We recruited 29 healthy adult participants, most of whom were students at the University of Lübeck. The inclusion criteria encompassed right-handedness as tested by the Edinburgh Handedness Inventory (Oldfield, 1971), a normal or corrected-to-normal visual acuity and intact color vision as tested by the Ishihara’s Test (Kanehara & Co., Ltd., Tokyo, Japan). Exclusion criteria were a known neurological, psychiatric, or ophthalmological disease.

Before participation, written informed consent according to the Declaration of Helsinki and its later amendments was obtained from each participant. The study was approved by the local Ethics Committee of the University of Lübeck (14-189).

Two of the participants had to be excluded due to excessive head motion in the MRI scanner (see section “Materials and Methods” on quality control of rs-fMRI data), leaving 27 subjects [female: n = 16 (59%); mean age: 24.2 ± 0.8 years, range: 19–38 years] for final analysis.

Participants viewed the stimuli on a screen (NNL LCD monitor, NordicNeuroLab, Bergen, Norway; active TFT, 69.8 × 39.3 cm visual area), positioned on the front end of the MRI scanner, via a mirror that was attached to the head coil (eye-to-screen distance: 130 cm, i.e., visual area 30.1° × 17.2°). A keypad with buttons for the manual responses was fixed at the participant’s right thigh using hook-and-loop tape.

Three different tasks were presented, each addressing a specific subcomponent of spatial attention. The (I) Posner task tests covert reflexive shifts of attention (orienting and re-orienting) (Posner et al., 1984), the (II) Landmark task visuospatial judgments (Milner et al., 1993) and the (III) VS task (Machner et al., 2018) overt shifts of attention by exploratory eye movements. All these tasks are known to activate the bilateral FEF and IPS as core regions of the DAN (Corbetta et al., 1998; Fink et al., 2000; Doricchi et al., 2010).

For our experiment, the original Posner paradigm was adapted to the scanner environment in analogy to a previous study (Thiel et al., 2004). The basic display showed a dark-gray background with a light-gray diamond at the center and two light-gray square frames (size 2°), positioned on the horizontal meridian at either side with an eccentricity of 12° from the center. As an attentional cue, one side of the central diamond briefly (200 ms) lighted up. After a delay of either 150 or 400 ms (stimulus onset asynchrony), the target (a light gray asterisk) appeared for 100 ms within one of the two peripheral frames. The inter-trial interval varied between 1650 and 1900 ms. The participant was instructed to press the left button on the keypad with the index finger as quick as possible upon target appearance. One run of the Posner task included 48 valid (cue and target on the same side) and 12 invalid (target on the opposite side as indicated by the preceding cue) trials, corresponding to a cue validity of 80%, which allows assessment of endogenous attention shifting (Chica et al., 2011).

The Landmark task represents a “perceptive version” of the established line bisection task (Schenkenberg et al., 1980) but is better suited for the scanner environment (Fink et al., 2000). A white line (20° or 24° of visual angle) appeared on a dark-gray background. It was pre-bisected with a vertical dash either directly in the middle, i.e., with both parts of the line having the same length, or with a leftward or rightward shift of either 0.25° or 0.5°. The vertical dash bisecting the horizontal line was always aligned with the screen’s center so that the absolute spatial position of the bisection mark on the screen could not be used as a cue for the final spatial judgment of the line length (Ellison et al., 2004). The participants were asked to judge which half of the line was longer or whether both were equal in length. If the left part of the line was estimated to be longer, they should press the left button on a keypad with their right index finger. In case of a longer segment on the right, they should press a button with their fourth finger and in case of equal lengths of both halves of the line they used their middle finger. The line was presented for 300 ms and participants had a maximum time of 2000 ms to give their response by pressing the respective button. One run of the Landmark task consisted of 80 trials, 32 of which showed a correctly bisected line with both halves having the same length (“equal”), 24 trials with a longer left segment and 24 trials with a longer right segment.

