Towards cascading genetic risk in Alzheimer’s disease

Abstract Alzheimer’s disease typically progresses in stages, which have been defined by the presence of disease-specific biomarkers: amyloid (A), tau (T) and neurodegeneration (N). This progression of biomarkers has been condensed into the ATN framework, in which each of the biomarkers can be either positive (+) or negative (−). Over the past decades, genome-wide association studies have implicated ∼90 different loci involved with the development of late-onset Alzheimer’s disease. Here, we investigate whether genetic risk for Alzheimer’s disease contributes equally to the progression in different disease stages or whether it exhibits a stage-dependent effect. Amyloid (A) and tau (T) status was defined using a combination of available PET and CSF biomarkers in the Alzheimer’s Disease Neuroimaging Initiative cohort. In 312 participants with biomarker-confirmed A−T− status, we used Cox proportional hazards models to estimate the contribution of APOE and polygenic risk scores (beyond APOE) to convert to A+T− status (65 conversions). Furthermore, we repeated the analysis in 290 participants with A+T− status and investigated the genetic contribution to conversion to A+T+ (45 conversions). Both survival analyses were adjusted for age, sex and years of education. For progression from A−T− to A+T−, APOE-e4 burden showed a significant effect [hazard ratio (HR) = 2.88; 95% confidence interval (CI): 1.70–4.89; P < 0.001], whereas polygenic risk did not (HR = 1.09; 95% CI: 0.84–1.42; P = 0.53). Conversely, for the transition from A+T− to A+T+, the contribution of APOE-e4 burden was reduced (HR = 1.62; 95% CI: 1.05–2.51; P = 0.031), whereas the polygenic risk showed an increased contribution (HR = 1.73; 95% CI: 1.27–2.36; P < 0.001). The marginal APOE effect was driven by e4 homozygotes (HR = 2.58; 95% CI: 1.05–6.35; P = 0.039) as opposed to e4 heterozygotes (HR = 1.74; 95% CI: 0.87–3.49; P = 0.12). The genetic risk for late-onset Alzheimer’s disease unfolds in a disease stage-dependent fashion. A better understanding of the interplay between disease stage and genetic risk can lead to a more mechanistic understanding of the transition between ATN stages and a better understanding of the molecular processes leading to Alzheimer’s disease, in addition to opening therapeutic windows for targeted interventions.


