Effective connectivity reveals a shift to left-hemisphere dominance for spatial attention with diminishing arousal

In healthy righthanders, spatial attention typically becomes biased to the right side when alertness wanes. The underlying neural mechanisms of healthy spatial attention and these drowsiness-related spatial biases remain contested. Competing theories propose that attention is governed either by the interaction of two bilateral networks (dual-network model) or by a dominant right hemisphere (right hemisphere dominance model). To adjudicate between these models, we investigated how effective connectivity within the auditory cortical hierarchy is modulated by drowsiness. We recorded high-density Electroencephalography (EEG) in 32 healthy participants performing a lateralised auditory localisation task in both awake and drowsy states. Time-resolved multivariate pattern analysis revealed that arousal states become decodable approximately 150 ms post-stimulus. We used Parametric Empirical Bayes and Dynamic Causal Modelling to map arousal-dependent changes in effective connectivity between left-sided and right-sided auditory stimuli. Consistent with the right hemisphere dominance model of spatial attention, right-sided stimuli elicited stronger bidirectional information flow between bilateral inferior frontal gyri during wakefulness. However, we observed a significant reduction in right-hemispheric frontoparietal connectivity alongside a strengthening of a left-hemispheric pathway from the inferior parietal cortex to the inferior frontal gyrus during drowsiness. This finding supports the dual-network theory and conflicts with the right hemisphere theorys prediction of a bilateral decrease in activity during drowsiness. Overall, our findings provide a mechanistic account of how diminishing arousal shifts cortical processing from a bilaterally integrated network to a left-lateralised state. Our results therefore support the dual-network model of spatial attention while retaining elements of the right hemisphere model.