In the VS task a computerized but naturalistic image of a desk scene was presented, in which participants were asked to find a paperclip (target) among different other everyday objects (e.g., a pen, coin, key, etc.) that served as distractors (Machner et al., 2018, 2020a). Each VS trial started with a central fixation cross presented on a black background, followed by the appearance of a desk image containing 30 different objects (see Figure 1 for an example). Participants were instructed to press a response button on a keypad as soon as they found the target. If there was no paperclip to be found, participants were asked to press a different button. A trial ended upon the button press or after a maximum time of 5000 ms. One run of the VS task included 30 trials of different desk images, 80% of which contained a paperclip (“target trials”).

The experimental design is depicted in Figure 1. Before entering the MRI scanner, subjects were briefly familiarized with the three tasks outside the MRI scanner. In the MRI scanner, subjects first received a structural MRI scan of the brain. Then, the PRE resting-state fMRI run was conducted. Next, there was the task block (total duration 24 min), in which participants performed three runs of each task in a predefined order (see Figure 1). Afterwards, the session was completed with a second rs-fMRI (POST) run.

To control for adherence to the tasks as well as for wakefulness during the rs-fMRI runs (eyes open), we continuously recorded and monitored eye movements of the participants using an MRI-compatible, remote eye tracker with a sampling rate at 1000 Hz (Eyelink 1000 Plus, SR Research, Ottawa, ON, Canada). This monitoring could exclude that participants fall asleep during the rs-fMRI sessions, also proven by offline analyses of eye position signals that showed only a small number of lacking eye signals in the PRE (14 ± 2%) and POST (18 ± 3%) rs-fMRI run, which were largely due to transient lid closure and artifacts.

From the Posner task, we analyzed the mean reaction time (“RT mean”) of the responses in all the valid and invalid trials. Furthermore, we calculated the difference in RT between the invalid and valid trials (“RT invalid-valid”) as a more specific indicator for attentional reorienting (Rengachary et al., 2011). From the VS task, we analyzed the mean RT for trials, in which a target was present and the response button was correctly pressed. The performance in the Landmark task was assessed by calculating the “error rate (ER).” Therefore, the number of trials incorrectly judged was divided by the total number of trials presented ×100.

For each task, we additionally assessed the individual improvement by calculating the difference in the subject’s RT (ER, respectively) between the first and the last (third) run.

Statistical analyses, apart from the fMRI analyses described above, were performed using the SPSS software package (version 22.0.0.2; IBM Corp., Somer, NY, United States).

Unless otherwise reported, data in the manuscript are presented as mean ± standard error of the mean (SEM). Differences in the ROI-to-ROI FC values between the PRE and the POST rs-fMRI run were assessed using paired t-tests. Correlation analyses between rs-fMRI (ROI-to-ROI FC) and behavioral (RT and ER, respectively) parameters were performed using the non-parametric Spearman’s rho correlation coefficient. The results of the rs-fMRI-behavior analyses, which tested whether behavioral performance was correlated to the DAN-FC in a hypothesis-driven set of four predefined ROI pairs, were corrected for multiple comparisons by applying the Bonferroni–Holm correction (p-value × 4). Furthermore, significant FC-behavior correlations in one rs-fMRI session (e.g., PRE) were tested for a significant difference to the corresponding rs-fMRI session (e.g., POST). Therefore we assessed the interaction FC × behavior by weighting the individual FC value with the linear regression coefficient from the FC-behavior correlation and subsequently performed a paired t-test on the weighted FC between PRE and POST. Regression coefficients were calculated using “robustfit” function within Matlab (DuMouchel and O’Brian, 1989). This function uses iteratively reweighted least squares to compute the coefficients, which makes it robust against extreme values and outliers. The level of significance was set at p < 0.05.

The participants’ mean RT in the Posner task was 435 ± 11 ms and there was significant improvement from the first to the last run (d = −32 ± 6, p < 0.001). The validity effect, i.e., the difference in RT between invalid and valid trials, was on average 32 ms ± 6. The mean RT in the VS task was 1500 ± 54 ms, again with significant improvement over time (d = −270 ± 49, p < 0.001). The mean ER in the Landmark task was 39 ± 1%. In this task, the participants revealed no improvement between the first and the last run (d = −1 ± 1%, p = 0.522).

