Mendelian randomization suggests a bidirectional, causal relationship between physical inactivity and obesity

This is a repository of the Mendelian Randomization analysis ran for the publication: Carrasquilla, G. D., García-Ureña, M., Fall, T., Sørensen, T. I. A., & Kilpeläinen, T.O. (2022). Mendelian randomization suggests a bidirectional, causal relationship between physical inactivity and adiposity. eLife, 11, e70386.

The repository consists of two sections:

1. The folder structure and code to reproduce our analysis in the /R folder.
2. The variants and summary statistics used to produce our results, plus a code to easily navigate through them in /data_&_results

GWAS summary statistics used:

In this publication we are testing whether there is any causal association between adiposity and physical activity or inactivity. To do so we used the latest and largest GWAS summary statistics for each of the traits. The main analysis are performed for Body Mass Index (BMI) and secondary analysis for three other anthropometric traits.

Physical activity and inactivity traits

For physical activity we decided to use accelerometer data for vigorous physical activity, moderate physical activity and sedentary time from two different sources.

Vigorous physical activity from the acc425 model from Klimentidis et al 2018 available here: (https://drive.google.com/drive/folders/1p2-aKT6GgOv4425yaIcvO0nN30O4mpPh).

Moderate physical activity form Doherty et al available here: (https://ora.ox.ac.uk/objects/uuid:ff479f44-bf35-48b9-9e67-e690a2937b22/download_file?file_format=gzip&safe_filename=Doherty-2018-NatureComms-moderate.csv.gz&type_of_work=Dataset).

And sedentary time from Doherty et al available here: (https://ora.ox.ac.uk/objects/uuid:ff479f44-bf35-48b9-9e67-e690a2937b22/download_file?file_format=gzip&safe_filename=Doherty-2018-NatureComms-sedentary.csv.gz&type_of_work=Dataset).

Anthropometric traits:

As explained before, the main analysis are for BMI as the main adiposity trait, though we also ran analysis for other traits that inform about central obesity or abdominal fat accumualation: body fat percentage, waist circumference adjusted for BMI and waist-to-hip ratio adjusted for BMI.

BMI GWAS summary statistics from Pulit et al can be found here: (https://zenodo.org/record/1251813/files/bmi.giant-ukbb.meta-analysis.combined.23May2018.txt.gz?download=1)

WHRadjBMI GWAS summary statistics from Pulit et al can be found here: (https://zenodo.org/record/1251813/files/whradjbmi.giant-ukbb.meta-analysis.combined.23May2018.txt.gz?download=1)

WCadjBMI GWAS summary statistics from Shungin et al 2015 can be found here: (https://portals.broadinstitute.org/collaboration/giant/images/7/73/GIANT_2015_WCadjBMI_COMBINED_EUR.txt.gz)

BFP GWAS summary statistics from Elsworth et al 2018 can be found here: (https://gwas.mrcieu.ac.uk/datasets/ukb-b-8909/)

All code in this github relies on having the raw data from these sources in the RAW_DATA folder

Running CAUSE:

The analysis for this publication started in 03/2020 and, thus, we used the first version of CAUSE, which the authors used to publish in Nature Communication: (link.) Thus, to reproduce our results you should have the versions of the following packages installed:

devtools::install_version("mixsqp", version = "0.1-97", repos = "http://cran.us.r-project.org")
devtools::install_version("ashr", version = "2.2-32", repos = "http://cran.us.r-project.org")
devtools::install_github("jean997/cause@v1.0.0")

Curating WCadjBMI and BFP data:

While BMI and WHRadjBMI data from Pulit et al 2018 are relatively straighforward to use, WCadjBMI and BFP are not. In the following section, we describe what do you need in order to curate this data and obtain the data that you can find in data_and_results for these two traits.

Curating WCadjBMI:

For curating WCadjBMI you will need to follow the code on /R/Curating_GWAS/2_Curating_WCadjBMI.R. Checking that code you will realise that you will need additional data. Why is that? Because WCadjBMI GWAS summary statistics only present rsIDs and in this publication we merge GWAS summary statistics using chromosome and basepair positions in build 37 since the overlap with rsID might not be ideal due to mismatches caused by merged rsIDs.

This problem can be solved using the chromosome and basepair position from the WHR GWAS from the same publication as WCadjBMI, allowing for an almost perfect match. Please, follow the code in /R/Curating_GWAS/1_Curating_WHR.R to obtain the curated WHR GWAS summary statistics that will be used to obtained the curated WCadjBMI summary statistics. The GWAS summary statistics for WHR from Shungin et al 2015 can be found in: https://portals.broadinstitute.org/collaboration/giant/images/5/54/GIANT_2015_WHR_COMBINED_EUR.txt.gz

Importantly, while we are trying to match with chromosome and position, not all GWAS summary statistics possess them and, even if we do our best to obtain them, we might miss some of them. For those rsIDs without chromosome and position in build 37, the matching is done through rsID..., but as stated before, there might be mismatches due to merged rsIDs (rsIDs that reference the same basepair position, but one is deprecated and no longer used in recent reference panels.)

To allow the best matching with rsID, we always use a dictionary of synonim rsIDs before running CAUSE. The dictionary can be downloaded: ftp://ftp.ncbi.nlm.nih.gov/snp/organisms/human_9606/database/organism_data/RsMergeArch.bcp.gz

Curating BFP:

For BFP, you just need to follow the code on /R/Curating_GWAS/BFP_curation/Curating_BFP.R to clean the original vcf file.

Running 2SMR:

The 2SMR analysis are sensitive to sample overlap, so we decided to use GIANT BMI summary statistics, since moderate and vigorous physical activity and sedentary time GWAS are from UK Biobank, exclusively. You can find the original GIANT BMI summary statistics here: https://portals.broadinstitute.org/collaboration/giant/images/1/15/SNP_gwas_mc_merge_nogc.tbl.uniq.gz

To run the analysis following the code in the folder /R/2SMR/running_2SMR, you will need to curate the data using the code in /R/2SMR/curating_data_4_2SMR/.

Packages requiered

All 2SMR codes start by loading several libraries.

All of the following can be downloaded using install.packages() function:

library(ggplot2)
library(ggrepel)
library(dplyr)
library(gridExtra)
library(rmarkdown)
library(data.table)
library(jsonlite)
library(httr)
library(tidyverse)
library(phenoscanner)

Though TwoSampleMR needs remotes:

install.packages("remotes")
remotes::install_github("MRCIEU/TwoSampleMR") #analysis were performed with version 4.26

Curating BMI data

The GIANT BMI summary statistics do not have chromsome and position. It also presents old rsIDs that have been merged into newer ones. To clean this data, we used the chromosome and positions and rsIDs from Pulit et al BMI, dbSNP147 and the merged variants dictionary that can be downloaded from: ftp://ftp.ncbi.nlm.nih.gov/snp/organisms/human_9606/database/organism_data/RsMergeArch.bcp.gz.

Importantly, in the code found in /R/2SMR/curating_data_4_2SMR/ the function PS_query() uses dbSNP147 data that can only be obtained by asking permission to PhenoScanner. The code in this repository does not showcase any of that data, in agreement with PhenoScanner. Importantly, to reproduce correctly how PS_query works, dbSNP147 data should be organized in a specific way. How to handle that data and organize it in the same way to obtain the results can be found in this other repository: https://github.com/MarioGuCBMR/local_SNP_query_in_PhenoScanner. Once again, neither in this repository nor in local_SNP_query_in_Phenoscanner dbSNP147 data can be found or obtained by using theirs codes, as agreed with PhenoScanner after obtaining it.