Cognitive Capacity Genomewide Polygenic Scores Identify Individuals Resilient to Cognitive Decline in Aging

The genetic underpinnings of cognitive resilience in aging remains unknown. Predicting an individual’s rate of cognitive decline—or cognitive resilience—using genetics will allow personalized intervention for cognitive enhancement and optimal selection of target samples in clinical trials. Here, using genome-wide polygenic scores(GPS) of cognitive capacity as the genomic indicators for variations of human intelligence, we examined the genetic liability of cognitive abilities in the behavioral/cognitive phenome to understand individual phenotypic differences over time. We analyzed the 18-year records of the cross-sectional and longitudinal sociogenomic data of 8,511 European-ancestry adults from the Wisconsin Longitudinal Study (WLS), especially focusing on the cognitive assessments that were repeatedly administered to the participants at their average ages of 64.5 and 71.5. Our linear mixed-effects model identi�ed a signi�cant interaction effect between age and cognitive capacity GPS, which indicates that a higher cognitive capacity GPS signi�cantly correlates with a slower cognitive decline in the domain of immediate memory recall (p-value = 1.79E-03, β = 1.86E-01). Also, the phenome-wide analysis identi�ed several signi�cant associations of cognitive capacity GPSs on the cognitive and behavioral phenome, such as Similarities task (p-value = 3.59E-74, β = 1.36, 95% CI=(1.22, 1.51)), Number Series task(p-value = 2.55E-78, β = 0.94, 95% CI=(0.85, 1.04)), IQ scores(p-value = 7.74E-179, β = 1.42, 95% CI=(1.32, 1.51)), high school class rank (p-value = 3.07E-101, β = 1.86, 95% CI=(1.69, 2.02), Openness from the BIG 5 personality factor(p-value = 2.19E-14, β = 0.57, 95% CI=(0.42, 0.71)), and social participation of reading books (p-value = 2.03E-21, β = 0.50, 95% CI=(0.40, 0.60)), attending cultural events, such as concerts, plays or museums (p-value = 2.06E-23, β = 0.60, 95% CI=(0.49, 0.72)), and watching TV (p-value = 4.16E-18, β=-0.48, 95% CI=(-0.59, -0.37)). As the �rst phenome-wide analysis of cognitive and behavioral phenotypes, this study presents the novel genetic protective effects of cognitive ability on the decline of memory recall in an aging population.


Introduction
The magnitudes of cognitive decline in aging, a major health concern in contemporary society, differ substantially across individuals [1,2].Unraveling the genetic underpinning for individual variations of cognitive decline with aging, particularly those associated with cognitive resilience, could help develop individualized interventions for cognitive decline and allow better sample selection in clinical trials in dementia research.Despite studies reporting the genetic risk factors of accelerated cognitive decline among individuals with dementia [3][4][5], we know little about the genetic protective factors of cognitive resilience in the normal aging population.
Genome-wide Polygenic Scores (GPS) leverage the fact that most human traits are developed from the aggregated in uence of many genetic variants, both common and rare [6][7][8].By aggregating the miniscule effects of millions of genetic variants into a single score, GPS allows researchers to stratify individuals by their genomic propensity for a particular trait and to select individuals with extremely high or low GPS for further research.The recent large genome-wide association studies (GWAS) of educational attainment, an often-used proxy phenotype for human intelligence, identi ed 1,271 independent autosomal loci reaching genome-wide signi cance [9].These ndings suggest several biological pathways related to brain development or neuron-to-neuron communication contribute to human intelligence.While the GWAS revealed many genetic variants associated with cognitive phenotypes (such as cognitive performance, math ability, highest math class taken) [9][10][11][12][13][14][15][16], the genomic contribution to speci c cognitive domains remains unknown, as does their relationship to cognitive changes with aging.
Since general cognitive ability is known to be highly heritable (50-70%) and polygenic [17,18], we utilize GPS to account for the genome-wide factors underlying cognitive capacity and its secular changes.We leverage the expansive phenotype information of a 50+ year social longitudinal database for phenomewide association studies (PheWAS).The Wisconsin Longitudinal Study (WLS), the longestrunning social longitudinal study in the United States [19,20], encompasses a detailed and broad lifelog of cognition, personality, nancial, health, and socioeconomic status.The surveys have repeatedly administered the same cognitive ability tests with the time interval of ~10 years in their latest survey rounds, as well as collected the genotype data of the participants, which creates a deep genotypephenotype catalogue of an individual's cognitive and behavioral traits over their adult lives.
Herein, we hypothesize that the polygenic in uence of the cognitive capacity can explain certain patterns of cognitive abilities and their declines in aging as well as other socio-behavioral phenotypes that might be affected by genetics of cognitive abilities.We tested the associations between longitudinal observations of individual cognitive/behavioral phenomes and the GPSs of four different cognitive phenotypes (educational attainment, cognitive performance, math ability, highest math class taken) with a focus on the secular changes of cognitive test scores.The approach was designed to systematically address the following research questions: rstly, whether a certain cognitive domain is more impacted by polygenic in uence than other cognitive domains; secondly, whether individuals with different GPS show different patterns of cognitive decline in aging; and, thirdly, the extents to which phenotypic variances of behavioral/cognitive phenotypes can be explained by genetic liability of cognitive capacities.

