Brain function relies on coordinated activity in local neural circuits, from which large-scale functional brain networks are composed (Park and Friston, 2013). Such changes in the activity of local neural circuits and large-scale systems are paralleled by the modulation of metabolic processes (Riedl et al., 2016). Recent work has assessed the dynamic changes in functional brain network architecture associated with cognitive task engagement. Results from these investigations have challenged the prior notion (Smith et al., 2009) that functional interactions at resting-state are relatively stable and sufficient to support goal-directed behavior (Cocchi et al., 2013). For example, studies have highlighted dynamic brain network reconfigurations when switching from a state of rest to that of cognitive engagement (Braun et al., 2015). Solving tasks of increasing cognitive complexity has further been linked to complex changes in the functional interplay between cortical regions that otherwise comprise distinct networks at rest (Hearne et al., 2017). These findings are in line with the proposal that brain networks exhibit a flexible modular architecture, with highly interconnected hub regions changing their functional network assignment according to task demands (Cole et al., 2013).

Task-induced neural activation in segregated brain regions and corresponding network dynamics have, of late, been largely assessed using the blood oxygen level dependent (BOLD) signal. The BOLD signal represents a non-specific proxy of activation, which is directly mediated by hemodynamic factors including changes in blood flow and oxygen content rather than by neural metabolism (Heeger and Ress, 2002). As a result, the metabolic underpinnings accompanying large-scale functional network reconfigurations that support cognition have received little recent research focus (Horwitz et al., 1995; McIntosh et al., 1994). It also remains unknown if the association between functional connectivity and metabolism is confined to specific task-related brain networks or extends more globally. Furthermore, the variation of this association as a function of task load is poorly understood. To address this gap in knowledge, we analyzed simultaneously acquired positron emission tomography (PET) and functional magnetic resonance imaging (fMRI, i.e., BOLD and arterial spin labeling (ASL)) data while healthy participants performed a well-validated cognitive task comprising two levels of difficulty (Haier et al., 2009). By combining these different imaging modalities we were able to (i) comprehensively map whole-brain network reconfigurations as a function of varying cognitive demands; (ii) assess the coupling between task-driven metabolic increases and modulations in functional connectivity; and (iii) model the neural dynamics underpinning changes in functional connectivity and related metabolic demands. We hypothesized significant task-induced changes in the link between metabolic and neural factors, supporting our ability to flexibly engage in cognitive processes. Previous work has shown that the bulk of functional network reconfigurations occur when participants engage in a cognitive task, with minor reorganizations as a function of cognitive load (Hearne et al., 2017). We therefore expect greater metabolic demands to be mobilized to establish task-specific patterns of functional brain interactions.

Simultaneous functional MRI (BOLD and ASL) and [¹⁸F]FDG PET data were acquired while 22 healthy adult participants performed a cognitive task (Tetris). The task had two different levels of difficulty and involved rapid visuo-spatial processing and motor coordination (see Materials and methods and Figure 1A and B).

Figure 1

Download asset Open asset

We first integrated information regarding task-related changes in local BOLD signals, cerebral blood flow (CBF) and glucose consumption to identify functionally segregated brain regions involved in task performance (Figure 1C). This analysis leveraged the recently introduced approach of functional PET (fPET) imaging to determine task-specific changes in the cerebral metabolic rate of glucose (CMRGlu) (Hahn et al., 2016; Rischka et al., 2018) as well as canonical analyses for BOLD and ASL signal changes. We then assessed task-based network reconfigurations between the ensuing identified brain regions using metabolic connectivity mapping (MCM) (Riedl et al., 2016; Figure 1C). MCM evaluates the association between regional patterns of BOLD-derived functional connectivity and glucose metabolism (see Materials and methods for rationale underlying this approach). To this end, task-based functional connectivity was computed from data collected during periods of continuous task performance (6 min, Figure 1A). This approach minimizes the influence of nonspecific factors including episodic rest-to-task transitions and major changes in visual luminance (Cole et al., 2019). We finally tested the link between metabolic and neural dynamics by comparing results from dynamic causal modeling (DCM) to those obtained using MCM (Figure 1C).

