not in the CN subgroup, which could potentially be due to a stagedependent beneficial effect of Klotho. However, an alternative explanation is that the levels of both amyloid-and tau-PET are lower in CN compared to those in MCI, and thus any protective effect is likely to be of smaller size and would require a larger sample size to detect. Together, these results support the notion that KL-VS het is associated with an Aβ-related rather than agerelated reduction of tau pathology.
In amyloid-positive individuals, we found KL-VS het to be associated with better memory performance, mediated by the effect of KL-VS het on tau-PET. Our results are broadly consistent with those from studies on healthy aging, reporting KL-VS het to be associated with better cognition 5,8-10 , and lower risk of conversion from cognitively normal to mild cognitive impairment or AD dementia in ApoE ε4 carriers 13,16 . Our findings suggest that the association between KL-VS het and lower neurofibrillary tau pathology is of central importance for the association found between KL-VS het and less cognitive impairment. A previously reported absence of an association between KL-VS het and cognitive decline in asymptomatic participants with elevated levels of Aβ 44 did not assess the presence of abnormal neurofibrillary tau, which may have hampered to detect an effect of KL-VS het on cognitive decline in subjects at risk of AD 35 .
Previous studies reported KL-VS het to be associated with lower amyloid-PET in ApoE ε4 carriers (but not in ApoE ε4 non-carriers), which was strongest in the age range between 60 and 80 years 13,16 . In our primary analysis, we did not confirm age-or ApoE ε4-dependent effects of KL-VS het on tau-PET. By investigating a larger sample of all participants with available amyloid-PET regardless of the availability of tau-PET (N = 1067), we were able to substantiate those earlier findings 13,16 . Specifically, we showed reduced amyloid-PET burden in younger KL-VS het carriers (<80 years) and, in accordance with previous work, this association was mainly driven by cognitively unimpaired ApoE ε4 carriers rather than non-carriers or MCI patients. Comparing effect sizes, Cohen's f = 0.061 for the association between KL-VS het and lower amyloid-PET versus f = 0.114 for the association with lower tau-PET, strengthens the important role of changes in tau pathology for understanding the role of Klotho in AD.
The mechanisms linking Klotho to tau pathology remain elusive. Klotho is a pleiotropic protein that has been implicated in multiple biological processes including insulin regulation 4 , growth factor functions, in particular of FGF23 45 , regulation of members of the redox system 46 , and calcium signaling 47 . One possibility of how the Klotho protein might be linked to reduced neurofibrillary tau is its involvement in autophagy 48 , a mechanism that is involved in the clearance of AD pathologies 49 . Lentiviral overexpression of Klotho protein in an APP-PS1 mouse model of Aβ deposition reduced Aβ plaque load in aged mice and rescued the impaired autophagy possibly by modulating the Akt/ mTOR pathway 11,50 . Since APP-PS1 mice do not develop tau pathology, it remains, however, to be tested whether Klothoinduced autophagy reduces tau pathology. Those mechanistic explanations remain speculative at this point and the current work encourages future studies to investigate the mechanism that could underlie the protection Klotho exerts against the development of Aβ-related tau pathology.
Our findings of the spatial correspondence between the strength of the effect of KL-VS het on regional tau-PET and the spatial distribution of KL mRNA suggest a local effect of Klotho on the development of fibrillar tau, especially in temporal brain areas. Alternative splicing of the human KL mRNA results in both a membrane-bound and a secreted transcript of Klotho 1,4 , indicating that Klotho may act both in a cell-autonomous manner and as a humoral factor. Therefore, differences in gene expression in KL in the brain and/or different circulating levels of Klotho linked to KL-VS het5 may influence the development of pathological tau 11 , but this link remains to be investigated.
Our results have important implications for clinical trials in AD. Since tau pathology correlates more closely with clinical symptoms than Aβ, tau-targeted therapies seem a promising approach to arrest disease progression 51 . The common KL-VS genotype may inform those clinical trials that target tau pathology. Especially when anti-tau trials aim to include amyloid-positive or ApoE ε4 carrying participants, group differences in the KL-VS het variant may be taken into account when estimating the expected change in tau pathology over time, which would be useful in the computation of statistical power to detect a treatment effect. Furthermore, the current findings encourage future studies to test whether enhancing Klotho protein levels could reduce the development of tau pathology in amyloidpositive participants. The Klotho protein is druggable and could thus be made a target in the development of disease-modifying therapeutic approaches.
