To our knowledge, this is the first study specifically examining the effects of CPAP on TPN and DMN networks of OSAS patients and comparing results to those observed in HC and in patients with sham CPAP exposure, using fMRI and a working-memory task with 4 levels of increasing load.
The main finding of our study was that active and sham CPAP have opposite effects on behavioral performance and cerebral activation, particularly apparent at the limit-of-capacity task level (3-back).
Most differences between our HC and untreated OSAS patients are found in the deactivation of the temporal regions of the DMN (tDMN) and in TPN activation of bilateral frontal eye fields and of the ventral frontal cortex (VFC) of the right-hemisphere ventral frontal network. On the contrary, TPN activation within the bilateral fronto-parietal network, including dlPFC, was comparatively relatively preserved. Differences are more pronounced in the 3 vs 0-back contrast, where both patient groups performed significantly worse than HC.
After 2 months of treatment, active- and sham-CPAP groups showed opposite trends of cerebral activation: this is particularly apparent at the 3-back task level, suggesting a positive effect of CPAP on the active-treatment group while the sham-CPAP group seems to sustain additional effects of untreated OSAS, or unreported effects of sham-CPAP exposure. An alternative interpretation of our results obtained on the sham group, is that sham-CPAP per se has a negative impact as it may lead to more mouth-breathing and sleep disruption. We could not directly assess the effects of sham-CPAP on the sleep continuity in our patients, but the significant worsening of RTs and decreased TPN activation together with decreased DMN deactivation in the temporal regions at the 3-back after 2 months of sub-therapeutic exposure in the sham group reflects a rather faster evolution than what one would expect of untreated OSAS-related effects.
In a previous report we found a positive correlation between the duration of nocturnal desaturation and cerebral activation in the medial and anterior regions of temporal lobes of untreated OSAS patients. We interpreted these findings as resulting from the particular vulnerability of the temporal structures to hypoxic stress 
. In the present study we improved upon our previous analysis by examining the effects of AHI and nocturnal hypoxemia on cerebral activation specifically within TPN and DMN, which revealed an even more extensive impact of hypoxemia on temporal lobe structures, while AHI effects were comparatively modest (, ). Our findings of opposite response in deactivation of the temporal brain regions after 2 months of active and sham-CPAP treatment support our initial hypothesis: that OSAS has a negative impact particularly on the temporal lobe part of the DMN, possibly via the effects of nocturnal hypoxemia. Our results are corroborated by the findings of the recent multicentric APPLES study showing hypoxemia as the major OSAS-related factor to correlate with behavioral cognitive deficit in OSAS patients 
and by the fact that overall, the most consistent findings in structural neuroimaging studies of OSAS patients are signs of hippocampal injury (atrophy or metabolic anomaly) 
. Similarly, in a recent study, O'Donghue et al have found metabolic abnormalities in the hippocampus of OSAS patients comparable in magnitude to changes found in Alzheimer's disease 
. Those changes were no longer detectable after 6 months of treatment, suggesting that the hippocampus, even if sensitive to OSAS-related injury, has the potential for recovery with CPAP treatment.
Defective deactivation of DMN regions is associated with worse behavioral performance in a number of conditions associated with cognitive impairment and sleep disturbances (normal aging, Parkinson disease, schizophrenia, ADHD), suggesting that inability to reallocate neuronal resources and to suppress activity in neighboring regions plays a role in cognitive impairment 
. Disruption of task-related deactivation within the midline regions of DMN is seen after acute sleep deprivation 
; and slower reaction times on a psychomotor vigilance task are also associated with increased activation in midline DMN regions after total sleep deprivation 
, suggesting that sleep deprivation has a direct impact on DMN.
Sleep deprivation (SD) induces a decrease of cerebral activation in the fronto-parietal and ACC regions of the TPN and reduces DMN deactivation during WM tasks 
. The improvement that we observed after CPAP treatment (in cerebral task-related activation-deactivation and behavioral parameters), associated to partial recovery of task-related deactivation in tDMN regions and in midline core DMN regions, likely reflects two aspects of CPAP action: elimination of episodes of nocturnal hypoxemia and decrease in sleep fragmentation with subsequent improvement of vigilance levels. The “improvement of alertness” effect.is also supported by the higher activation shown by HC compared to OSAS patients in the bilateral anterior insular/inferior frontal regions (BA 47/13)(which constitute the “alerting” node of the TPN 
), as well as in the right VFC 
and by the accentuation of these differences after sham CPAP treatment.