Cognitive/Behavioral PheWAS
Linear regression was used to investigate the associations between the four types of cognitive capacity GPS (EA, CP, HM, MA) and the normalized variables of the cognitive and behavioral phenotypes.We adjusted for biological sex, age and the rst 10 PCs of genetic ancestry and estimate each GPS' signi cance (p-value), effect size (β), 95% con dence interval (CI), and proportion of variance explained (R 2 ) for the target outcomes.

Cognitive changes
Among aforementioned cognitive assessments, we have selected 7 repetitive measures administered to the participants with an average 6.5-year interval and investigated its interaction effects with cognitive capacity GPSs as the participants aged: Similarities, Letter Fluency, Category Fluency, Immediate Recall, Delayed Recall, Digit Ordering and HUI3 cognition level (timepoint 2/3).Linear mixed-effects regressions were nested by individual ID and each survey round (random effect) and included the following xed covariates in the analysis: age at the survey time point, biological sex, rst 10 ancestrally-informative principal components(PC1-10) of genotype data and years of education.Bonferroni-correction was used to adjust for multiple testing and the scores were normalized except for an ordinal variable, HUI3 cognition level.We hypothesized that the contribution of genetic factors to cognitive phenotypes are associated with different degree of cognitive decline in certain cognitive domain.The analyses were performed in R 3.5.1 environment, and linear mixed-effect model was run with lme4 package [35].

Participant Demographics
Our study included 8,511 European-ancestry individuals with DNA genotype data, behavioral questionnaire data and cognitive assessment data available, including 7 different cognitive ability tasks which were administered repetitively with an average 6.5-year interval (SD=1.25 year).The average age of the study participants was 48.6 years at the time of the rst round of cognitive assessment (WLS survey round 4, 1992-1994, SD=15.  1, Supplementary Table 1).The variance of IQ scores explained by CP GPS was 10.4% (Adjusted R 2 ) whereas the baseline covariate model without GPS variable explained 0.8% of IQ score variance.

Cognitive GPSs correlate with the changes of Immediate Recall
Our linear mixed effect model identi ed the signi cant age-x-GPS interaction effect in the changes of Immediate Recall task with the participants' age.All four kinds of cognitive capacity GPS showed signi cant interaction effects with participants' age (Age:GPS) for the Immediate Recall test scores (strongest with EA GPS, p-value=1.79E-03,β=1.86E-01) (Table 2).Their positive effect sizes suggest that an individual who has higher EA GPS tend to show less changes in cognitive assessments as an individual ages.Table 2. Results of the linear mixed effects model analysis explaining the temporal changes of immediate recall assessment score of the WLS participants by the GPSs of Educational attainment (EA), Cognitive Performance (CP), Math Ability (MA), Highest Math Class (HM) with age interaction effect.Effect size of each variable were presented with 95% con dence interval within the parentheses.
Compared to the individuals in the lowest GPS quartile (EA GPS mean=-0.480),individuals in the highest GPS quartile (EA GPS mean=-0.089)showed less decrease in Immediate Recall score changes in later survey rounds.Slope of participants' age in the lowest GPS quartile (β= -1.97E-01, 95% CI=(-0.231,-0.163), p-value of slope=8.61E-31)distinctively showed more intense decrease compared to the highest GPS quartile group (β=-1.24E-01,95% CI=(-0.158,-0.090), p-value of slope=6.67E-13)(Figure 2(a)).The pseudo-R 2 of our linear mixed models in explaining Immediate Recall by cognitive capacity GPS was up to 0.063 with xed effects and 0.170 with both xed and random effects (both with EA GPS).
To visually depict the degree of cognitive changes according to GPS, we divided the cohort into 4 quartiles based on the GPS of each individual and analyzed the average phenotypic changes of each group over time.The average Immediate Recall task scores of the individuals in the highest GPS quartile were 0.044 (z-score) at timepoint 2 (average age of participants 64.5), and increased to 0.073 (z-score) at timepoint 3 (average age 71.5) (1.65 fold increase).In contrast, the average task scores of the individuals in the lowest GPS quartile were -0.031 (z-score) at timepoint 2 and decreased to -0.077 (z-score) at timepoint 3 (2.48 fold decrease) (Figure 2(b)).