Several caveats should be considered when interpreting the current results. First, the human KL gene consists of three polymorphic variants. We decided to focus on the KL-VS haplotype given the existing evidence of its beneficial influence on Aβ and cognition in both mice and humans 5,[8][9][10]13,52 . While the second variant C1818T (rs564481) is located on the fourth exon and likely has no functional consequences itself, the third variant G395A (rs1207568) is located in the promoter region and may be a potential regulatory site of KL. The two latter variants appear more frequent in Asian populations, where they have been linked to cardiovascular risk factors 53 . Related to the current research question, an investigation across three independent cohorts of oldest-old Danes found different polymorphic variants of KL, besides KL-VS, to be associated with better cognitive functions 7 . It has yet to be proven whether these other KL variants also support resilience in AD. Another caveat is that we did not measure Klotho protein levels in the serum or CSF. Circulating levels of Klotho decrease during aging 54 and are associated with cognitive performance 5 and gray matter volume 55 in cognitively unimpaired individuals. In patients with AD, CSF levels of Klotho are reduced 52 , where the experimental reversal of reduced Klotho expression in transgenic mouse models exerted beneficial effects on Aβ and cognition 8,11,30 . While the KL-VS het variant has been associated with higher circulating levels of Klotho 55 , it remains to be investigated whether the association between KL-VS het and pathological tau are mediated by higher protein levels in the CSF and brain tissue.
In summary, our findings revealed a protective association of KL-VS het on tau accumulation that particularly manifested in amyloid-positive individuals, where lower tau pathology was related to better cognitive functions. These findings may be particularly informative for clinical anti-tau trials 56 and may encourage future studies on enhancing Klotho protein levels as a therapeutic intervention to slow down the development of tau pathology and dementia in AD.

Methods
Sample characteristics. A total of 551 participants were selected from ADNI phase 3 (ClinicalTrials.gov ID: NCT02854033) based on the availability of KL-VS and ApoE ε4 genotyping, T1-weighted MRI, 18 F-flortaucipir (FTP) tau-PET and 18 F-florbetapir (FBP) or 18 F-florbetaben (FBB) amyloid-PET. MR and PET imaging had to be acquired during the same study visit. In addition, a subsample of 200 participants with a followup tau-PET assessment was selected for the longitudinal analyses. The two singlenucleotide polymorphisms for KL-VS (rs9536314 for F352V, rs9527025 for C370S) and ApoE (rs429358, rs7412) were genotyped using DNA extracted by Cogenics from a 3 mL aliquot of EDTA blood. Participants were assigned to the heterozygous KL-VS group when they carried 1, but not 2, copies of the KL-VS haplotype. ApoE ε4 carriers were defined as individuals carrying at least one ε4 allele. Clinical classification was performed by the ADNI centers, dividing participants into cognitively normal (CN, Mini-Mental State Examination [MMSE] > 24, CDR = 0, non-depressed), mild cognitively impairment (MCI; MMSE > 24, CDR = 0.5, objective memory-loss on the education adjusted Wechsler Memory) or AD dementia (ADD; 19 < MMSE < 24, CDR = 0.5-1.0, NINCDS/ADRDA criteria for probable AD are fulfilled). All participants provided written informed consent and all work complied with ethical regulations for work with human participants.
MR and PET acquisition and preprocessing. All imaging data were downloaded from the ADNI loni image archive (https://ida.loni.usc.edu).
All PET images were coregistered to the corresponding T1-weighted image to make use of Freesurfer-derived masks in participants' high resolution, native space. SUVR scores were obtained by normalizing tau-PET images to the inferior cerebellar gray matter and amyloid-PET images to the whole cerebellum, following the previous recommendations 58 . In order to make FBP and FBB amyloid-PET measures comparable, we transformed SUVR scores into centiloid (CL) units using the established transformation formula (http://adni.loni.usc.edu/wp-content/ themes/freshnews-dev-v2/documents/pet/ADNI Centiloids Final.pdf). For the analysis of longitudinal tau-PET, we additionally calculated annual tau-PET SUVR change rates as the difference between tau-PET SUVR scores measured at the follow-up versus baseline visit divided by the follow-up time in years.
Tau-and amyloid-PET regions of interest. For the analyses of tau-PET, we extracted mean SUVR scores from bilateral inferior temporal gyri marking Aβrelated increase of tau pathology to neocortical structures 20,26,[35][36][37] . In addition, we assessed global tau-PET burden 59 as the size-weighted mean SUVR score across all Freesurfer regions, excluding hippocampus, thalamus, and basal ganglia due to commonly reported tracer off-target binding 60 .