Corbetta and colleagues have proposed that the right hemisphere-lateralized frontal network's function is to detect behaviorally relevant stimuli and works as an alerting mechanism or “circuit-breaker” for the bilateral fronto-parietal network when these stimuli are detected outside of the focus of cognitive processing 
. The frontal component of this network is affected in OSAS patients and may explain further the increased accident risk reported in these patients: in addition to impaired vigilance OSAS patients may suffer from an impaired ability to detect behaviorally relevant stimuli outside of their attention set and in their ability to shift attention in order to make an adequate response.
Two other studies have investigated the effect of CPAP on cerebral activation of OSAS patients. Thomas et al. (2005) have found no recovery of dlPFC activation on a verbal 2-back WM task in 6 OSAS patients after CPAP treatment, although there was some increase of parietal activity 
. Since their patient group performed significantly worse than HC at baseline and did not improve after treatment, the lack of cerebral activation changes in the dlPFC was interpreted as an indication of a more permanent cerebral damage. However, no information was reported on the deactivation pattern.
More recently Castronovo et al. (2009) examined the effect of CPAP on cerebral activation on a verbal 1- and 2-back WM task in 17 OSAS patients 
. Greater activation with WM load was reported in a number of prefrontal regions, bilateral precuneus, left putamen, left hippocampus in never treated OSAS patients as compared to HC and was interpreted as a compensatory over-recruitment. If more extensive activation in the PFC is indeed suggestive of compensatory recruitment of the TPN, more important activation with WM load in the hippocampus can be better explained by defective deactivation with WM load. Together with their finding of lesser activation with WM load in bilateral hippocampi after CPAP treatment, these observations are consistent with our findings of defective DMN functioning in OSAS. Considering that OSAS patients of Castronovo et al performed similarly to HC on all task levels, their patient population can be more directly compared with ours, who did not reveal behavioral differences with HC at 1 and 2-back levels. It appears that in this subset of OSAS patients as opposed to subjects with behavioral deficit studied by Thomas et al, dlPFC function is still intact whereas hippocampal deficits are already present, even though partially reversible with treatment. But, an important methodological point, only the inclusion of limit-of-capacity 3-back level allowed us to observe behavioral differences between patients and HC, as well as the interaction in cerebral activation/deactivation between the two patient' groups (active and sham CPAP).
Our study has several limitations: the fact that we had no untreated patient control group does not allow us to firmly tease apart the possible negative effects of sham-CPAP and the effects of the natural evolution of untreated OSAS. Similarly, the fact that HC have undergone only one fMRI session does not allow complete control of possible learning effects when comparing them to OSAS groups after treatment. In order to minimize this potential bias, we have trained all our subjects twice before the baseline scan at 1-week interval, until maximal performance could be achieved (>75% for 3-back). Moreover, when comparing pre ant post-treatment changes in brain activity, we only report within-subjects contrasts. Finally, the fact that the 2 patient groups exhibited opposite trends of activation and deactivation also suggests that any potential learning effects after 2 months were minimal.
The relatively small sizes of our patient' groups allowed for differences in BMI and behavioral performance at the intermediate (2-back) and maximal WM loads (3-back) between sham and active CPAP groups at baseline despite randomization. It is increasingly recognized that excess body weight is associated with brain structural and functional alterations 
, a higher risk of developing dementia in later life 
, as well as cognitive dysfunction in otherwise healthy adults or after controlling for BP, age, and diabetes 
. Therefore higher BMI in the active CPAP group might account for failure of this group to completely normalize their behavioral performance at the 3-back level after treatment. However, this did not obscure the opposite evolution of the two groups and allowed a further insight into the differences between HC and OSAS patients with different degrees of performance impairment.
Finally, we used relatively short treatment duration of 2 months and it is possible that further improvements in behavioral performance and cerebral activation/deactivation patterns could be seen with longer treatment durations, such as 6 months.