Dynamics of cognitive variability with age and its genetic underpinning in NIHR BioResource Genes and Cognition cohort participants

A leading explanation for translational failure in neurodegenerative disease is that new drugs are evaluated late in the disease course when clinical features have become irreversible. Here, to address this gap, we cognitively profiled 21,051 people aged 17–85 years as part of the Genes and Cognition cohort within the National Institute for Health and Care Research BioResource across England. We describe the cohort, present cognitive trajectories and show the potential utility. Surprisingly, when studied at scale, the APOE genotype had negligible impact on cognitive performance. Different cognitive domains had distinct genetic architectures, with one indicating brain region-specific activation of microglia and another with glycogen metabolism. Thus, the molecular and cellular mechanisms underpinning cognition are distinct from dementia risk loci, presenting different targets to slow down age-related cognitive decline. Participants can now be recalled stratified by genotype and cognitive phenotype for natural history and interventional studies of neurodegenerative and other disorders.

TMN and TMA.The trail making test consisted of two parts: (i) TMN and (ii) TMA.In the TMN task, participants were shown numbers 1-25 pseudo randomly on the screen and instructed to touch them in numeric order as quickly and accurately as possible.In the TMA task, participants were shown numbers 1-13 and letters A-L together pseudo randomly on the screen and instructed to touch first the number and then letter in numeric and alphabetical order (i.e., 1-A-2-B-3-C) as quickly and accurately as possible.The main outcome for both parts of the tests was mean response time in milliseconds to sequentially identify the correct element of the sequence, which was log10 transformed.Both tests assess executive functioning.
In the analysis, we considered both parts of the test separately.

MX.
Participants were shown a 3x3 matrix of tiles with one missing piece and provided with eight options at the bottom of the screen to select the missing tile that completed the 3x3 matrix displayed on top.There were 15 questions of this nature, each of which was timed for 60 seconds.If the time for a question expired, the app automatically progressed to the next question and considered the expired question incorrect.The primary outcome of this test was the total number of correct answers made, which was reversed prior to the analysis by using the total number of incorrect answers.This test measures non-verbal fluid intelligence.
WM.This test involved a number shown to participants for memorization, and a proportionate amount of time was given for memorization and recall depending on the length of the number (number of digits).
Correct recall of the number resulted in the number string expanding in length (by one digit).All numeric strings of numbers were randomly generated with an algorithm that prevents repeated numbers.Before starting the test, the participants were allowed to have unlimited attempts to practice with single-and double-digit length numbers.The test ended either when the participant made two incorrect recalls of number string consecutively or when the participant reached the end of the test (a 24-digit length string).
The primary outcome of this test was the total number of correct responses made by participants, transformed with log10 and reverse-scored in the similar way to SD prior to analysis.The WM test is also known as the numeric memory test.
QZ.This test comprised of 13 questions, which required participants to choose the correct answer from multiple choices (5 options, with one exception having 4 options).The test had a fixed duration of 2 minutes.After that, participants were not allowed to complete the rest of the test.The time taken to answer each question was recorded in milliseconds.The primary outcome of this test was the total number of correct responses made within 2 minutes.Prior to the analysis, test scores were reversed by using the total number of incorrect answers.This test is also recognized as a measure of fluid intelligence, which measures verbal and numerical reasoning abilities.
VY.This test involved answering 20 questions (stimuli provided in Supplementary Table 28).For each question, a word was shown on the left side of the screen and participants were provided with 6 words on the right side of the screen to choose the correct synonym.The time limit for each question was 21 seconds.
The primary outcome of this test was the total number of correct answers made, which was reversed by using the total number of incorrect answers.This test assesses crystallized intelligence.
PR.This test involved participants recalling the positions of different shapes.First, participants were presented with 7 different shapes in 7 different locations, one after another.Once all shapes were displayed, one of the shapes was shown in the middle of the screen and participants were invited to identify its correct location based on its previous appearance.The primary measure for this test was the total number of guesses required to complete the task which underwent a log10 transformation.This test is designed to measure episodic memory and attention.
G4 and G6.Initially, we performed PC analysis using all cognitive test scores.Eigenvalues depicted on the scree plot indicated three PCs where 6 tests (RT, TMN, TMA, SB, SI and SD) fall onto the 1 st PC (explaining 41.3% variance), 4 tests (WM, MX, QZ and VY) fall onto the 2 nd PC (explaining 15.6% of the variance) and PR falls onto the 3 rd PC.Based on the initial observations, we decided to create two measures of general cognitive ability.We used scores of RT, TMN, TMA, SB, SI and SD tests to create PCs.Scores on the first rotated PC, which accounted for 66.5% of the total variance (Extended Data Fig 3A and C), were used as a measure of general cognitive ability (we named it G6).The expected contribution of TMN, TMA and RT to G6 was lower than SB, SI and SD (Extended Data Fig 3B).Likewise, we used WM, MX, QZ and VY to create PCs and the first rotated PC, which accounted for 46.6% of the total variance (Extended Data Fig 4A and C), were used as the 2 nd measure of general cognitive ability (we named it G4).
The expected contribution of MX and WM to G4 was lower (Extended Data Fig 4B).We used the "prcomp" function from the stats package in R for PCA analysis.Note, PR was not included in the calculation of general cognitive ability.

