There are now many fMRI studies which investigate functional or effective connectivity between brain regions during task performance or in the resting state. Most of these analyses have tried to investigate the linear or non-linear functional interactions between predefined brain regions and used different measures of statistical dependency like simple linear or partial correlations, mutual information, Granger causality, or coherence (Smith et al., 2011).

Analyses of functional or effective connectivity have been only used in few pharmacological fMRI studies. We here shortly summarize the existing studies in the cholinergic and noradrenergic system. Wink et al. (2006) investigated the muscarinic receptor antagonist scopolamine on functional connectivity in a resting state condition and focused their analysis on connectivity of the hippocampus with six, large brain regions including temporal, parietal and frontal cortex. They found that scopolamine enhanced the connectivity between the frontal cortex and hippocampus, brain regions which were also affected by ageing. Effects of the cholinergic agonist nicotine have also been investigated in the resting state. Tanabe et al. (2011) focused their analysis on brain regions belonging to the so-called default mode network and on extrastriate regions. They provide evidence that nicotine reduced neural activity in the default mode network, and increased activity in the extrastriate resting state network in non smokers. The coupling between the default mode network and an executive resting state network was analyzed in smokers under placebo and nicotine by Cole et al. (2010). Their findings show both, nicotine-induced increases and decreases in coupling of those two networks with different brain regions. Others have focused their connectivity analyses on brain regions such as the cingulate cortex and have shown enhanced functional resting state connectivity of the cingulate cortex with several fronto-parietal brain areas under acute nicotine (Hong et al., 2009). A different approach was taken by Balsters et al. (2011) who combined EEG/fMRI with donepezil administration to investigate cholinergic modulation of oscillatory brain activity which was then related to changes in the BOLD signal. Behaviorally donepezil impaired performance. Neurally drug effects were evident as reductions in alpha and increases in beta and delta power. These maladaptive oscillatory changes were associated with BOLD signal changes within the hippocampus, fronto-parietal brain regions and the so-called default mode brain network.

While connectivity studies in the cholinergic system were exclusively performed in the resting state, most of the studies on the noradrenergic system analyzed changes in connectivity during task performance. Wang et al. (2011) used an effective connectivity analysis to assess the effects of the selective noradrenaline reuptake inhibitor reboxetine in stroke patients within a finger tapping task. They found that reboxetine increased the connectivity between supplementary motor areas and the primary motor cortex. Increases in effective connectivity after reboxetine were also reported in healthy volunteers in a visuomotor control task. Here regions showing increased connectivity were the right visual, intraparietal and superior frontal cortex (Grefkes et al., 2010). McCabe and Mishor (2011) investigated the effects of reboxetine within a resting state condition and focused their analysis on regions of the limbic system. Their data provides evidence that reboxetine reduced the striatal–orbitofrontal cortex connectivity.

Hermans et al. (2011) analyzed the effects of stress and of propranolol, a beta-adrenergic receptor blocker, and found that propranolol reduced functional connectivity in a salience network but not visual network during subjects saw aversive cinematographic material and feared mild electrical shock. Coull et al. (1999) analyzed the effects of clonidine (a2 adrenoceptor agonist) on effective connectivity during a rest condition and during the performance of a visual attention task. During the rest condition they found decreased connectivity between the frontal cortex to thalamus and in pathways to and from visual cortex. In contrast, during the attentional task, clonidine increased the functional connectivity from and to the parietal cortex. These findings resemble the opposing effects of reboxetine found by Wang et al. (2011) and Grefkes et al. (2010) during task performance as opposed to those found by McCabe and Mishor (2011) in the resting state and indicate again the necessity to differentiate between effects of drugs during task performance vs. rest.

Due to the diversity of the administrated drugs, task conditions and applied analysis techniques which often focused on certain brain regions only, it is difficult to integrate the reported results. Furthermore, drug-induced increases or decreases of functional connectivity in specific brain regions can induce quite different changes in functional brain topology and it is unclear whether these changes improved or hindered the brains' capacity for information processing.

Recently, an increasing number of studies extended connectivity analyses by using graph-theoretical measures (see Box 1) to interpret the interactions of brain regions (Achard and Bullmore, 2007; Xia and He, 2011). In contrast to the above approaches complex network- or graph analyses often investigate the whole host of functional connections of the entire brain to describe the topology or organization of functionally connected brain networks. Thereby, the brain is analyzed as a complex network of highly interconnected processing nodes. The nodes often represent anatomically-defined brain regions and edges embody the functional interaction between them. In contrast to normal functional connectivity analyses, graph theory provides a full range of measures to describe drug-induced changes in topology. Thus, with graph theoretical measures it is possible to detect not only focal, but also subtle, widely distributed drug effects on functional network topology. Most important, graph theoretical measures also quantify whether drug-induced changes in topology or network organization facilitate or hinder information processing (Bassett and Gazzaniga, 2011) and thus provide a biological explanation for drug effects on behavioral performance. For example, while Furey et al. (2000) hypothesized in their seminal study that cholinergic neurotransmission increases the efficiency of information processing based on focal activity within extrastriate and prefrontal cortex, a graph analytical approach would enable to quantify processing efficiency with mathematical measures. Currently few studies have used graph and complex network analysis to investigate drug effects on functional brain networks.