Introduction
Alzheimer's disease is characterized, at the neuropathological level, by the build-up of two proteins: amyloid plaques and neurofibrillary tangles of phosphorylated tau. 1 Both these pathological features can be observed long before the memory loss and decline in executive function that is characteristic of patients with Alzheimer's disease.Accumulation of amyloid plaques in the brain predates the clinical symptoms of Alzheimer's disease by two decades, 2 whereas the spatial distribution of tau tangles reflects more closely the reported cognitive deficits and neurodegeneration. 3he amyloid cascade hypothesis postulates that the deposition of the amyloid-β protein (the main component of the amyloid plaques) is the cause of Alzheimer's disease and that neurofibrillary tangles, cell loss, vascular damage and dementia are a direct consequence. 4In keeping with the amyloid cascade hypothesis, a theoretical framework for the progression of biomarkers during the course of Alzheimer's disease has been developed. 5Here, amyloid pathology is the first to appear, followed by tau pathology, neurodegeneration and, finally, cognitive decline.][8][9][10] In an attempt to operationalize this, a simplified ATN model has been proposed. 11The components of the ATN model refer to the status of three different key biomarkers in Alzheimer's disease: amyloid (A), tau (T) and neurodegeneration (N).In this approach, each of the three biomarkers can be either positive or negative.Exceeding the centiloid threshold in amyloid PET imaging would place a participant into the A-positive (A+) group, whereas a scan slightly below the threshold would be considered amyloid negative (A−).One practical advantage (but also a major source of criticism) of this model is that the biomarker status can be assessed using a variety of methods: wet biomarkers (CSF or plasma) or brain imaging (centiloids or visual reads). 12Although the progression from A−T−N− to A+T−N− to A+T+N− to A+T+N+ would be the most typical progression in Alzheimer's disease and in keeping with the theoretical framework of biomarker progression, all combinations of biomarker statuses emerge in observational cohorts. 13,14enome-wide association studies (GWASs) have broadened the understanding of the genetic basis of late-onset Alzheimer's disease over the last decades. 157][18] Recently, these GWASs have been enriched with participants with a family history of Alzheimer's disease (diagnosis-by-proxy), [19][20][21] leading to a drastic increase in sample size and expanding the set of genetic risk loci for Alzheimer's disease to 90. 15 These loci have been linked to various molecular processes, such as immunity, cholesterol processing and endocytosis. 22rther studies investigated the genetic effects on Alzheimer's disease-related biomarkers, ranging from tau and amyloid levels in CSF 23,24 or in the brain 25,26 to MRI-based measures, such as hippocampal volume 27 and phenotypes derived from disease progression modelling. 28The strongest common genetic risk factor for Alzheimer's disease is the e4 allele of the APOE gene: carriers of the e4 allele have a 2-to 4-fold increased risk of developing Alzheimer's disease, and e4 homozygous subjects have an 8-to 12-fold increased risk. 29The genetic risk outside the APOE region is often summarized using polygenic scores, which have been shown to improve predictions of clinical diagnosis 30 and pathologyconfirmed cases. 31][37] For Alzheimer's disease and for other disorders, genetic risk is often considered as a time-invariant constant.That is, genetic risk identified through case-control studies is assumed to affect both onset and progression.However, given that Alzheimer's disease is now understood to unfold in stages, we hypothesized, in line with the amyloid cascade hypothesis and the A/T/N framework, that genetic risk in Alzheimer's disease is disease-stage dependent; i.e. some genetic risk factors will aid the transition from A− to A+, whereas other, distinct genetic risk factors will increase the risk to transition from T− to T+.
In this work, we explore whether genetic vulnerability to Alzheimer's disease varies with disease stage.Using longitudinal data from the Alzheimer's Disease Neuroimaging Imitative (ADNI) and survival analysis, we show that APOE contributes to progression from A−T− to A+T−, but only marginally from A+T− to A+T+.Conversely, polygenic risk contributes to the progression from A+T− to A+T+, but not from A−T− to A+T−.

Data
Data used in the preparation of this article were obtained from the ADNI database (http://adni.loni.usc.edu).The ADNI was launched in 2003 as a public-private partnership, led by Principal Investigator Michael W. Weiner, MD.The primary goal of the ADNI has been to test whether serial MRI, PET, other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment and early Alzheimer's disease.For up-to-date information, see https://adni.loni.usc.edu/about/.ADNI study data were accessed through the R package ADNIMERGE (accessed: 20 July 2023).