Changes in local metabolism and neural activity as a function of task difficulty

Relative to resting-state, task performance elicited significant increases in CMRGlu, CBF and BOLD-signals (all p_FWE < 0.05, Figure 2). These cross-modal changes showed a high spatial overlap among each other (Figure 2, Dice coefficient = 0.49–0.55). Significant changes across the three modalities overlapped in the occipital cortex (Occ), the supplementary motor area (SMA), the intraparietal sulcus (IPS) and the frontal eye field (FEF, Figure 2 conjunction map). Increases in task difficulty (hard vs. easy) caused a significant increase in the imaging parameters in these brain regions (all p_Bonf-Holm = 10⁻⁷ to 0.020), with the exception of CMRGlu in SMA as well as CBF in SMA, IPS and FEF. To allow for a focus on cross-modal changes, the regions included in MCM and DCM analyses were Occ, FEF and IPS as obtained from the conjunction analysis (Figure 2). The size of these three regions was 14.4 cm³ (FEF), 33.2 cm³ (IPS) and 22.1 cm³ (Occ), thus providing stable estimates for MCM spatial correlations between functional connectivity and glucose metabolism. The average distance between any voxel of one region to the closest voxel of another region was 46 ± 23 mm, mitigating potential cross-talk between the regions introduced through spatial smoothing.

Figure 2 with 1 supplement see all

Download asset Open asset

Figure 2—source data 1

Task-induced changes CMRGlu, CBF and BOLD signal.

https://cdn.elifesciences.org/articles/52443/elife-52443-fig2-data1-v2.txt

Download elife-52443-fig2-data1-v2.txt

Using BOLD-derived regions as priors for the fPET analysis may bias the direct comparison between the two imaging modalities. Hence, we also used a baseline definition for the fPET data that is independent from the BOLD data. Here, the baseline was modeled by a third-order polynomial, while modeling the task effects as nuisance variables (Hahn et al., 2016). Calculation of CMRGlu changes independent from the BOLD data showed similar but slightly more liberal task-specific changes in CMRGlu (Figure 2—figure supplement 1). The overlap between CMRGlu and BOLD-derived changes was comparable to that detected in the original analysis (Dice coefficient = 0.53).

Task-specific association between glucose metabolism and functional connectivity

We next used MCM to assess the coupling between metabolic processes and BOLD-derived functional connectivity in the key task-related brain regions identified by our previous local analysis. At rest, significant associations between patterns of CMRGlu and functional connectivity were only observed for the link from IPS to the FEF (p_Bonf-Holm = 10⁻⁷, Figure 3). This relationship was stable across the two task conditions (p_Bonf-Holm = 10⁻¹¹ to 10⁻⁷), supporting the notion that the functional interplay between IPS and FEF reflects an intrinsic and task-invariant propriety of brain organization (Fox et al., 2006).

Figure 3 with 3 supplements see all

Download asset Open asset

Figure 3—source data 1

Metabolic connectivity mapping data.

https://cdn.elifesciences.org/articles/52443/elife-52443-fig3-data1-v2.txt

Download elife-52443-fig3-data1-v2.txt

Engagement in the cognitive task showed an interaction effect of condition X connection (p = 0.016) and a main effect of condition for Occ -> IPS, Occ -> FEF and FEF -> IPS (all p_Bonf-Holm = 0.005 to 0.026). Post-hoc paired t-tests revealed that task performance induced increases in the associations between CMRGlu and functional connectivity patterns for Occ -> IPS, Occ -> FEF and FEF -> IPS connections as compared to rest (all p_Bonf-Holm = 0.0004 to 0.02, Figure 3). This result highlights that task performance was specifically supported by interactions from the occipital cortex to IPS, either directly or indirectly via the FEF (Figure 3). A putative mediation effect of FEF was suggested by the observation that Occ -> IPS MCM values decreased in the easy condition (from 0.17 to 0.14, p_Bonf-Holm = 0.017) when controlling for the indirect pathway (Occ -> FEF -> IPS).

Notably, functional connectivity was also sensitive to task performance with significant increases from rest to task for all three connections (Figure 3—figure supplement 1). Taking into account the underlying metabolic demands using MCM allowed us to infer directionality on these task-induced changes in functional connectivity, that is the influence one region exerts on another (Occ -> IPS, Occ -> FEF), as well as a separation of task effects (FEF -> IPS) from intrinsic connectivity (IPS -> FEF).

