Structural connectivity mediates the relationship between blood oxygenation and cognitive function in sickle cell anemia

Abstract

In sickle cell disease, the relative importance of reduced hemoglobin (Hb) and peripheral oxygen saturation on brain structure remains uncertain. We applied graph-theoretical analysis to diffusion magnetic resonance imaging data to investigate the effect of structural brain connectivity on cognitive function, alongside the presence or absence, number, and volume of silent cerebral infarction. In patients, we investigated the relationships between network properties, blood oxygenation, and cognition (working memory and processing speed indices). Based on streamline counts and fractional anisotropy, we identified a subnetwork with weakened connectivity in 92 patients with sickle cell disease (91 homozygous for HbS [HbSS], 1 heterozygote with HbSβ0 thalassemia; 49 males; aged 8.0 to 38.8 y), compared with 54 control subjects (22 males; aged 6.7 to 30.6 y). Multiple regression analyses showed a significant effect of Hb on full-network edge density (P < .05) and of peripheral oxygen saturation on streamline-weighted subnetwork efficiency (P < .01). There were effects of fractional anisotropy-weighted full-network and subnetwork efficiency on working memory index (both P < .05), and of streamline-weighted subnetwork efficiency on processing speed index (P = .05). However, there were no effects of presence, number or volume of silent cerebral infarcts. Streamline-weighted efficiency was progressively lower with lower oxygen saturation, with a downstream effect on the processing speed index. In path analysis, indirect relationships between blood oxygenation and cognition, mediated by network properties, were better supported than direct alternatives, with an indirect relationship between low oxygen saturation and processing speed index in patients, mediated by structural connectivity efficiency in a subnetwork of the brain differing from control subjects. Our findings are consistent with the notion that cognitive impairment is primarily mediated by hypoxic-ischemic effects on normal-appearing white matter and highlight the utility of network-based methods in providing biomarkers of cognitive dysfunction in patients with sickle cell disease.

Oxygen saturation, brain white matte connectivity and processing speed in sickle cell disease

Figure 1.

Recruitment flowchart for the study.

Figure 2.

Topological and standard-space representations of the subnetworks differing between patients and control subjects for the streamline-weighted case (top, cyan) and the fractional anisotropy-weighted case (bottom, magenta). Left column: axial topological view of the whole network, with subnetwork nodes and edges colored. Edge opacity is proportional to weight. Middle column: axial view of the brain, with regions involved in the subnetwork colored. Right column: sagittal view of the brain, with regions involved in the subnetwork colored. The left of the brain is shown on the left in axial views; sagittal views are from the left.

Figure 3.

Matrix of marginal Pearson correlation coefficients between blood oxygenation, network, and cognitive variables in patients. Pale colors indicate weak correlations; stronger negative correlations are shown in darker shades of blue, and stronger positive correlations in darker shades of red. Absolute r-values >0.2 are also annotated in the appropriate cell. fED, full-network edge density; fGE, full-network global efficiency; FSIQ, full-scale intelligence quotient; PSI, processing speed index; sGE, subnetwork global efficiency; SpO_2, peripheral oxygen saturation; WMI, working memory index.

Figure 4.

Boxplots of subnetwork efficiency in each of 3 patient subgroups, divided in order by measured SpO₂, and for the control group for comparison. SpO₂, peripheral oxygen saturation.

Figure 5.

Path analysis diagram showing direct and indirect relationships between blood oxygenation measures (red boxes, left) and cognition (orange boxes, right) in patients. Mediation pathways via unweighted (blue), streamline-weighted (green), and fractional anisotropy-weighted (lavender) network measures appear toward the center of the diagram. Standardized coefficients are shown on solid lines for relationships with P < .05. The double-headed arrow between processing speed index and working memory index reflects a residual covariance term. FA, fractional anisotropy; fED, full-network edge density; fGE, full-network global efficiency; FSIQ, full-scale intelligence quotient; PSI, processing speed index; sGE, subnetwork global efficiency; SpO₂, peripheral oxygen saturation; WMI, working memory index.