Preparation of genetic data
The genetic data preparation followed the procedure described by Altmann et al. 32 The additional genetic data contributed by the ADNI-3 cohort was integrated with the existing data using the same processing pipeline.Briefly, single nucleotide polymorphism (SNP) genotyping data were available for n = 2001 subjects across ADNI phases 1, 2, GO and 3. Genotyping was conducted using four different platforms: Human610-Quad, HumanOmniExpress, Omni 2.5 M and Illumina Infinium Global Screening Array v.2 (Illumina). 38Prior to imputation, we applied subject-level quality control (QC) steps based on call rate (10% cut-off) and concordance between chip-inferred sex and self-reported sex separately for the four genotyping arrays; all subjects were retained.At the SNP level, we conducted standard QC steps ensuring compatibility with the reference panel used for imputation [strand consistency, allele names, position, reference/alternative allele assignments and minor allele frequency discrepancy (0.2 cut-off)].Imputation was carried out using the Sanger Imputation Server (https://imputation. sanger.ac.uk/), with SHAPEIT for phasing, 39 positional Burrows-Wheeler transform 40 for imputation and the entire Haplotype Reference Consortium (release 1.1) reference panel. 41Data from the four different genotyping platforms were imputed separately.As part of post-imputation QC, multi-allelic variants and SNPs with imputation INFO score of <0.3 were removed, and genotype calls with a posterior probability of <0.9 were set to missing (i.e.hard called).Following the initial QC, genotypes from the four platforms were merged.Additional information on the imputation and QC process is detailed by Scelsi et al. 28,42 Using the merged data, we retained SNPs with minor allele frequency ≥ 1% and genotyping rate of >0.9.
SNPweights 43 was used to infer genetic ancestry from genotyped SNPs.The reference panel comprised Central European, Yoruba Africans and East Asian samples from HapMap 3 44 and native Americans from Reich et al. 45 For this study, only participants with predicted central European ancestry of ≥80% were retained (n = 1851).Next, using the imputed and merged data genetic relatedness between central European participants was computed.Initially, the SNP content was restricted to SNPs with minor allele frequency ≥ 5%, and linkage disequilibrium (LD) pruning was carried out in PLINK v.1.9(-indep-pairwise 1000 50 0.1).The genetic relatedness matrix was computed using the remaining autosomal SNPs, and the dataset was trimmed to remove subects with relatedness of >0.1 (-rel-cutoff 0.1), leading to n = 1833 unrelated participants.

Definition of genetic risk
In this study, we focused on two sources of genetic risk: (i) the risk conferred through the APOE gene based on the genetic markers for APOE-e2 and APOE-e4; and (ii) the polygenic risk conferred by the remaining genome.As described previously, 32 polygenic risk scores (PRSs) were computed using the software PRSice v.2.1.9. 46As base GWAS, the stage 1 results of the Alzheimer's disease GWAS featuring a clinically defined Alzheimer's disease phenotype was used. 18or PRS computation, SNPs with minor allele frequency ≥ 5% were considered, and SNPs were selected using LD clumping (1000 kb, R 2 of 0.1 and P-value threshold of 1.0) within the ADNI cohort, missing SNPs were simply ignored at the subject level (using the setting -missing SET_ZERO), and the APOE region was excluded (hg19 coordinates chr19 from 44 400 000 to 46 500 000).For this study, we used only the P-value cut-off of 1.0 × 10 −8 to build the PRS (Supplementary Table 1).PRSs were computed for all ADNI participants with genome-wide genotyping data.Of the remaining subjects, n = 417 ADNI-1 participants who contributed to the Alzheimer's disease GWAS 18 were excluded from the analysis to ensure independence between training and application dataset for PRS.Thus, n = 1416 unrelated participants with central European ancestry were eligible for inclusion in the study.

Definition of amyloid status
For this project we relied on two modalities to define amyloid status: amyloid-β PET using the 18 F-florbetapir and 18 F-florbetaben PET tracers, and CSF measures of amyloid-β(1-42) using the Roche Elecsys® immunoassay.We used data processed by ADNI for both modalities.Detailed information on the PET processing is available elsewhere (https://adni.loni.usc.edu/methods/petanalysis) 47; and information on CSF amyloid-β(1-42) processing is detailed by Bittner et al. 48and Hansson et al. 49 For CSF amyloid-β(1-42) we used the cut-off of 880 pg/ml, 49 and for amyloid-β PET we used the tracer-specific standardized uptake value ratio (SUVR) (whole cerebellum reference) cut-offs of 1.11 and 1.08 for 18 F-florbetapir and the 18 F-florbetaben, 50 respectively.A participant's visit was labelled as A+ if either the PET result or the CSF result indicated a positive amyloid finding (i.e. in cases where the results were discordant, the visit would be labelled as A+).Visits with only negative amyloid findings were labelled as A−, and visits without any information on amyloid (i.e.neither a PET nor a CSF result) were labelled as 'amyloid missing'.