We further sought to exclude a potential influence of the preprocessing order of functional connectivity (Carp, 2013) and the subsequent MCM estimates. Thus, functional connectivity was also computed with an approach that includes all processing steps (regression against nuisance variables, filtering, motion scrubbing) in a single calculation using one large regression matrix (Hallquist et al., 2013). These analyses yielded similar MCM values as compared to the original computation (Figure 3—figure supplement 2).

A control analysis was performed to assess the impact of spatial smoothing on the MCM estimates (Figure 3—figure supplement 3). As expected, spatial correlation of randomly permuted voxels on unsmoothed data yielded MCM values close to zero (z = 0.00005 ± 0.004). Smoothing still yielded average correlations around zero but with a slightly higher variance (z = −0.001 ± 0.03). This suggests that smoothing does not systematically inflate the MCM estimates, which is also in line with our observation that half of the MCM values were approximately zero despite using smoothed data (Figure 3).

Relationship between metabolic connectivity mapping and effective connectivity

We employed DCM to formally assess the association between MCM and neural dynamics. To achieve this, we constructed a space of all possible directed (effective) connections between key task-related regions (Figure 4—figure supplement 2). Bayesian model selection indicated that the most plausible model of neural interactions across both task conditions (easy and hard) converged with the one identified by MCM (posterior probability of 0.69 vs. 0.13 for the 2^nd best model, Figure 4). To assess the specificity of this result, we also evaluated the relevance of each connection individually using family-wise inference. For a certain connection, all models comprising a task modulation were compared to those models without such modulation. Results showed high probabilities (posterior probability >0.99) for the majority of task-specific modulations (Occ -> FEF, FEF -> IPS) and connections (IPS -> FEF). An exception to this was observed for Occ -> IPS linkage (posterior probability = 0.85, Figure 4). The reduced probability for task-based effective connectivity from occipital (Occ) to parietal (IPS) cortices mirrors the mediation effect of FEF on this connection observed with MCM (previous section). Finally, a contextual modulation of task-specific effective connections as a function of cognitive load was observed for the link FEF -> IPS (p = 0.027). Notably, the DCM results did not change when using a larger ROI size of 8 mm (Figure 4—figure supplement 3).

Figure 4 with 3 supplements see all

Download asset Open asset

Figure 4—source data 1

Probability of DCM models.

https://cdn.elifesciences.org/articles/52443/elife-52443-fig4-data1-v2.txt

Download elife-52443-fig4-data1-v2.txt

Figure 4—source data 2

Probability of DCM family inference.

https://cdn.elifesciences.org/articles/52443/elife-52443-fig4-data2-v2.txt

Download elife-52443-fig4-data2-v2.txt

Using simultaneous PET/MR imaging and multimodal brain network analyses, we studied interdependent changes in neural activity, neurovascular coupling and metabolic demands as a function of cognitive task engagement and load. In addition to increments in regional neural activation and energy consumption, the execution of a cognitively challenging task elicited a marked increase in the association between glucose metabolism and functional network dynamics. In our task, requiring rapid and complex visuo-spatial reasoning, this effect was specific to feedforward connections linking the visual and dorsal attention systems. Once established, the task-based association between metabolic and neural demands reached a plateau and only few effective connections were further modulated by additional cognitive load. These findings highlight the key role of metabolic factors in supporting brain network reconfigurations as a function of cognitive engagement (Bassett et al., 2011; Cocchi et al., 2014; Cole et al., 2013; Hearne et al., 2017).

Task performance led to regional increases in different parameters of energy consumption including CMRGlu, CBF and BOLD-inferred activation. To date, only a few studies using [¹⁸F]FDG fPET imaging have assessed the possible functional relationships between these diverse parameters of brain activity (Horwitz et al., 1995). For example, the association between task-induced BOLD signal changes and glucose metabolism (but not CBF) has been previously studied using simple visual stimulation and finger-tapping (Rischka et al., 2018) and with a visual-perceptual task (Jamadar et al., 2019). Here we extend upon previous work by providing a comprehensive assessment of the neural and metabolic processes supporting performance in a complex cognitive visuo-motor task. We demonstrated that the technique of fPET is sufficiently sensitive to capture metabolic differences between multiple levels of cognitive load. Moreover, our results directly support previous findings, suggesting that regional neural and metabolic changes converge in spatially circumscribed and functionally specialized brain regions (Jamadar et al., 2019; Riedl et al., 2014; Rischka et al., 2018).

