ENIGMA DTI protocol development Working Group - Teleconference
April 5, 2012, 9am PST
Topics: Registration to the new DTI template created from 4 different cohorts (N=400), covariates to test, covariates to control for, creation of an
elderly template
●Participants:
○Peter Kochunov, David Glahn, Emma Sprooten, Andrew McIntosh, Mark Bastin, Ian Deary, Neda Jahanshad, Paul Thompson
●Introduction (Paul)
○ENIGMA paper: Thanks everyone for all the simultaneous edits on Google Docs. We returned corrections yesterday (Thursday), and the expected date of publication is April 15th.
●Status of template:
○Peter – template has been completed from 400 healthy subjects, aged 18-85, with 100 contributed from each cohort
■Average Jacobian of warps (when mapping individuals to a template) has been greatly reduced for LBC and UCLA
■Enigma template (L) much improved periphery relative to the JHU template (R)– previous warping to JHU template tries to warp peripheral tracts to essentially zero
■JHU ROIs fit very well on top ENIGMA template (see picture)
○Mark – also LBC is much improved because JHU template is all younger subjects and LBC is older
○Peter – also the resolution for JHU template is very low
○Neda – we can also use the new JHU “Eve” atlas parcellations for the ENIGMA template (see below).
■All in the same space as the prior JHU atlas ROI labels, so they should overlay nicely
■Has more peripheral regions segmented
○Peter – need to clean up skeleton first
○Paul - the low average Jacobian to this template would also allow us to do an unbiased TBM analysis and compare the results to DTI if we wanted to; it opens up that possibility as we have many studies of TBM heritability and SNP effects
●Tests to run on template and publications:
○Paul – we should try and publish several lines of work with this template to prove its utility; this will help the template be accepted when we suggest it for use by more groups in ENIGMA. Many will have their own template or may be using the JHU atlas, so if we publish a paper showing this helps, it will encourage people to use it.
○We could also jointly publish
■Voxelwise analysis
■Several lines of work can be done such that each group takes the lead on one project:
■for example
●Peter – main paper looking at age
●Neda – heritability
●Edinburgh – ApoE4
■Heritability can be run using Mx for UCLA, SOLAR for GOBS and Visscher’s method for BFS and LBC.
○David – using 3 different heritability estimates and getting 3 different results may bring about a lot of concerns and may not be the best approach
○Paul – template can be used for other (non-genetic) studies
■Age effects are known – great place to start
■Sex effects -less clear what they should be, and are debatable
■UCLA – ages 20-29 so may not be the best sample for picking up age effects, although there are 100+ additional subjects at ages 12 and 16
○Andrew – BFS has a 3-year age range so may not be useful for looking at age effects
○David – can try one atlas, or different ones for each age group, and compare the results
○Ian – also has narrow age range samples
■Each study (age/heritability/APOE) will be impactful
○Paul – we should try them all and have a distributed processing where different sites might take the lead on different papers
■May want to combine the main template and heritability paper, Peter could take the lead on it, with heritability estimates from twins and the pedigree. This would be interesting as there is some theoretical work on why heritability estimates may differ (slightly) when based on twin vs. family designs
○Template will be useful in 2 ways
■Regardless of ROIs – on a voxelwise level, we can use it to pool voxelwise analyses, done separately at each site using this template; voxelwise statistics can be pooled using conjunction tests or replications within an ROI, or even voxelwise meta-analysis
■And also additional studies can be done, with the availability of the ROIs
○Ian – would be interested in cognitive correlation maps, e.g. IQ
○Paul - yes we could certainly run those and contribute to a paper led in Edinburgh on that
○Peter – for heritability analysis will UCLA be willing to use SOLAR
○Neda - yes we could implement it to be runnable in the LONI pipeline
■Not able to integrate software at this time, as there have been many tweaks to the scripts, will need to be run by Peter/David
○Paul – we can compare this to standard ACE
○David – kinship matrices from both sites can be combined for a large heritability study
○Paul – generally an ACE approach is taken, can SOLAR output common environmental term (C)?