For the analysis of amyloid-PET images, we computed mean amyloid-PET levels from a global ROI spanning lateral and medial frontal, anterior and posterior cingulate, lateral parietal, and lateral temporal regions. Mean SUVR from these regions was also used for sample stratification into amyloid-positive participants based on established thresholds (SUVR FBP ≥ 1.11 or SUVR FBB ≥ 1.08; see "ADNI_UCBERKELEY_AV45_Methods_12.03.15.pdf" and "UCBerkeley_FBB_Methods_04.11.19.pdf" on the ADNI website).
mRNA expression levels of KL. Regional gene expression was obtained from publicly available microarray measurements of regional mRNA expression based on post-mortem data from the Allen Brain Atlas (http://human.brain-map.org). The Allen Brain atlas is based on more than 60,000 microarray probes collected from 3700 autopsy-based brain tissue samples from a total of six individuals aged 24-57 without a known history of neurological or psychiatric diseases 33,34 . Microarray-based log2 expression values of 20,737 genes within each of the 3700 samples were mapped back into MNI standard space by the Allen Brain Institute using stereotactic coordinates of the examined probes. The whole gene expression data have been recently mapped to the Freesurfer-based Desikan-Killiany atlas as median gene expression for probes falling within each of the 68 atlas ROIs 38 . Here, we specifically extracted median expression of KL mRNA within these Desikan-Killiany ROIs, to test a spatial correlation between KL expression and KL-VS het effects on local tau-PET uptake. Since microarray assessments and thus KL mRNA expression of all six Allen brain atlas subjects were available only for the left hemisphere (vs. two subjects for the right hemisphere), we restricted the analysis of KL mRNA expression data to the more robust estimates of the left hemisphere in line with previous studies 61,62 .
Neuropsychological assessment. The ADNI neuropsychological test battery contains multiple indicators for memory functions, on which basis a composite score (ADNI-MEM) has been established 39 . ADNI-MEM summarizes test performance on the Rey Auditory Verbal Learning Test, elements from the AD Assessment Scale-Cognitive Subscale, word recall from the MMSE, and the Wechsler Logical Memory Scale II. In the exploratory analysis, we also used established ADNI summary scores of executive functions (ADNI-EF), language (ADNI-LAN), and visual-spatial abilities (ADNI-VS) (see ADNI_Methods_UWNPSYCHSUM_March_2020.pdf on ADNI webpage). Note that 2 participants had no neuropsychological tests available resulting in a sample of 549 participants for this part of the analysis.
Statistical analysis. All statistical analyses were conducted with R statistical software (version 3.6.1). P values were considered significant when meeting a twotailed alpha threshold of 0.05. Baseline tau-PET SUVR values were entered as log- transformed values into the statistical models to approximate normality. All interaction analyses were controlled for the main effect of the interaction terms. Group demographics were compared between KL-VS het carriers versus noncarriers using Welch T-tests for continuous measures and χ 2 tests for categorical measures.
KL-VS heterozygosity × amyloid interaction on tau pathology. In our main analysis, we tested whether KL-VS het moderates the relationship between amyloid-and tau-PET. To this end, multiple linear regression analyses were used to estimate the interaction effects between KL-VS het and global amyloid-PET uptake on tau-PET levels in the inferior temporal ROI and the global ROI (N = 551). Age, sex, diagnosis, education, and ApoE ε4 carrier status were considered as covariates. In secondary analyses, we accounted for potential biases due to unequally sized KL-VS het groups by repeating the same interaction analysis in matched groups of equal size (N = 288). For this purpose, 144 out of 407 KL-VS het non-carriers were selected based on propensity score matching for global amyloid-PET levels and clinical diagnosis using the matchit R package. To ensure that our results were not affected by the skewed distribution of PET data or outliers, we iteratively determined the t-statistic of the Kl-VS het × amyloid interaction effect on tau-PET levels using 1000 bootstrapping iterations (i.e., random sampling from the subject pool with replacement using the boot R package). As a reference, we generated a null distribution of the t-statistic using the same approach, but randomly reshuffling the KL-VS het labels on every iteration. We compared the mean tvalue of the bootstrapped interaction effect to that of the null distribution using Welch T-tests. The significance of the bootstrapped interaction effect was determined by onesamplet-tests estimating whether the resulting distributions of t-values significantly differ from zero and by confirming that the 95% confidence intervals did not overlap with zero. In addition, we accounted for potential influences of clinical diagnoses by repeating the same interaction analyses in the subsample of only MCI (N = 156) or CN (N = 347) participants. Finally, we repeated the analysis in a subsample excluding participants with tau-PET SUVR values in the two ROIs smaller than the reference region (i.e., SUVR < 1).