Our results provide novel information regarding the link between energy demands and neural dynamics on a whole-brain systems level. We demonstrated that changes in metabolic demands and large-scale network dynamics converge when participants engage in a cognitive task. In other words, the association between glucose metabolism and BOLD-derived functional connectivity strengthened during cognitive performance, compared to a state of rest.

The relationship between blood flow and metabolic factors has been demonstrated to be in balance at rest, whereas task-induced neural activations induce a mismatch (Gusnard et al., 2001; Raichle et al., 2001). Specifically, CMRGlu and CBF markedly increase compared to oxygen consumption in response to increases in local neural activity. The resulting difference between CBF and oxygen consumption is thought to be the major driver of changes in the BOLD signal (Fox and Raichle, 1986; Raichle, 1998). Increases in CBF are linked to glutamate release via both neuronal and astrocytic pathways (Attwell et al., 2010; Mishra et al., 2016), supporting the contribution of glutamate-mediated post-synaptic processing to BOLD signal changes (Logothetis et al., 2001). The coupling between BOLD changes, local high frequency neurophysiological activity and function (Engell et al., 2012) also implicates the contribution of local recurrent feedback to BOLD, as well as activity reflecting efferent output (Siero et al., 2013). On the other hand, the relationship between CBF and CMRGlu is attributable to the downstream effects of glutamate release associated with neuronal activation. In fact, glutamate release also leads to an increased glucose uptake into neurons (Lundgaard et al., 2015) and astrocytes (Zimmer et al., 2017), to provide energy for the reversal of ion gradients and glutamate recycling (Harris et al., 2012; Magistretti and Allaman, 2015; Raichle and Mintun, 2006). These considerations highlight the linked but distinct mechanisms underpinning BOLD fMRI and [¹⁸F]FDG PET signals. The asymmetry of these two different imaging techniques motivates the assumptions underlying MCM. Although the physiological mechanisms coupling CBF and CMRGlu most likely remain the same, our findings indicate that the time-dependent, nonlinear interactions between synaptic processing, energy consumption and neurovascular responses across macroscopic brain regions become apparent when engaging in a cognitively demanding task. The ability of functionally specialized brain regions to dynamically change their connectivity patterns is essential for cognitive processing (Cocchi et al., 2013; Cole et al., 2013; Hearne et al., 2019). Our results extend this knowledge by highlighting that external task engagement is accompanied by the convergence between metabolic factors and functional brain network reconfigurations.

In a state of rest, the association between glucose metabolism and functional connectivity was confined within a given functional system, namely the dorsal attention network. In contrast, task performance elicited the emergence of marked interactions between brain regions known to be part of distinct resting-state networks including the visual and dorsal attention systems. These observations support previous results on the emergence of between-system interactions during visual task execution (Riedl et al., 2016). In contrast to the substantial changes associated with the transition from rest to task, we observed that additional cognitive load does not substantially influence the association between metabolic and neural processes (Figure 3). These findings suggest that departing from the intrinsic (resting-state) network configuration carries the bulk of the metabolic cost associated with external task engagement. Considering that the availability of glucose in the brain is carefully maintained (unless experimentally or pathologically altered) (Dienel, 2019), our results put forward the hypothesis that the functional limits of efficient cognitive performance are largely related to neural processes supporting dynamic interactions between remote brain regions. Our work provides a strong motivation for future studies aiming to test this hypothesis and assess to what extent cognitive limits may depend on the efficiency of neural ensembles to use the underlying metabolic resources.