○David – yes this can be done, lets see what data looks like first
○Ian – yes I would expect to see a standard ACE test in a heritability paper
○Peter – using FDR to control for multiple comparisons across a skeleton derived from TBSS has concerns.
■Will work with Tom Nichols to validate a modified version of FDR for the skeleton and run permutations
○David – is on a consortium call (later today) about the genetics of alcoholism, as part of the PGC
■Do other groups have information on alcohol intake
■UCLA – no alcohol consumption (will double check) but alcohol related SNPs are genotyped and would be promising
■LBC may have some data
○Ian – interested in cognition
■LBC has cognition information in the 73 year olds, but also when the subjects were 11
■UCLA has ~70 12 year olds and ~70 16 year olds with IQ information but no GWAS yet, we could look at the IQ effect though easily
○Paul – Neda has a promising project on iron and would be interested in looking at serum iron measures including transferrin as we had a strong association between FA and the hemochromatosis gene HFE.
■David – GOBS has this info but no one has looked at it, will have to dig it up.
○WISHLIST up here on Google DOCS—please add to it and let other groups know if you have (and will share :) ) the information for joint analysis
○When FA and MD are aligned to the template, let’s do voxelwise tests of:
○age, sex
○heritability from ACE
○FSIQ/VIP/PIQ, any iron measures (if available)
○ApoE (if available)
○anyone’s favorite SNPs (e.g. Alzheimer’s risk SNP in CLU from UCLA paper Braskie et al., J. Neurosci., 2011)
○ICV (this might be interesting)
○Neurotcism, other personality variables
○Psychopathology: symptoms (e.g. hallucinations) syndromes (e.g. BD, MDD, SCZ)
○Substance misuse
○Polygenetic scores for MDD, BD, SCZ based on PGC data
●Covariates to control for:
○Neda – Peter mentioned last time that the amount of head motion may be a good covariate
○Peter – motion had no direct relation on FA, maybe not necessary to use it. When we adjusted for it the correlations with other covariates went up a bit from 0.37 to 0.41
○Neda – many sites have narrow age ranges and the sex-by-age interaction term or the age^2 terms may introduce a multicollinearity problem
■Should sites include covariates based on what fits?
■Should we also center the covariates?
○David – all sites should include the same set of covariates, and should be mean-centered
■Keep them in if it is not hurting the model
○Paul – we can try with and without ICV as well as there might be some interesting relation to FA
○David – include global FA as a covariate in the same sense as ICV was used as a covariate for hippocampus
○Paul - that’s a great idea as we could see if the effects are global or regionally specific - also the results may go away but then we know there’s only evidence for a global effect
●Other diffusion measures:
○Paul – should we look into other diffusion measures? MD/radial/axial? Reviewers often ask for these.
○Peter – Not me! Not a fan. FA is the most heritable among the most commonly used measures
■Axial/radial (the ratio) is more robust than each alone, as they appear to be correlated
■Diffusivity measures have a lot more subject–to– subject bias
○Mark – FA is also less affected by partial voluming with CSF
●Elderly template:
○Mark – going back to an elderly template, would it be possible to make one? Possibly using all 200 healthy and randomly chosen mixed indivuals
○Paul – ADNI also has elderly subjects with DTI without GWAS that can be added to the template, and we are now analyzing their DTIs; there are hundreds and we could try the elderly template versus the all-age tepm
○Peter – does not like the idea of having many templates, but it can be done and has all the data.
●Direct FA-skeleton Comparison between templates :
○Emma – has LBC TBSS data using the default template.
■Can we compare global FA measures taken with each template?
○Neda – sounds like a good idea
○Peter – not sure it is comparable as the skeletons can be different, we may be comparing 2 different signals
●Conclusions -- for the next call, in one month’s time
○Neda will clean up skeleton by next week
○All groups will have images registered to template using registration algorithm of choice and project on to ENIGMA skeleton; let’s try some covariates too
○Neda to send out ‘Doodle’ poll for May
○Don’t forget to add to the wishlist!