3862
Identifying confounds in human brain functional connectomes: physiological and cognitive contributions1Monash Institute of Cognitive and Clinical Neurosciences and School of Psychological Sciences, Monash University, Melbourne, Australia, 2Monash Biomedical Imaging, Monash University, Melbourne, Australia, 3Centre of Excellence for Integrative Brain Function, Australian Research Council, Melbourne, Australia
Synopsis
Inter-subject analysis of brain function using fMRI is hampered by physiological factors and the neurovascular origin of the semi-quantitative blood oxygenation level dependent signal. A common technique to address these issues is to compute the individual subject fMRI signal correlations between brain regions and compare the functional connectomes across subject groups. In this work, we examined the effect of physiological factors that can contaminate functional connectomes in a large cohort of healthy elderly people. To quantify the impact of the physiological variance we computed the physiological connectome and modelled its impact on the relationship between the functional connectome and cognitive performance.
Introduction
The blood-oxygen level dependent (BOLD) effect is used to infer brain function from the haemodynamic response to neural activity. Each individual’s BOLD signal is dependent upon several physiological variables, including the concentration of red blood cell proteins (haemoglobin).1 Temporal dynamics from non-cerebral origins, e.g. heart-rate, can induce artefacts in derived measures of the BOLD signal.2 While physiological dynamics are commonly filtered, additional variance from static physiological factors may remain. In order to compare BOLD derived measures between individuals, network models are computed, e.g. functional connectomes (FCs). These approaches propose that temporal correlations represent neuronal coherence,3 and do not contain variance from temporally invariant physiological factors (e.g. haemoglobin). In this study, we examine whether comparisons between functional connectomes are confounded by physiological variance, and whether these confounds impact the relationship between functional connectomes and cognitive measures.
Methods
The study used data from 518 participants (age=73.8±3.5, 248 females) from the ASPREE-NEURO study,4 a sub-study of a clinical trial of aspirin in neurotypical adults aged 70 years and over.5 Here, data was acquired prior to the administration of study medication using a 3T Skyra MRI with a 32-channel head and neck coil (Siemens, Erlangen, Germany). The fMRI protocol was an eyes-open resting-state multi-band EPI sequence (multiband factor=3, TE=21ms, TR=754ms, voxel=3.0mm isotropic, matrix 64×64, slices=42, TA=5.16mins). A T1-Weighted MPRAGE scan was acquired for registration (TE=2.13ms, TR=2300ms, TI=900ms, voxel=1.0mm isotropic, matrix=240x256x192, flip angle=9). Haemoglobin level was assessed as part of a full blood work up.
The fMRI data was pre-processed for geometry distortions, motion (MCFLIRT),6 brain extraction (BET),7 high-pass filtering (0.01 Hz), de-noising (FIX and MELODIC),8–10 and linearly registered to T1 (FLIRT).6,11 The T1 was non-linearly registered to MNI (FNIRT, http://fsl.fmrib.ox.ac.uk). A full description of the fMRI pre-processing can be found in a previous publication.12 The brain was parcellated into 82 regions (Desikan-Killiany atlas).13,14 Per-subject FCs were calculated using temporal correlations (Pearson) between each pair of regions as edge-weights.
To identify physiological residuals within the FCs, inter-subject correlations between FC edge-weights and haemoglobin level were estimated to infer a physiological connectome.
To quantify the impact of physiology on a functional interpretation of the connectome, a moderation model was tested. Recall rates on the Hopkins verbal learning test (HVLT)15 were correlated with each FC edge, with and without an interaction term (FC x haemoglobin). A permutation test was performed to estimate the likelihood of the observed change in functional correlation with and without this second term, i.e., the change in the correlation coefficient between HVLT and FC in the two models.
In all connectomes, the network-based statistic was used to assess the significance of effects at a level of 0.05.16
Results
A physiological connectome with a large number of significant edges was identified (Figure 1). 806 (12%) of the possible connectome edges correlated with haemoglobin levels beyond a significance threshold. The distribution of the edges appeared to be widespread.
The moderation connectome is displayed in Figure 2, depicting correlations between task performance and the moderation term (FC x haemoglobin). 276 of the possible connections (4%) show a significant correlation, indicating that the relationship between the FC and task performance is indeed moderated by haemoglobin levels.
When accounting for the physiological contribution (Figure
3), 500 (7.4%) of the total connections in the brain are significantly
different to those calculated using a direct task to FC model.