Definition of tau status
In keeping with our definition of amyloid positivity, we used available data from CSF and PET imaging to define the tau status.More precisely, we used the Phospho-Tau( 181 P) CSF Roche Elecsys® immunoassay and PET imaging using the 18 F-flortaucipir tracer.Details on the processing are available elsewhere (https://adni.loni.usc.edu/methods/pet-analysis). 48,49,51For CSF we used phosphorylated 181 P tau (pTau) with a cut-off of 34.61 pg/ml, and for tau PET we used the cut-off of 1.42 in the meta temporal region of interest comprising the amygdala, entorhinal cortex, fusiform gyrus, inferior and middle temporal gyri 52 when normalized to the inferior cerebellar grey matter. 53Both cut-offs were data driven: (i) the tau PET cut-off corresponds to a z-score of 2.0 in the cognitively normal participants in ADNI (n = 506) and is close to the 'high tau' cut-off of 1.43 defined by Jack Jr et al. 54 ; and (ii) the pTau cut-off was set to maximize the Youden's index between CSF pTau and tau PET (at the 1.42 cut-off) in (n = 502) ADNI participants with concurrent CSF and PET measurements.The same labelling scheme as for amyloid was applied: a visit was labelled as T+ if either the CSF or the PET indicated a positive finding, T− if there were only negative tau findings, and 'tau missing' if neither data on CSF pTau nor on tau PET were available.

Statistical modelling
We used Cox proportional hazards models to investigate the genetic contribution of progressing (i) from A−T− to A+T−; and (ii) from A+T− to A+T+.For (i), we included every eligible participant with genetic data and who was A−T− based on their biomarker results as described above.This earliest A−T− visit was considered the 'start' visit (i.e. the status of previous visits was ignored).A subject was considered a converter when the available biomarkers indicated A+T− at a later visit.The time of the first A+T− biomarker finding after their initial A−T− visit was used as the conversion time.For non-converters, we recorded the last visit where both amyloid and tau biomarker information was available to define the maximal follow-up time.Likewise, for (ii) we included every participant with genetic data and who was A+T−.This first A+T− visit was considered the 'start' visit.A subject was considered a converter when the available biomarkers indicated A+T+ at a later visit.The time of the first A+T+ biomarker finding after their initial A+T− visit was used as the conversion time.As before, for nonconverters we recorded the last visit when amyloid and tau biomarker information was available as their last point of contact.
For the analyses, the Cox proportional hazards model included age at the inclusion visit (i.e.either the A−T− or the A+T− visit), sex and education.As variables of interest, we also included genetic variables for PRS, in addition to allele counts of APOE-e2 and APOE-e4.The proportional hazards assumption was tested for each covariate and for the global model using statistical tests and graphical diagnostics based on scaled Schoenfeld residuals.As a measure of overall model performance, the concordance index (C-index) was computed for the following: (i) full models; (ii) models without APOE; (iii) models without PRS; and (iv) models without any genetics.Statistical tests were carried out in R (v.4.1.0)using the survival (v.3.5.5), survminer (v.0.4.9) and rms (v.6.8.0) packages.

Sensitivity analyses
In addition to the two main analyses, we conducted a series of sensitivity analyses addressing the conversion definition, biomarker cut-offs, biomarker source and polygenic score source.

Conversion definition
To maximize the available data, we relaxed the requirement for both amyloid and tau biomarkers to be available at the same visit to define conversion.As before, both biomarkers at the same time were required to define the 'start' visit as either A−T− or A+T−.However, for defining progression from A−T− or A+T−, a single A+ or T+ visit was sufficient, respectively.

Biomarker cut-offs
We varied the tau PET cut-off from 1.0 to 3.0 standard deviations (SDs) above the mean in the cognitively normal ADNI participants.Notably, the lowest cut-off resulted in 1.31 and was close to the neuropathologically defined cut-off of 1.29 by Lowe et al. 55 The pTau cut-offs were adjusted accordingly to maximize Youden's index between CSF pTau and tau PET.