Discussion
In this study, we assessed the genetic in uences of general cognitive abilities on cognitive and behavioral phenome using a combinational approach of GPS-based PheWAS on longitudinal observations in the aging population.We hypothesized that the contribution of genetic factors to cognitive capacities are associated with certain cognitive or behavioral phenotypes, and even different degree of cognitive decline in certain cognitive domains.
Our study identi ed that the effect of the age-x-GPS interactions were signi cantly positive across all four cognitive capacity GPSs (Table and individuals with a higher cognitive GPS presented a slower trajectory of memory decline than those with a lower GPS (Figure 2(a)).This result indicates that the portions of cognitive ability under genetic in uence may serve as a 'buffer' against memory decline in aging.These observations align well with existing studies on the protective effect of education and intelligence on the occurrence of dementia [36].A close relationship between early-life education and intelligence with cognitive decline have been reported for dementia and AD [37].Even though it is not yet clear how early-life education and intelligence moderate the risk for dementia, our ndings suggest that individual variations of memory decline are closely associated with polygenic in uences of cognitive abilities.
Among the repeated assessments of 7 cognitive domains with an average 6.5-year interval, a decline of immediate memory recall in aging signi cantly correlated with the cognitive GPS but not the other cognitive domains did.Memory recall, assessed by immediate and delayed recall tests of words, is hippocampus dependent [38][39][40].We did not observe the signi cant interaction effect in the domain of delayed recall.It is interesting that the discovered genetic protective effect exerted speci cally on the hippocampus-related immediate memory recall.There are two implications worth noting.Firstly, given the speci city of the correlations among various cognitive domains, the genetic protective factor of immediate memory decline may be mediated via the hippocampus.Indeed, the hippocampus is the primary mediator of interventions for the cognitive wellness or dementia, such as aerobic tness [41], diet [42], and medication [43][44][45].This is closely related to the unique role of the hippocampus in neurogenesis and synaptic plasticity [46,47].Future research should thus test whether the hippocampus and hippocampal network underlies the genetic projective effect on immediate memory decline, but not in delayed recall, and if so seek to elucidate the mechanisms involved.Secondly, given the role of the hippocampal memory impairment in the pathophysiology of Alzheimer's disease (AD), our nding may lead to the potential link of the inherited genetic factor of cognitive resilience to the individual differences in hippocampal degeneration, as well as memory decline in AD [48,49].Testing this link will allow better strati cation of AD and monitoring the course of the disease by the individual-speci c genetic pro les of cognitive resilience.
Our PheWAS identi ed several phenome-wide associations between cognitive capacity GPSs and cognitive assessments.The Similarities task from WAIS, Number Series task and Digit Ordering task showed strongest associations across the four cognitive capacity GPSs in terms of effect size (β) and pvalue (Figure 1, Figure 1).These ndings suggest that the cognitive components required to successfully complete the Similarities, Number Series, or Digit Ordering tasks might strongly overlap with genetic components of cognitive capacities primarily exhibited by the domain of uid intelligence.We assume that the series of cognitive components involved in the Similarities and Number series tasks, such as logical memory, symbol search, and reasoning, might be closely linked to early-life cognition, all of which may serve as phenotypic indicators for uid intelligence.Our ndings are backed up by the previous knowledge that uid intelligence is considered to be more dependent on biological in uences and less dependent on past learning experiences than crystalized intelligence [50].
Our analysis identi ed several phenome-wide signi cant associations of cognitive capacity GPSs with several early-life cognitive phenotypes including IQ scores, educational attainment, or high school class rank.The signi cant genetic association between cognitive capacity and IQ scores or educational attainment has been well established in several GWAS studies on human intelligence [9,[13][14][15][16].IQ scores of the WLS respondents were derived from the Henmon-Nelson test of mental ability, which is regarded as a general measure of overall intelligence, capturing both uid and crystallized intelligence.