Figure 2 depicts the statistical maps showing the results of the seed-based (ROI-to-voxel) FC analysis of the four predefined DAN-ROIs, separately for the PRE and POST rs-fMRI run.

For each of the four seeds, we consistently observed functional connections to the (bilateral) FEF, IPS, middle frontal gyrus (MFG), and MT+ (middle temporal complex) as well as to the basal ganglia and the thalamus.

Statistical comparison of the different seed-based FC maps between the PRE and the POST run did not reveal supra-threshold voxels. Hence, FC of the predefined ROIs to whole brain did not change significantly from PRE to POST.

Table 1 depicts the group-mean of Fisher’s z-transformed ROI-to-ROI FC results of the different intra- and inter-hemispheric ROI pairs, separately for the PRE and POST rs-fMRI run. Pairwise comparisons did not reveal significant differences between the PRE and POST rs-fMRI run (p always > 0.104).

Correlation analyses were performed between behavioral results (mean RT and RT improvement in the Posner and VS task, ER in the Landmark task) and ROI pairs’ FC results in the PRE and the POST run (Table 2).

The following significant rs-fMRI-behavior correlations were revealed for the PRE run (Figure 3): The intra-hemispheric FC between left FEF and IPS correlated with the RT improvement in the VS task (r = 0.521, p = 0.024) as well as with the RT improvement in the Posner task (r = 0.496, p = 0.052, statistical trend). Thus, the stronger the participant’s FC between left FEF and IPS, the better was the individual learning effect in both tasks. Furthermore, the FC between right and left FEF was correlated with the mean RT in VS (r = 0.527, p = 0.020), i.e., participants with stronger interhemispheric FEF-FC needed on average more time to detect the target in the VS task.

In the POST run (Figure 3), the FC between right FEF and IPS was found to be inversely correlated with the mean RT in VS (r = −0.502, p = 0.032). Thus, participants with faster VS revealed stronger intrahemispheric FC in the right DAN after task performance. Moreover, the POST run FC between right FEF and IPS was negatively correlated with the RT improvement in the Posner task (−0.676, p < 0.001) and also the average DAN-FC was negatively correlated with the RT improvement in the Posner task (−0.581, p = 0.001). Hence, participants with greater improvement in the Posner task subsequently revealed more decreased intrahemispheric FC in the DAN.

When comparing these correlation results between PRE and POST, the interaction FC*behavior became significantly different in four (p < 0.001) and showed a statistical trend (p < 0.01) in one of the comparisons (Figure 3).

Finally, the individual difference in FC between the POST and PRE run was correlated to the different behavioral parameters (Table 2, POST-PRE lines). This yielded significant inverse correlations with the RT improvement in the Posner task for the two ROI pairs FEF_L–FEF_R and FEF_L–IPS_L as well as for the DAN-FC (Table 2, corrected p always < 0.05).

The strength of FC in several DAN-ROI pairs was found to be related to the behavioral performance in different tasks of spatial attention. For example, an increased FC between left-hemisphere DAN regions (left FEF and IPS) was associated with better learning (improvement of RTs over time) in the VS task. Furthermore, stronger inter-hemispheric DAN-FC (bilateral FEF) predicted better learning (RT improvement) in the Posner task but counterintuitively it was also correlated with slower (not faster) RTs in the VS task. There was no significant correlation between DAN-FC and behavioral performance in the Landmark task. However, this task appeared to be too demanding with high ERs (about 40% on average) and without significant improvement (learning) over the runs. Thus, this behavioral parameter might not have been an optimal candidate for FC-behavior correlations.

As mentioned above, the correlations between PRE-task DAN-FC and behavioral parameters in VS and Posner were not unidirectional (same accounted for the POST task FC results), i.e., stronger DAN-FC was often – but not always – associated with better performance (faster mean RT or greater improvement over the runs). One reason may be that the FC between brain regions indicates a functional connection but the direction/type of influence cannot directly be inferred (Friston, 2011), being either beneficial (excitatory) or disturbing (inhibitory). Hence, an increase of FC between two connected brain areas, in which one region functionally inhibits the other, may result in worse behavioral output, whereas the FC increase between functionally synergistic regions can lead to better performance.