Biomarker source
To maximize data and follow-up time, the main analyses combined data from two biomarker sources: CSF and PET.Additional sensitivity analyses relied exclusively on either CSF biomarkers or PET biomarkers.For this analysis, pTau cut-offs were varied in the range from 22 to 31, covering the values of 24.25 and 29.19, which were found to indicate tau PET positivity in Braak III/IV and Braak V/VI regions, respectively. 56Tau PET cut-offs were varied, as before, from 1.0 to 3.0 SD above the mean in cognitively normal participants (i.e. in the same range as the main analysis).

Polygenic score source
To include more SNPs in the PRS, we explored the summary statistics on Alzheimer's disease and related dementias by Bellenguez et al. 21We followed the same PRS pipeline as above and applied a P-value threshold of 5.0 × 10 −8 (i.e.genome-wide significant), leading to 77 included SNPs (Supplementary Table 2).

Results are independent of biomarker source
The main analysis combined different biomarker sources to maximize the available data and the observation time for the conversion analysis.Relying on PET biomarkers alone, only 7.5% of visits (1237 of  5 and 6).Despite the reduced statistical power, these sensitivity analyses confirmed the pattern observed in the main analysis: APOE-e4 contributed mainly to the A−T− to A+T− conversion, whereas PRS contributed to the A+T− to A+T+ conversion (Fig. 3 and Supplementary Tables 5 and 6).