PheWAS of the behavioral phenome identi ed several behavioral traits having high relatedness to the genetic factors of cognitive capacity.All of the tested cognitive capacity GPSs positively correlated with Openness among the Big 5 Personality factors, and some GPSs negatively correlated with Neuroticism (Supplementary Table 1).The nding presents an interesting cross-trait hypothesis in which the variances of personality dimensions may be partially explained by genomic components of cognitive capacity, or vice versa.'Openness' could be regarded as the attitude and tendency to explore, detect, understand, and appreciate complicated patterns of new information through both the senses and in the abstract [51].Previous studies support our ndings, concluding that an overall open-minded attitude might positively in uence the long-term variances of cognitive abilities with willingness to explore [52].
No signi cant associations between Spouse IQ and cognitive abilities were identi ed, which indicates that the behavioral associations between assortative mating and cognitive abilities is unclear.In addition, a strong relationship between occupational income and several cognitive capacity GPS were found, which supports the existing studies on a strong association between general mental ability and job performance [53].
A few limitations of this study should be noted.The WLS had two time points for measuring changes in their cognitive assessments with an average 6.5 year interval.Adding more cognitive measurements through time will strengthen our ndings by more thoroughly monitoring cognitive changes over the lifetime.Also, the unexplored impact of other sociodemographic variables such as socioeconomic status, educational environment, lifestyles, or family structure, should be considered to better connect our theorical ndings with phenome-wide expression of cognitive abilities.In addition, we used Europeanancestry speci c summary statistics for constructing the cognitive capacity GPSs and applied them to participants with European-ancestry.Researchers should note that the translational application to non-European individuals could be different, and the results should be interpreted with caution.Future investigation is needed to elucidate heterogeneity between ancestry groups for the genetic underpinnings of cognitive abilities.Our ndings could serve as the rst cognitive-phenome map that describes the functional boundaries of human cognition from a genetic perspective, and the map could be further expanded with the advanced phenotyping of human cognition and behavior traits.<p><strong>Graphical results of our linear mixed-effect model analysis showing that individuals with a higher cognitive GPS presented a slower trajectory of memory decline than those with a lower GPS.&nbsp;</strong>Changes in seven cognitive assessments (<em>immediate recall task, category uency task, digit ordering task, delayed recall task, letter uency task, similarities task, and health utility index (HUI) level 3 cognition leve</em>l) and interaction effects of CP GPS were shown.The selected seven cognitive assessments were repeatedly administered to 8,511 European ancestry individuals between the average age of mid-50s (survey timepoint 1) and mid-70s (survey timepoint 3).<strong>&nbsp;&nbsp; </strong><strong>(a) Interaction plots showing the different slopes of age-dependent interaction effects by cognitive capacity GPS on the cognitive assessments.&nbsp;</strong>TheX-axis indicates the age of the WLS participants at survey timepoint, while the y-axis indicates each cognitive assessment score (zscored).The four lines indicates different slopes of individuals&rsquo; cognitive changes strati ed by GPS.The gray area represents 95% con dence interval of each slope.Similarities task were the only tasks that were repeatedly administered to the participants since timepoint 1 (Average participants&rsquo; age 48.6).<strong>(b).Bar plots showing the strati cation performance of cognitive

Declarations Figures
4 years), 64.2 years at the second assessment (WLS survey round 5, 2003-2005, SD=4.1 years), and 70.7 years at the time of the last assessment (WLS survey round 6, 2011, SD=4.2 years).The sample was 51.8% female, 47.8% completed high school education or less than one year of college (number of years of education), 78.2% were born in Wisconsin, USA.PheWAS of Cognitive GPSs in the Cognitive/Behavioral Phenome 1. Cognitive Phenotypes Across all of the PheWAS results, IQ score showed the strongest associations with the four cognitive GPSs in terms of p-value and the increase proportion of variance explained (strongest with CP GPS, p-value=7.74E-179,β=1.42, 95% CI=(1.32,1.51)) (Figure 1, Table

Table 1 .
). Phenome-wide association studies (PheWAS) signi cant results of the four cognitive capacity GPSs with the cognitive/behavioral phenotypes.The table presents only the phenotypes signi cantly associated with all four cognitive capacity GPSs (Educational attainment (EA), Cognitive Performance (CP), Math Ability (MA), Highest Math Class (HM)).The strongest cognitive capacity GPS-phenotype associations were presented from the full PheWAS results (available in Supplementary Table