Next, we tested whether KL-VS het moderates the relationship between amyloid-and tau-PET accumulation over time (N = 200). Separate linear regression models were used to estimate the KL-VS het × global amyloid-PET interaction effect on tau-PET annual change rates in the inferior temporal ROI and the global ROI, controlling for age, sex, and diagnosis.
KL-VS heterozygosity × ApoE interaction on tau pathology. Additional exploratory analyses were run to determine the influence of ApoE ε4 carrier status. To this end, we examined the interaction between KL-VS het and ApoE ε4 status on tau-PET levels in the whole sample (206 ApoE ε4 carriers and 345 non-carriers), controlling for age, sex, diagnosis, education, and global amyloid-PET uptake.
Spatial match between KL mRNA expression and the association of KL-VS heterozygosity on tau pathology. Next, we tested whether the favorable influence of KL-VS het on local tau-PET levels overlapped within those brain regions showing higher local KL mRNA expression levels. To this end, we determined KL mRNA expression using the Allen brain atlas data in all 34 left-hemispheric Desikan-Killiany atlas regions and determined the KL-VS het × amyloid-PET interaction effect on tau-PET uptake for corresponding anatomical regions. We then tested the ROI-to-ROI Pearson-moment correlation between regional KL mRNA expression and the interaction effect test statistic (not restricted to regions showing a significant interaction effect).
KL-VS heterozygosity-memory relationship and lower tau pathology as a mediator. We tested whether KL-VS het was associated with better memory functions, and whether this association was mediated by reduced tau-PET levels. For the main analysis, mediation analysis (causal mediation R package) was conducted in which KL-VS het variant was treated as a predictor, global tau-PET levels as a mediator, and ADNI-MEM scores as an outcome. Mediation analysis was performed in the subsample of amyloid-positive participants (N = 229) since we found KL-VS het to be associated with lower tau-PET levels specifically in individuals with elevated amyloid-PET uptake. Note that, since we conditioned the mediation effect on amyloid-PET levels, this is formally a moderated mediation analysis that we conducted only for one level of the moderator (amyloid status = positive) following our hypothesis. The significance of the mediation effect was determined using 10,000 bootstrapped iterations, where each path of the model was controlled for age, sex, diagnosis, education, ApoE ε4 carrier status, and global amyloid-PET levels.
We ran an alternative analysis strategy in the whole sample (including amyloidpositive and -negative participants, N = 549) that estimated the KL-VS het × amyloid-PET interaction effect on ADNI-MEM scores. Importantly, the interaction analysis was once run without and once with controlling global tau-PET levels. We specifically hypothesized that if the beneficial influence of KL-VS het is dependent on lowering tau accumulation, then the interaction effect should be diminished in the tau-controlled analysis. Other covariates considered in the multiple regression models were age, sex, diagnosis, education, and ApoE ε4 carrier status. In secondary analyses, we repeated the mediation analysis for ADNI summary scores of other cognitive domains.
Is KL-VS heterozygosity associated with lower Aβ accumulation? Lastly, we performed an exploratory analysis with the aim to confirm previously observed ageand ApoE-dependent associations between KL-VS het and lower amyloid-PET burden 13 .
For this purpose, we tested for a KL-VS het × age effect on global amyloid-PET levels in the current sample (N = 551) and in a larger ADNI sample (N = 1067) including all participants with amyloid-PET assessment and KL-VS status (regardless of whether or not they underwent tau-PET assessment). Sex, education, and diagnosis were considered as covariates. Comparable to the original report, we also investigated ApoEdependent effects of KL-VS het on amyloid-PET levels in a subgroup including only CN participants aged between 60 and 80 years. Age, sex, and education were considered as covariates.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data that support the findings of this study were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) and are available from the ADNI database (adni.loni.usc.edu) upon registration and compliance with the data use agreement. A list including the anonymized participant identifiers of the currently used sample and the source file can be downloaded from the ADNI database (tau-PET data release in May 2020; UCBERKELEYAV1451_05_12_20.csv). The Allen Brain Atlas (http://human. brain-map.org) and Freesurfer-mapped transcriptomic data from the Allen Brain Atlas (http://figshare.com/articles/A_FreeSurfer_view_of_the_cortical_transcriptome_ generated_from_the_Allen_Human_Brain_Atlas/1439749) are freely available online. Source data underlying Fig. 2 are provided with this paper.

Code availability
The R code pertaining to the figures in this manuscript is provided at https://github.com/ njulianeitzel/NatCommun2021_KL-VS. Costume R code can be obtained from the first author upon request.