Discussion
This longitudinal survival analysis demonstrated that APOE-e4 plays an important role for the progression from A−T− to A+T−, but APOE is of only marginal importance in A+T− participants who progress to A+T+.Conversely, polygenic risk for Alzheimer's disease exhibited the inverse pattern: there was no contribution to the progression from A−T− to A+T−, but a significant contribution to faster progression from A+T− to A+T+.This held true for an alternative PRS defined using a different genetic study and involving a larger number of genetic loci.Notably, when assessing covariates, a differential effect of years of education was observed: higher education had no effect (Fig. 1) or was marginally harmful (Supplementary Fig. 2) for the progression from A−T− to A+T− but was protective for the conversion from A+T− to A+T+ (Fig. 2).The bisection of the genetic risk by disease stage was largely independent of the applied biomarker cut-offs.Moreover, relying on only a single source of biomarkers for defining stage and conversion confirmed the findings of the main analysis despite reduced sample size and observation time.
8][59] Given that amyloid deposition occurs 10+ years before other Alzheimer's disease processes 2 and APOE-e4 is the strongest common genetic risk factor for amyloid deposition, it would be expected for APOE-e4 to exert its maximum effect in younger people.Still, APOE remains the strongest risk factor in individuals ≥80 years old. 58Thus, the agedependent heterogeneity of APOE is likely to be compounded by a survivor bias: individuals with a very late onset despite carrying APOE-e4 might harbour protective variants, 60 such as KLOTHO-VS, where a protective effect on amyloid deposition and Alzheimer's disease was observed in only 60-to 80-year-olds, but not in the 80+ group. 57,61he findings from this longitudinal analysis are also in line with previous reports in the ADNI cohort of APOE and polygenic risk on amyloid and tau.For instance, cross-sectional amyloid biomarkers in the CSF and in the brain were mainly driven by APOE, whereas cross-sectional CSF tTau and pTau measurements were associated with PRS beyond the APOE locus. 32Moreover, APOE was found to predict amyloid status, whereas polygenic risk for Alzheimer's disease improved predictions of diagnosis and of clinical progression from mild cognitive impairment to Alzheimer's disease above APOE alone. 35These observations extend to plasma markers of tau pathology: PRSs (that excluded the APOE region) were found to be associated only with plasma p-tau181 in A+ participants. 62This association between polygenic risk (beyond APOE) and CSF tau biomarkers rather than with amyloid and neurodegeneration was also observed outside of the ADNI study. 63Polygenic risk (beyond APOE) was associated with non-amyloid endophenotypes in a large cohort of people with mild cognitive impairment.This suggests that these variants are more closely linked with neuronal degeneration than with Alzheimer's disease-related amyloid pathology. 64ll these previous studies made the connection between existing amyloid pathology and correlations between polygenic risk and tau pathology using cross-sectional study designs.In a recent longitudinal study of tau PET in the ADNI cohort, higher polygenic risk for Alzheimer's disease was associated with accelerated increase in tau signal in the brain, and this effect was modulated by amyloid pathology: A+ participants showed a stronger effect of PRS on tau accumulation. 65Our longitudinal analysis, which combined CSF and PET data to maximize the sample size, confirms these observations and indicates that polygenic Alzheimer's disease risk (outside the APOE region) contributes to tau pathology in A+ participants but has no meaningful contribution in A− participants.The genetic architecture of Alzheimer's disease, as captured by the PRS, involves multiple different pathways, mainly amyloid-β processing, tau, immunity and lipid processing. 18,21The PRS that we used covers three genes that have been associated with tau binding: BIN1, CLU and PICALM.BIN1 mediates Alzheimer's disease risk by modulating tau pathology, 66 and BIN1 risk variants increase tau PET (but not amyloid PET) 67 in an amyloid-dependent fashion. 68ur observations of an amyloid-dependent effect of the PRS align with these earlier single-gene studies.Although there are currently no mechanistic analyses that explain the amyloid-dependent effect of BIN1 on tau pathology, recent data from animal models in Alzheimer's disease suggest a state-dependent effect of genetic risk factors related to microglia: deletion of Trem2 in mouse models exacerbated tau accumulation and spreading, leading to brain atrophy, but only in the presence of existing amyloid-β pathology. 69long the same lines, physical contact between microglia and plaques in addition to a functioning TREM2 gene are necessary for the appropriate microglial response to amyloid pathology. 70Thus, defects in TREM2 can contribute to neurodegeneration only once amyloid pathology has been established.Consequently, other genes contributing to the PRS might also exert their effect in an amyloid-dependent fashion.Our longitudinal analysis presented here is the first to support such a state-dependent genetic risk model in humans, and further fine-grained examination of how the pathways involved in the PRS contribute to sequential disease progression are needed.
The partition of Alzheimer's disease genetic risk into APOErelated and polygenic risk beyond APOE is a simplification in this analysis.Recent works have linked established Alzheimer's disease risk loci outside APOE, such as CR1, to amyloid biomarker levels. 23onversely, studies of biomarker levels of tau repeatedly highlight the APOE locus. 23,24However, if being A+ were a prerequisite to exhibit pathological accumulation of tau, then the strong genetic association with APOE in these GWASs would merely reflect the necessary condition rather than a genuine direct molecular process that affects tau levels.The strong dependence of tau levels on established amyloid pathology is supported by mediation analyses in recent cross-sectional 63 and longitudinal 65 studies.Moreover, the known Alzheimer's disease genetic risk variants are contributing differently to molecular pathways, 18,21,22 where each pathway in turn will exercise differential effects on the Alzheimer's disease biomarkers, including markers for vascular pathology. 37Therefore, a pathway PRS comprising only genes associated with the regulation of the amyloid precursor protein catabolic process (e.g.Gene Ontology term GO:1902991) might contribute significantly to the conversion from A−T− to A+T−.Likewise, a pathway PRS using only genes known to bind the tau protein (GO:0048156) might exhibit an even stronger association with A+T− to A+T+ conversion than the general PRS used here.
In addition to the cascading effect of genetic risk in Alzheimer's disease, we also observed a stage-dependent effect of non-genetic risk factors.Education has been shown to have a protective effect against dementia 71 : here, we show that higher rates of education do not influence transition to amyloid positivity but do slow progression from A+T− to A+T+.Consequently, other non-genetic risk factors might show a similarly state-dependent effect on the pathological pathway from A−T− to A+T+ and further neurodegeneration.
The study has several limitations.First, the present analysis was limited to two biomarkers in Alzheimer's disease: amyloid-β and tau.It would be desirable to include neurodegeneration (N; of the ATN framework) or, potentially, more fine-grade staging from advanced data-driven disease-progression modelling. 7However, at this point adding further stages would reduce the available sample size.Second, we partitioned the genetic risk in Alzheimer's disease into two components: APOE and other top variants combined into a single polygenic risk score.Further work should explore a more fine-grained partition of the polygenic risk into individual SNPs or into pathway PRS. 37Third, the study population was of central European ancestry; therefore, it is unclear whether the findings would generalize to other genetic backgrounds.Finally, although the ADNI cohort is a large cohort, the number of subjects who were eligible for our analysis was reduced owing to the requirement for concordant and longitudinal recordings of multiple biomarkers in addition to genetics.The available sample size might have limited statistical power to render the estimated hazards ratios significant in some settings.However, the two conversion analyses were based on similar sample sizes (∼300 participants), thus allowing us to make a relative comparison between the genetic effect (of APOE or PRS) under two different biomarker-defined disease stages.Moreover, uncertainty of the estimated effects might also be increased owing to disease heterogeneity in Alzheimer's disease, [72][73][74][75] which is likely to be underpinned by differences in genetic architecture.Thus, analyses in further large longitudinal cohorts are required to confirm the observation of stage-dependent genetic vulnerability in Alzheimer's disease and to uncover more finegrained associations with Alzheimer's disease subtypes.

