We present the first investigation of the cerebral correlates of SD in a complex, open-ended task. Based on previous SD brain imaging in more standardized laboratory tasks, we hypothesized an adaptive cerebral response following SD. We observed significant activity differences in a meaningful network of brain regions (see below) during SD relative to NS, thereby extending support for the adaptive cerebral response hypothesis to the domain of complex, open-ended tasks.
The SD-related cortical modulations revealed a coherent network of areas involved in a compensatory cerebral response to SD. The most robust cortical finding, observed in six of six subjects, was a large area of SD > NS activity around the left pSTS (the junction of BAs 39, 22 and 37). This region was also identified by the globally sensitive cerebral metabolic rate for glucose (CMRglu) study of Thomas et al. (2000) as well as a recent fMRI study (Drummond et al., 2004). The pSTS has been associated with non-biological, rigid motion inter-relationships as well as cognitive planning or simulation of complex motion (Cohen et al., 1996; Dukelow et al., 2001). We also observed dorsolateral prefrontal cortex (DLPFC) activity (including BA 9), which has been implicated in the working and episodic memory processing and cognitive control (Buckner et al., 1999), consistent with remembering how to operate the joystick controls, and expectations of the visual display's response to those controls. Activity in the right temporal lobe (BA 21) has been associated with general visuospatial processing, and selective spatial attention in particular (Waberski et al., 2002). Right BA 39 around the angular gyurs has also been associated with spatial attention (Chambers et al., 2004). Together, these cortical regions, each exhibiting SD > NS activity, support visuospatial processing, memory and attention – all conceptually important components of the docking task.
One unique finding was in the posterior cingulate (BA 31). Multiple neuroimaging studies of mild cognitive impairment and dementia have shown that reduced blood flow in this region is an early indicator of cognitive decline (Devous, 2002; Wolf et al., 2003). Here, we observed significant modulation of this region following NS but none (or perhaps activity decreases relative to fixation; Fig. 3d) following SD. Thus, we tentatively propose that the loss of task-related BA 31 modulation following SD might directly reflect declining cognitive function. This remains to be investigated in more detail.
Parietal cortex and SD
Over a series of three studies, Drummond and Brown (2001) previously proposed that SD induces a switch from temporal to parietal processing, at least in verbal learning and divided attention tasks, further suggesting that engagement of parietal regions represented a less efficient and less accurate method of performing these tasks. We did not find evidence to support this hypothesis in our docking task. Moreover, the time course data (Fig. 3) suggests that the changes in cerebral utilization following SD are considerably more complex than a simple ‘switch’ from activating one network to activating another.
However, in a more recent paper specifically investigating task difficulty, Drummond et al. (2004) found inferior parietal cortex significantly associated with successful completion of difficult variants of Baddeley's Grammatical Transformation (GT) task. This finding appeared in the absence of behavioral differences between SD and NS, and was located near the pSTS, the site of our most robust finding. The GT task, however, is quite different from SpaceDock: GT is event-oriented, with verbal cues, keypress responses, and requires substantial working memory, whereas SpaceDock is continuous, with implicit visuospatial cues, complex sensorimotor response requirements, and arguably depends less on working memory. The tasks share two characteristics, however: both have a significant spatial component, and both are challenging. From just these two studies, one cannot attribute activity in the pSTS specifically to spatial- or difficulty-related causes. However, the replicated finding in two highly distinct tasks highlights Drummond et al.'s suggestion that more focus should be placed on the parietal lobes in overcoming the effects of sleep loss.
Other sources of SD versus NS differences
A number of alternative, (non-exclusive) hypotheses can be put forth to explain the observed differences in cerebral activity between SD and NS conditions. First, task performance levels may have differed in the two conditions. In fact, only two prior SD neuroimaging studies reported no observable behavioral deficit (Drummond et al., 2004; Portas et al., 1998), and Portas’ study found only thalamic responses to SD. Although our study is not immune from such a performance-based interpretation (some significant behavioral effects were observed, and we had relatively low power for detecting between-group differences in behavioral performance), we nevertheless found little evidence of consistent behavioral impairment following SD, as previously observed in complex tasks (Harrison and Horne, 2000).
Alternatively, subjects may have engaged in an effortful attempt to maintain performance under SD. Support here comes from our debriefing sessions, in which every subject described task performance as requiring more effort following SD. The simplest effort-based model would propose global upregulation of brain activity as effort increased, and would be consistent with SD > NS activity. The time course findings in Fig. 3, however, suggest that this simplest effort-based explanation is inadequate. In particular, the different time courses suggest that SD induced at least four qualitatively different regional responses: upregulation (3a), downregulation (3b), recruitment (3c) and dropout (3d). The NS > SD finding in BA 31 also challenges the simplest effort hypothesis. Alternative, regionally specific formulations of an effort hypothesis, however, remain as potentially viable explanatory candidates.
Third, it has been suggested that performance strategies may change with SD (Drummond and Brown, 2001). In fact, complex tasks can provide more strategic opportunities for enhancing (or maintaining) performance (Harrison and Horne, 2000), and SpaceDock provides several such options. We suggest, however, that our findings are not solely strategy related based on three observations: (1) we focused our analyses specifically on brain regions exhibiting consistent activity differences across subjects, and one would expect subjects naïve to the task to select different maintenance strategies, (2) we found no consistent behavioral changes across subjects suggesting consistent strategic alterations; and (3) at debriefing, our subjects revealed no awareness of changing their task performance strategy. Thus, while it is always difficult to eliminate the strategy change explanation, we believe such changes played a limited role in this study.
Finally, it is possible that certain regions of the brain are inherently more sensitive to SD. The fact that this and prior studies show substantial task-specificity of cerebral activity following SD argues partly against an inherent sensitivity hypothesis. However, relative measures such as fMRI (as opposed to, e.g. CMRglu) do not probe the brain uniformly – they are biased towards regions most involved in the task at hand. Thus, inherent, regional cerebral sensitivity to SD remains a viable explanation of available data.
In summary, while our data are inconsistent with several hypotheses alternative to cerebral compensation, some such explanations cannot be rejected. Looking at the qualitative range of observed time courses actually suggests a multiplicity of explanations: some regions display changes in the magnitude of otherwise significant activation (possibly reflecting cognitive reserve (Stern, 2002); others appear to exhibit compensatory recruitment/dropout, as discussed by Drummond et al. (2004). These along with inherent sensitivity, effort, and strategic hypotheses remain to be rigorously teased apart with additional SD studies on both simple and complex, open-ended tasks.
There are three principal limitations of our study. First, while behavioral performance was found to be reasonably equated across sleep conditions, the between-subject behavioral analyses suffered from low power for detection of significant behavioral effects. Moreover, some behavioral differences were observed in both within- and between-subject analyses. In particular, there was some evidence of an unresolved learning curve (our Order by Condition interaction). Given our counterbalanced orders, an unresolved learning curve would tend to reduce the observed performance differences. Regardless, behavioral differences across conditions cannot be excluded as potentially contributing to the neuroimaging results. Also, we did not address the role of expertise or automaticity in task performance, which could affect neurophysiological responses in the face of SD (Peres et al., 2000). Finally, and most limiting is the small subject pool, which precluded standard random effects analysis. A major change in scanner hardware (including field uniformity and receiver coils) as well as limited funds precluded addressing this low-N directly. Instead, we restricted our interpretation to only those regions exhibiting reliable activation changes across subjects. While appropriate, this provides only an initial measure of inferential support, warranting further investigation of complex tasks in SD to assess the robustness and the generality of these findings.