Despite some dissociations, our findings principally support the previous proposal of RSNs representing individual traits that may determine the personal cognitive ability of humans (Harmelech and Malach, 2013; Sadaghiani and Kleinschmidt, 2013). Our results underline that such a relation is also found for the DAN, which is recruited during behavioral engagement in the related functional domain (visuospatial attention). Different previous studies on the relation of pre-task FC and attentional performance stressed the importance of between-networks FC, specifically the functional interaction of prefrontal DAN areas and visual cortices (Baldassarre et al., 2012), of right parietal DAN regions and dorsal anterior cingulate cortex belonging to the default mode network (Bueicheku et al., 2020) or between parietal DAN regions and remote subcortical/medial-temporal/orbito-frontal regions belonging to several different RSNs (Chou et al., 2013). In our study, the within-network FC of the DAN was shown to be predictive of the individual behavioral performance in spatial attention tasks. This confirms previous findings where performance in a visual signal detection task was related to pre-task FC between parietal and dorsolateral prefrontal DAN regions (Berry et al., 2017) and that attentional performance of children between 4 and 7 years was correlated to the individual FC in fronto-parietal DAN regions (Rohr et al., 2017). Studies assessing FC by use of electroencephalography (EEG-FC) also revealed the behavioral relevance of pre-task EEG-FC in DAN regions for attentional tasks involving audio-visual stimuli (Hipp et al., 2011) or visuo-motor performance (learning of mirror drawing skills) (Manuel et al., 2018). That the ongoing “spontaneous” activity in the RSNs may even be a prerequisite for the learning of skills is supported by a recent study using magnetoencephalography during engagement in a motor task. It showed that performance improvement relied on micro-offline gains acquired during the phases of rest interspersed between bouts of motor practice (Bönstrup et al., 2019).

In summary, our findings together with those of previous studies underline the predictive value of both within- and between-network FC of the DAN for the individual ability in visuo-spatial attention.

As revealed by ROI-to-whole brain FC analyses, the predefined DAN core regions (bilateral IPS and FEF) were functionally connected with each other as well as with the bilateral MFG and MT+, regions that are usually considered to build the DAN (Fox et al., 2006; He et al., 2007; Hacker et al., 2013). These brain regions are also recruited during active engagement in tasks that require shifts of spatial attention (Corbetta and Shulman, 2002). The ROI-to-whole brain FC as well as the ROI-to-ROI FC did not change from the PRE- to the POST-task recording. These findings are in line with previous studies reporting the DAN to be an inter-individually reproducible (Fox et al., 2006) and intra-individually highly stable RSN (Choe et al., 2015; Finn et al., 2015). The within-network FC of the DAN remains even stable when its state changes from rest to action, as previously shown in a study that compared resting-state FC to intra-task FC while participants performed a visuospatial attention task (Spadone et al., 2015).

That there was no significant change of absolute FC values from PRE to POST in our study does not mean that the DAN is not malleable or influenceable by experience. Of note, there was a certain change in the DAN-FC from PRE to POST in that several significant FC-behavior relations were only present in the PRE but not the POST rs-fMRI session and vice versa.