Conclusion
In this work we demonstrated, in a simplified setting, that genetic risk for late-onset Alzheimer's disease unfolds in a disease stagedependent fashion.A better understanding of the interplay between disease stage and genetic risk can lead to a better understanding of the molecular processes leading to Alzheimer's disease, in addition to opening therapeutic windows for targeted interventions and personalized approaches to dementia prevention.

Data availability
Data used in the preparation of this article were obtained from the ADNI database (http://adni.loni.usc.edu) and are freely available

Figure 2
Figure 2 Hazard ratios for the conversion from A+T− to A+T+.(A) Forest plot depicting the hazard ratios for all covariates in the model.(B) Estimated survival curves stratified by APOE-e4 genotype.(C) Estimated survival curves stratified by PRS percentile (5%, 50% and 95%).APOEe2 = number of APOE-e2 alleles; APOEe4 = number of APOE-e4 alleles; Edu = years of education; PRS = polygenic risk score, scaled to zero mean and unit standard deviation.
401) had both biomarkers, and it resulted in shorter observation times for A−T− to A+T− conversion [3.62 (SD = 1.07) years] and A+T− to A+T+ conversion [2.94 (SD = 1.13) years] compared with the main analysis.Relying solely on CSF biomarkers, 19.2% of visits (3155 of 16 401) had both biomarkers.Overall, this led to a shorter observation time with respect to the main analysis for A−T− to A+T− conversion [3.94 (SD = 2.7) years] and A+T− to A+T+ conversions [3.6 (SD = 2.38) years].Furthermore, relying on a single source of biomarkers led to reduced sample sizes (from ∼300 in the main analyses to 70-250 in the sensitivity analyses) and observed conversions (Supplementary Tables

Figure 3
Figure 3 Survival curves for PET-only and CSF-only analyses.The top row (A-D) is based on results from AT(N) definitions based exclusively on PET biomarkers (amyloid and tau), with a cut-off of 1.45 for tau PET.The bottom row (E-H) relied on CSF biomarkers (ABETA42 and pTAU), with a cut-off of 26 for pTau.The two left columns (A, B, E and F) display the progression probability from A−T− to A+T− stratified by APOE-e4 genotype and PRS quantile (5%, 50% and 95%).The two right columns (C, D, G and H) display the progression probability from A+T− to A+T+.