The effects of genetic and modifiable risk factors on brain regions vulnerable to ageing and disease

We have previously identified a network of higher-order brain regions particularly vulnerable to the ageing process, schizophrenia and Alzheimer’s disease. However, it remains unknown what the genetic influences on this fragile brain network are, and whether it can be altered by the most common modifiable risk factors for dementia. Here, in ~40,000 UK Biobank participants, we first show significant genome-wide associations between this brain network and seven genetic clusters implicated in cardiovascular deaths, schizophrenia, Alzheimer’s and Parkinson’s disease, and with the two antigens of the XG blood group located in the pseudoautosomal region of the sex chromosomes. We further reveal that the most deleterious modifiable risk factors for this vulnerable brain network are diabetes, nitrogen dioxide – a proxy for traffic-related air pollution – and alcohol intake frequency. The extent of these associations was uncovered by examining these modifiable risk factors in a single model to assess the unique contribution of each on the vulnerable brain network, above and beyond the dominating effects of age and sex. These results provide a comprehensive picture of the role played by genetic and modifiable risk factors on these fragile parts of the brain.

Nested variables were resolved based on UK Biobank information.For example, subjects who answered "Never" for the variable "Frequency of drinking alcohol" were originally coded as having missing values for the subsequent variable "Frequency of consuming six or more units of alcohol".To resolve this issue, subjects who never drink alcohol were also coded as subjects never drinking six or more units of alcohol.Similarly, subjects who answered "No" for the variables "Ever had prolonged feelings of sadness or depression" and "Ever had prolonged loss of interest in normal activities" were originally coded as having missing values in the subsequent variable "Lifetime number of depressed periods".For the purpose of our study, these values were instead recoded as "0" for the last question.
Categorical variables were transformed into binary, either by merging the same variables together (e.g.regularly takes medication for diabetes, a question that is asked separately for each gender), collapsing similar set of answers within the same question (e.g.leisure/social activities: attending gym OR pub OR education class, etc.) or splitting the original variable with a given number of x answers possible into the same x number of binary variables (e.g.leisure/social activities: gym ONLY, pub ONLY, education class ONLY, etc.).
Probability for two hits to be in PAR1 If our 7 significant hits can be found anywhere across the whole genome, the probability that 2 out of 7 are in PAR1 is as follows: g = length of human genome in bp: 3,053,521,184 h = length of PAR1 in bp: 2,639,519 (g and h in h19 lengths accounting for double counting or not counting some of Y) Chance (as a probability between 0 and 1): (h/g)^2×(1-h/g)^5 = 7.4×10 -7 In addition, if we consider as null hypothesis that "hits are distributed between PAR1 and the rest of the genome according to the probabilities implied by a uniform distribution over all loci on the genome, or that hits are more likely to arise in the rest of the genome rather than in PAR1" (i.e., under H0 each hit has a probability <= h/g of being in PAR1, and under the alternate hypothesis each hit has a probability > h/g of being in PAR1), the frequentist test for this situation is the binomial test (one tailed, with alternative greater), with 7 trials, 2 successes,

Modifiable risk factors two-stage analysis
There is substantial redundancy within each MRF category.Moreover, not all UK Biobank participants provided data for all variables; an analysis limited to those with complete data would be biased, and based on a small, low-powered sample.
We addressed both issues via a two-stage analysis in which first we identified which variable within a category best represents eventual associations of that category with the LIFO brain network loadings.Once this had been established, we investigated the unique contribution of that category, over and above all other categories, to the LIFO loadings, while comprehensively correcting for multiple testing with the conservative Bonferroni method.
While the second stage consisted of only one model, that was one of a large set of models that could have been investigated if complete data were available and any single (i.e., not necessarily the "best") variable were allowed to be used to represent each of the 15 MRF categories in the second stage.The number of such tests is thus: where: Nk = total number of MRFs per category, k = category.Selection of the "best" variable provided an algorithmic shortcut that bypassed the need for these many tests and further addressed issues related to missingness.However, correction for these many tests was necessary, otherwise the screening provided by the first stage would render the analysis circular.While these many tests are not independent, we again took the conservative Bonferroni method, thus with a two-tailed significance cut-off of P = 0.05/(2×N) = 0.05/[2×(5.41×10 14)] = 4.62×10 −17 .

Full output of the mediation analyses on the dominant and recessive models
Lead bi-allelic variant from cluster 5 on Alzheimer's disease via LIFO probability of success h/g.It returns, when implemented in R: P = 1.56×10 -5 Testing for differences in the reduced sample of n = 35,527 In order to verify possible sub-sampling bias induced by the selection of the complete cases, two-sample Kolmogorov-Smirnov test was used to compare the distributions.None of the modifiable risk factors (n = 12) from the Stage 2 analysis were significantly different between the original sample and the reduced, complete sample after correction for multiple comparisons across those factors: diabetes diagnosed by doctor Puncorr = 1, nitrogen dioxide air pollution in 2005 Puncorr = 0.713, alcohol intake frequency Puncorr = 0.713, sleep duration Puncorr = 0.997, waist circumference Puncorr = 0.021, past tobacco smoking Puncorr = 1, medication for blood pressure Puncorr = 1, frequency of stairs climbing in last 4 weeks Puncorr = 0.882, hearing difficulty/background Puncorr = 1, medication for pain relief Puncorr = 1, pub or social clubs Puncorr = 1, medication for cholesterol Puncorr = 1.
brain network