Discussion
The identification of a physiological connectome provides the first evidence that even after filtering and de-noising the regional correlations between fMRI signals may be influenced by physiological factors. Co-variance between haemoglobin and edge-weights in this healthy aged cohort revealed a significant physiological contribution to functional connectome variability. Further, this contribution significantly altered the relationship between performance on the Hopkins verbal learning test and functional connectomes in healthy aged subjects. While it is plausible that inter-subject physiological variability may be an underlying factor that influences both cognition as well as haemoglobin levels, our conjecture is that the variance is due to incomplete modelling of the haemoglobin contribution to the BOLD signal model.1
Conclusion
The identification of residual physiological contributions to the human functional connectome suggests the current assumption that intra-subject functional correlations are insensitive to static physiological factors is incorrect. Functional connectomes may be confounded by inter-subject physiological variance, which contributes to the relationship between functional connectomes and cognitive measures. Extensions to incorporate temporally invariant physiological measures, such as haemoglobin, into network models are necessary, and further work is required to ascertain the influence of other physiological factors and their impact on human cognitive domains.Acknowledgements
We would like to thank the ASPREE Group, the radiographers and technical staff who collected the data, and all the participants for volunteering their time.References
1. Ogawa S, Menon RS, Tank DW, et al. Functional brain mapping by blood oxygenation level-dependent contrast magnetic resonance imaging. A comparison of signal characteristics with a biophysical model. Biophys J 1993;64(3):803–12.
2. Chang C, Cunningham JP, Glover GH. Influence of heart rate on the BOLD signal: The cardiac response function. NeuroImage 2009;44(3):857–69.
3. Biswal B, Yetkin FZ, Haughton VM, Hyde JS. Functional connectivity in the motor cortex of resting human brain using echo-planar mri. Magn Reson Med 1995;34(4):537–41.
4. Ward SA, Raniga P, Ferris NJ, et al. ASPREE-NEURO study protocol: A randomized controlled trial to determine the effect of low-dose aspirin on cerebral microbleeds, white matter hyperintensities, cognition, and stroke in the healthy elderly. Int J Stroke 2017;12(1):108–13.
5. McNeil JJ, Woods RL, Nelson MR, et al. Baseline Characteristics of Participants in the ASPREE (ASPirin in Reducing Events in the Elderly) Study. J Gerontol Ser A 2017;72(11):1586–93.
6. Jenkinson M, Bannister P, Brady M, Smith S. Improved Optimization for the Robust and Accurate Linear Registration and Motion Correction of Brain Images. NeuroImage 2002;17(2):825–41.
7. Smith SM. Fast robust automated brain extraction. Hum Brain Mapp 2002;17(3):143–55.
8. Salimi-Khorshidi G, Douaud G, Beckmann CF, Glasser MF, Griffanti L, Smith SM. Automatic denoising of functional MRI data: Combining independent component analysis and hierarchical fusion of classifiers. NeuroImage 2014;90:449–68.
9. Smith SM, Hyvärinen A, Varoquaux G, Miller KL, Beckmann CF. Group-PCA for very large fMRI datasets. NeuroImage 2014;101:738–49.
10. Beckmann CF, DeLuca M, Devlin JT, Smith SM. Investigations into resting-state connectivity using independent component analysis. Philos Trans R Soc B Biol Sci 2005;360(1457):1001–13.
11. Jenkinson M, Smith S. A global optimisation method for robust affine registration of brain images. Med Image Anal 2001;5(2):143–56.
12. Jamadar SD, Sforazzini F, Raniga P, et al. Sexual Dimorphism of Resting-State Network Connectivity in Healthy Ageing. J Gerontol Ser B [Internet] 2018;
13. Desikan RS, Ségonne F, Fischl B, et al. An automated labeling system for subdividing the human cerebral cortex on MRI scans into gyral based regions of interest. NeuroImage 2006;31(3):968–80.
14. Fischl B, Salat DH, Busa E, et al. Whole Brain Segmentation: Automated Labeling of Neuroanatomical Structures in the Human Brain. Neuron 2002;33(3):341–55.
15. Brandt J, Benedict RH. Hopkins verbal learning test–revised: professional manual. Psychological Assessment Resources; 2001.
16. Zalesky A, Fornito A, Bullmore ET. Network-based statistic: Identifying differences in brain networks. NeuroImage 2010;53(4):1197–207.
Figures