Supplementary Fig. 4 | (A) Heatmap of genetic correlation for 11 cognitive tests and two measures of general cognitive ability (G4 and G6). Cells highlighted in red circle indicate non-significant correlation (two-tailed p>0.05), and (B) Heatmap of phenotypic correlation for 11 cognitive tests and two measures of general cognitive ability (G4 and G6). (A-B)
QZ: Quiz; WM: Working memory; MX: Matrices; SD: Symbols Digit; VY: Vocabulary; RT: Reaction Test; PR: Pairing 7; SB: Stroop Box; SI: Stroop Ink; TMN: Trail Making Numeric; TMA: Trail Making Alpha Numeric.Phenotypic and genetic correlation analyses was confined to participants for whom genetic data was available and completed all cognitive tests.In both plots, each coloured cell indicates magnitudes of correlations.Each coloured cell indicates magnitudes of correlations.The corresponding colour scale is presented on the right side of the heatmap where dark green represents the highest correlation, and darker moderate pink represents highest negative correlation.On the x-and y-axis, all cognitive phenotypes are presented maintaining the order.Supplementary Fig. 10 | Fine mapped SNPs for G4 and their intercorrelation.The strength of correlation is presented at the bottom of the plot.This plot was created using the LDmatrix tool (https://ldlink.nih.gov/?tab=ldmatrix) 95 .Pairing7; G6: a summary measure computed using RT, SB, SI, SD, TMN and TMA; G4: a summary measure computed using WM, MX, QZ and VY; B: Beta; SE: Standard Error; P: two-tailed p-value; P-trend: two-tailed p-value for trend.b Reference group: Low Multiple Deprivation Group; Linear regression models were commonly adjusted for age, age-square (except VY and PR), gender and types of devices used to take the test.An age by gender interaction term/s was included in the model for RT, SB, SI, SD, QZ, VY and G4.c P-values were presented after Bonferroni Holm adjustment for 13 tests.d Log10 transformation was performed for these test scores.

Table 1 |
Self-reported clinical information available on G&C study participants.All diagnoses were self-reported in response to questionnaire provided either by NIHR BioResource or Cognitive Test application.

Table 2 |
Types of transformation performed to cognitive test scores, what domain of cognition these tests cover and what the score means.SD, MX, WM, QZ, and VY scores were reversed (following the transformation for WM) using formula: max (test score)-tests score to have same direction of interpretation as other tests.

Table 3 |
Median and mean values for different cognitive test scores for all participants and for those whose genetic data were available.

Table 4 |
The association between cognitive phenotypes (including measures of general cognitive ability) and device used to take the cognitive tests.
bReference Category for Devices: iOS users.c Model adjusted for age and gender.

Table 5 |
The differences in age, deprivation, and education among device users.

Table 6 |
Study characteristics of participants with genetically inferred European ancestry.

Table 8 |
The association between cognitive phenotypes (including measures of general cognitive ability), age, and sex.Linear regression models were commonly adjusted for age, age-square (except VY and PR), gender and types of devices used to take the test.An age by gender interaction term/s was included in the model for RT, SB, SI, SD, QZ, VY and G4.Log10 transformation was performed for these test scores.
b Reference Category for gender: Female.c a RT: Reaction Test; SB: Stroop Box; SI: Stroop Ink; SD: Symbols Digit; TMN: Trail Making Numeric; TMA: Trail Making Alpha Numeric; MX: Matrices; WM: Working memory; QZ: Quiz; VY: Vocabulary; PR: Pairing7; G6: a summary measure computed using RT, SB, SI, SD, TMN and TMA; G4: a summary measure computed using WM, MX, QZ and VY; B: Beta; SE: Standard Error; P: two-tailed p-value; P-trend: two-tailed p-value for trend.b Reference group: Highest education; c P-values were presented after Bonferroni Holm adjustment for 13 tests.d

Table 10 |
The association between cognitive phenotypes (including measures of general cognitive ability) and multiple deprivation.

