Functional connectivity analyses in imaging genetics: considerations on methods and data interpretation

Abstract

Functional magnetic resonance imaging (fMRI) can be combined with genotype assessment to identify brain systems that mediate genetic vulnerability to mental disorders ("imaging genetics"). A data analysis approach that is widely applied is "functional connectivity". In this approach, the temporal correlation between the fMRI signal from a pre-defined brain region (the so-called "seed point") and other brain voxels is determined. In this technical note, we show how the choice of freely selectable data analysis parameters strongly influences the assessment of the genetic modulation of connectivity features. In our data analysis we exemplarily focus on three methodological parameters: (i) seed voxel selection, (ii) noise reduction algorithms, and (iii) use of additional second level covariates. Our results show that even small variations in the implementation of a functional connectivity analysis can have an impact on the connectivity pattern that is as strong as the potential modulation by genetic allele variants. Some effects of genetic variation can only be found for one specific implementation of the connectivity analysis. A reoccurring difficulty in the field of psychiatric genetics is the non-replication of initially promising findings, partly caused by the small effects of single genes. The replication of imaging genetic results is therefore crucial for the long-term assessment of genetic effects on neural connectivity parameters. For a meaningful comparison of imaging genetics studies however, it is therefore necessary to provide more details on specific methodological parameters (e.g., seed voxel distribution) and to give information how robust effects are across the choice of methodological parameters.

Figure 1. Seed Voxel Localization.

Top: Frequencies of MNI-coordinates in the X, Y, and Z dimension for right DLPFC seed regions identified with the global maximum and the next local maximum starting at 52, 30, 30. Bottom: Individual seed-region localization in the right DLPFC for the functional connectivity analyses based on the global maximum (left) and the next local maximum approach (right).

Figure 2. Influence of Seed Voxel Localization on the Comparisons of Genetic Groups.

Left: Interaction between the factors method (next local maximum vs. global maximum) and rs1006737 genotype (G/G, G/A, A/A) in the left anterior HF. Right: A post-hoc analysis of the parameter estimates of each risk group separately for both methods. A significant linear gene-dosage effect is found only with one method of seed region selection, the next local maximum approach.

Figure 3. Influence of Noise Regression Methods on Connectivity Patterns.

Functional connectivity group results (using one sample t-tests) with respect to the right DLPFC for the “standard implementation” of the connectivity analysis (top) and the additional global signal correction by application of the global normalization procedure (bottom). Yellow-red: positive correlations with seed voxel, green-blue: negative correlations with seed voxel.

Figure 4. Influence of Noise Regression Methods on the Comparisons of Genetic Groups.

A linear increase with with rs1006737 genotype in functional coupling between the right DLPFC (seed voxel) and the right hippocampus is detected for both noise reduction algorithms. However, the interpretation of the changes depending on the underlying correlation profile. While in the “standard implementation” one would argue that the hippocampus is decoupled in GG carriers and that coupling increases with genetic risk (GG

Figure 5. Influence of Seed Voxel Coordinates as Covariates on the Group Level.

Reduction of genotype effects after entering seed-voxel localization as covariate on the second level. Left: A linear decrease in regional activation is found with number rs1006737 risk alleles in the right DLPFC (blue). A similar gene-dosage dependent effect in functional connectivity was found at the same location (red). After controlling for differences in the spatial distribution of seed-voxel localization with introduction of the MNI coordinates in the x-, y-, and z-dimension as covariate on the second level the associations of genotype with DLPFC connectivity was reduced and non-significant at p>.05 corrected