Hence, there must have been some kind of reorganization of the DAN’s FC between PRE and POST, either (I) in direct response to the task performance or (II) spontaneously in terms of state fluctuations. As an example, the FC-behavior relation of right FEF-IPS FC and the mean RT in the VS task was significant only in the POST session after task engagement, because only then (but not before) a faster search was associated with stronger intra-hemispheric connectivity in the right DAN. Following hypothesis I, this may indicate an early and specific adaptation of the DAN to the task demands. That a behavioral intervention (i.e., one session of task performance or a several-day cognitive training) can in principle change RSNs was previously shown for different cognitive domains. For instance, one session of a new language task was able to change the FC in the language network immediately after the task (Waites et al., 2005). A 6-weeks mnemonic training in naïve subjects induced persistent FC changes in and between the medial temporal network, DMN, and other RSNs, finally resembling those of memory athletes (Dresler et al., 2017). Regarding the DAN and visuo-spatial attention, an intense training over several days on a shape-identification task led to changes in FC between visual cortex and frontal-parietal DAN areas, which were correlated with learning effects (Lewis et al., 2009). Engaging in a visual classification task (faces or scenes) caused differential coupling between ventral frontal cortex and category-preferential visual cortex regions in subsequent periods of rest (Stevens et al., 2010). Furthermore, one session of a VS task was shown to increase the FC between the right posterior parietal cortex and the dorsolateral prefrontal cortex (Bueicheku et al., 2019). A visuo-perceptional task training (motion coherence discrimination) was shown to increase the FC of MT+ as the cortical region responsible for the processing of moving visual stimuli (Sarabi et al., 2018) but also the FC of the hippocampus as a region involved in the consolidation of early learning (Urner et al., 2013).

In the light of these studies, our findings of different FC-behavior relations in the DAN before and after performance of spatial attention tasks could be interpreted in support of task-induced DAN-FC changes. This would underline the hypothesis that the DAN – although intra-individually very stable – can be shaped by learning experiences and that RSNs may serve as a flexible, continuously updated “memory system” that helps the individual to be better prepared for upcoming cognitive challenges (Sadaghiani and Kleinschmidt, 2013; Manuel et al., 2018). It is also in line with the hypothesis that RSNs may – at least partly – be the result of learning, i.e., repetitions of task-related co-activations of different brain regions (Miall and Robertson, 2006; Harmelech and Malach, 2013; Guerra-Carrillo et al., 2014), and that this rather flexible system plays on the rigid backbone of structural connections in the brain (Sadaghiani and Kleinschmidt, 2013).

However, following hypothesis II, there is also an alternative explanation for the observed differences between the two rs-fMRI sessions (PRE/POST) regarding different FC-behavior correlations in our study: spontaneous state fluctuations over time. Previous studies accordingly emphasized the “intrinsic activity,” “ongoing dynamics,” or “time-varying FC” of RSNs fluctuating over seconds and minutes independent from specific inputs or outputs (Raichle, 2015; Sadaghiani et al., 2015; Kucyi et al., 2018). Ongoing intrinsic fluctuations over large-scale networks were also shown to determine different attentional states, such as in sustained attention and task-unrelated mind wandering (Yamashita et al., 2021; Zuberer et al., 2021). The extent of these fluctuations can still influence the upcoming behavioral performance in a cognitive task explaining inter-trial (and indirectly also inter-subject) variability (Coste et al., 2011). Interestingly, when correlating the change in FC from PRE to POST with behavioral performance parameters in our cohort, smaller DAN-FC changes were associated with larger improvement in the Posner task. Hence, one could speculate that participants with a very stable “less fluctuating” DAN had better learning in this specific task. This assumption is also supported by a previous study, which showed that task-related reduction of FC variability was associated with improved behavioral performance in a letter recognition task (Elton and Gao, 2015).

Due to the design of our study, we cannot exclude that small spontaneous fluctuations of DAN-FC over time finally led to different FC-behavior correlations in the POST than in the PRE rs-fMRI session. Task-independent spontaneous state changes could even resolve some of the discrepancies in our FC-behavior correlations, for instance, that stronger post-task FC between right FEF and IPS was related to shorter RTs in VS and at the same time to smaller RT improvement in the Posner task.

To thoroughly disentangle task-induced changes from time-varying spontaneous fluctuations of the RSN one would have required a control rs-fMRI experiment without an interleaved task performance, which was not part of this study. The modest sample size of this study might have prevented to detect smaller effects of PRE/POST FC changes due to a lack of statistical power. Furthermore, significant FC-behavior correlations in a rather small sample may be statistically inflated and may not be replicated in larger (>2000) brain-wide association studies (Marek et al., 2020). This also casts some doubt on the general assumption that rs-fMRI FC strongly indexes inter-individual differences in cognitive ability, as long as it is not replicated in larger brain-wide association studies.