Table 12 |
cohort.The Association between cognitive traits (including a measure of general cognitive ability), age and APOE.
a RT: Reaction Test; SB: Stroop Box; SI: Stroop Ink; SD: Symbols Digit; TMN: Trail Making Numeric; TMA: Trail Making Alpha Numeric; MX: Matrices; WM: Working memory; QZ: Quiz; VY: Vocabulary; PR: Pairing7; G6: a summary measure computed using RT, SB, SI, SD, TMN and TMA; G4: a summary measure computed using WM, MX, QZ and VY; B: Beta value from regression model; SE: Standard Error; P: two-tailed p-value.bEach of the linear regression model was adjusted for age (centred to mean), age square, gender and types of devices used.cP-valuepresentedafterBonferroni-Holmcorrectionsfor13 tests.Non-significant p-values are highlighted in black.dLog10transformationwasperformedforthesetest scores.Supplementarya RT: Reaction Test; SB: Stroop Box; SI: Stroop Ink; SD: Symbols Digit; TMN: Trail Making Numeric; TMA: Trail Making Alpha Numeric; QZ: Quiz; PR: Pairing 7; G6: a summary measure computed using RT, SB, SI, SD, TMN and TMA; B: Beta value from linear mixed-effect model; SE: Standard Error; P: two-tailed p-value; Associations were adjusted for sex, the first five genomic principal components, devices, batch, array (as random effect).be3/e3carrierswereservedas reference.cP-valueswerepresentedafterconsidering Bonferroni-Holm adjustment, considering the number of cognitive traits each covariate was tested against.P-values significant before correction are indicated with "#".dLog10 transformation was performed for these test scores.#Significantat<0.05beforeBonferroni Holm correction.bHighPRS group was used as reference.cP-valuesarepresentedafterBonferroni-Holm corrections for 9 tests.P-values significant before the multiple testing correction are indicated with " # ". d Log10 transformation was performed for these test scores.#Significantat<0.05before Bonferroni Holm correction.##Significant at <0.01 before Bonferroni Holm correction.Supplementary Table16| Heritability for cognitive tests and two summary measures of general cognitive ability (G4 and G6) in G&C cohort.
a RT: Reaction Test; SB: Stroop Box; TMN: Trail Making Numeric; VY: Vocabulary; B: Beta value from linear mixed effect model; SE: Standard Error; P: twotailed p-value; Associations were adjusted for sex, the first five genomic principal components, devices, batch, array (as random effect).b High PRS group was used as reference.c P-values are presented after Bonferroni-Holm corrections for 4 tests.d Log10 transformation was performed for these test scores.

Table 23 |
Replication of G4 and G6 associated independent SNPs in previously published genome-wide association meta-analysis of human intelligence or general cognitive ability.Beta: coefficient from BOLT infinitesimal model; SE: standard error; P: two-tailed p-value; INFO, estimated imputation score.We harmonised summary statistics for both SNPs manually.We harmonised summary statistics for both SNPs manually.The direction of beta values was reversed as scores for the measure of general cognitive ability or intelligence were calculated using cognitive test scores where higher scores indicated better performances.

Table 24 |
List of candidate SNP IDs for G4 annotated to have an exonic function.

Table 25 |
Genes prioritised by genome-wide gene-based association analysis, positional, eqtl and chromatin interaction mapping.

Table 26 |
genes prioritised by positional mapping strategy.Eqtl_mapping column presents genes prioritised by eQTL mapping strategy.Ci_mapping column presents genes prioritised by chromatin interaction mapping strategy.Top 10 clusters with their representative enriched terms for G4."Log10(P)" is the p-value (computed using the cumulative hypergeometric distribution) in log base 10. "Log10(q)" is the multi-test adjusted p-value in log base 10.

Table 27 |
Estimated genetic correlation for G4 and G6 with adulthood IQ, childhood IQ, educational attainment, and Alzheimer's Disease.PMID: PubMed Identifier; rg: estimate of genetic correlation; se: Standard Error; P: two-tailed p-value.*Positive correlations indicate worse performance of trait is positively associated with worse performance on G4 and G6. a

Table 28 |
Stimuli for the vocabulary test.