There is strong empirical evidence that variation in network integration and network efficiency correlates with cognitive performance. This correlation has been shown, for example, for higher IQ (Hampson et al., 2006; Li et al., 2009; van den Heuvel et al., 2009) and working memory performance (Bassett et al., 2009) and shorter reaction times in a Go/NoGo task (Zhou et al., 2012a). Further support for a relationship between brain network topology and behavioral performance derives from studies which investigated the network topology in elderly volunteers. These studies found that brain networks of elderly participants have less efficient and more modular brain topology with fewer long-distance and more short-distance connections which might explain age-related performance declines in multiple cognitive functions (Achard and Bullmore, 2007; Meunier et al., 2009; Wang et al., 2010). Recently, we could also show that experimentally induced changes in task performance correlates with the efficiency of brain networks. While the cholinergic agonist nicotine improved behavioral performance in a sustained attention task and network efficiency in the resting state, time-on-task and cognitive fatigue increased reaction times and induced higher modularity within the network (Giessing et al., unpublished data; Breckel et al., unpublished data). Most of the studies reported above investigated endogenous resting state data supporting the assumption that endogenous functional brain network topology interacts with task performance and it can be assumed that endogenous functional brain network topology also interacts with task-related neural processing.

The reported empirical correlations between functional brain network topology and behavioral performance fit with contemporary network theories of information processing in the brain, like the “brainweb” (Varela et al., 2001) or “workspace” theories (Dehaene et al., 1998; Baars, 2002). Baars et al. proposed the workspace theory of conscious perception of cognition (Baars, 1988, 2002, 2005; Shanahan and Baars, 2005) which has since been developed as global neuronal workspace theory by Dehaene and others (Dehaene and Naccache, 2001; Bartolomei and Naccache, 2011; Dehaene and Changeux, 2011). This theory predicts that conscious performance of demanding tasks requires the integration of otherwise segregated brain modules. The integration of segregated brain modules results from an ensemble of “workspace neurons” which is anatomically distributed throughout the brain (Dehaene and Naccache, 2001; Baars, 2005). It has been suggested that synchrony in the gamma band is important for the global neuronal workspace and that this gamma synchrony is regulated by GABAergic interneurons which are modulated by acetylcholine and dopamine (Changeux and Lou, 2011). Other theories like the “brainweb theory” also emphasize that behavioral performance is largely dependent on the integration of functionally specialized brain regions and that the synchronization or correlation of brain regions is a key mechanism for large-scale integration of brain networks (Varela et al., 2001). Recent evidence in humans by Bauer et al. (2012) indicated that the cholinesterase inhibitor physostigmine is able to modulate synchronization in the alpha and beta band but no evidence was found for a modulation in the gamma band. In their study, Bauer et al. (2012) have shown that the drug speeded performance and enhanced spatial attention effects on synchronization in visual cortex. In summary, the reported theories predict that higher cognitive performance goes along with higher functional integration of brain networks and at least for the cholinergic system there is first evidence that cholinergic drugs may promote the functional integration of brain networks by modulating synchronous brain activity.

Only a few studies have used complex network analyses to investigate drug effects. We will thus mention also studies outside the cholinergic and noradrenergic system to underline the feasibility of graph analytical approaches in understanding drug effects. Complex network analyses have been previously used to partition functional brain networks into communities of densely interconnected nodes after drug challenge (Schwarz et al., 2008, 2009; Bifone et al., 2010). The authors compared dopaminergic, serotonergic and cholinergic drugs. After nicotine challenge they found two networks, one including sensorimotor cortex, thalamus, hypothalamus, hippocampus, and inferior colliculus, the other including cingulate, pre- and orbitofrontal cortex, striatum, amygdale, piriform, and entorhinal cortex as well as visual and parietal regions. For the human brain, previous results revealed that the efficiency and capacity for parallel information transfer of brain networks are affected by normal aging and by pharmacological blockade with the dopamine receptor antagonist sulpiride (Achard and Bullmore, 2007). These results suggest that the dopaminergic antagonist declines the efficiency of small world networks similar to the effects of aging. For the cholinergic system we could recently show that the cholinergic agonist nicotine increased the efficiency, reduced the modularity and clustering of brain network and drove brain functional networks to a more integrated workspace configuration (Giessing et al., unpublished data). We would assume that improvements in attention and/or cognitive control with noradrenergic drugs should similarly increase network efficiency.

Even though different drugs or other experimental conditions might have similar effects on network topology on the global, whole brain level, the underlying mechanisms might be quite different. For example, Achard and Bullmore (2007) showed that ageing and the dopamine antagonist sulpiride impaired network efficiency, but both effects were differentially localized within the brain. While the effects of ageing were found in several brain regions including the frontal and temporal neocortex, the dopamine blockade reduced efficiency mainly in the dorsal cingulate and lateral temporal cortical areas. Thus it may be reasonable to assume that even though cholinergic and noradrenergic drugs will both increase global network integration since both have pro-cognitive effects, the localization of effects may be different. In Box 2 we suggest an experimental approach to study the interaction between task-related and resting state functional connectivity in pharmacological fMRI studies.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.