Within the task-based network (Figures 3 and 4), MCM and DCM allowed inferences regarding the directionality of functional connectivity between brain regions supporting task processes. Specifically, the use of these techniques draws upon metabolic knowledge (MCM) and computational modeling (DCM) to disambiguate the task-related changes in local engagement from inter-regional interactions, which compound classic linear measures of functional connectivity (Cole et al., 2019). Our results showed that task engagement was associated with the emergence of distinct patterns of effective connectivity from the visual to the dorsal attention networks. DCM further highlighted fine-grained modulations of specific connections as a function of increased task load. The observed bottom-up processing of visual input is in line with the role of attention networks in stimulus-driven control of attention (Corbetta and Shulman, 2002; Vossel et al., 2014) as well as the observed emergence of feedforward interaction with the FEF following stimulation of the visual cortex via transcranial magnetic stimulation (Castrillon et al., 2020; Cocchi et al., 2016). On the other hand, the lack of feedback control of the dorsal attention network over visual areas is consistent with studies showing that top-down modulation only emerges when distractors need to be filtered or context specific information extracted (Corbetta and Shulman, 2002; Vossel et al., 2014), which was not the case in our task.

Estimating measures of directed functional connectivity is a considerable challenge, particularly when those measures depend purely upon conditional likelihoods in the data (Smith et al., 2011), or in networks with closed cycles (Ramsey et al., 2010). In this context, it is worth considering the complementary nature of the two different approaches (DCM and MCM) used in this work to infer causal directionality. Whereas DCM requires careful a priori selection of regions and specification of a model space, MCM is less vulnerable to these issues (Riedl et al., 2016). MCM represents a simplified index of effective connectivity that draws upon the coupling between glutamate-mediated changes in the BOLD signal and metabolic demands. On the other hand, DCM provides a formal assessment of interactions between neural populations using a computational model that has been extensively validated (e.g., Stephan et al., 2008). By using a generative model and an observation (HRF) equation, DCM also avoids reliance upon making direct inference on conditional dependences in data, a challenge for many measures of directed functional connectivity (Smith et al., 2011).

In interpreting our findings, we acknowledge that the limited sample size of this study may have diminished the ability to detect more subtle changes in the coupling between metabolic and neural processes. Although small changes between the two task conditions may indeed be present or further brain regions may have been involved, these effects more likely play only a minor part in processing of the employed task and are therefore unlikely to substantially change the implications of our results. Also, our task only tested for a circumscribed set of visual-motor cognitive domains. While it is unlikely that the observed effects are specific to a given set of cognitive functions, future work is required to expand current findings to different domains. By successfully linking metabolic and functional network dynamics, the current study provides new methodological and neurobiological frameworks to understand the nature of pathological alterations of brain functions. Specifically, our paradigm and findings may help unfold the complex interdependent effects of cognitive, metabolic, and neural factors in pathologies such as epilepsy, schizophrenia and Alzheimer’s disease (Scherr et al., 2019).

We cannot publicly share raw data for reasons of data protection; processed data supporting the results of this manuscript have been deposited to Dryad Digital Repository under the DOI https://doi.org/10.5061/dryad.zcrjdfn7p.

1. 1. Hahn A
   2. Breakspear M
   3. Rischka L
   4. Wadsak W
   5. Godbersen GM
   6. Pichler V
   7. Michenthaler P
   8. Vanicek T
   9. Hacker M
   10. Kasper S
   11. Lanzenberger R
   12. Cocchi L

   (2020) Dryad Digital Repository

   Data from: Reconfiguration of functional brain networks and metabolic cost converge during task performance.

   https://doi.org/10.5061/dryad.zcrjdfn7p

1. 1. Aquino KM
   2. Schira MM
   3. Robinson PA
   4. Drysdale PM
   5. Breakspear M

   (2012) Hemodynamic traveling waves in human visual cortex

   PLOS Computational Biology 8:e1002435.

   https://doi.org/10.1371/journal.pcbi.1002435

   • PubMed
   • Google Scholar
2. 1. Attwell D
   2. Buchan AM
   3. Charpak S
   4. Lauritzen M
   5. Macvicar BA
   6. Newman EA

   (2010) Glial and neuronal control of brain blood flow

   Nature 468:232–243.

   https://doi.org/10.1038/nature09613

   • PubMed
   • Google Scholar
3. 1. Attwell D
   2. Laughlin SB

   (2001) An energy budget for signaling in the grey matter of the brain

   Journal of Cerebral Blood Flow & Metabolism 21:1133–1145.

   https://doi.org/10.1097/00004647-200110000-00001

   • PubMed
   • Google Scholar
4. 1. Baron RM
   2. Kenny DA

   (1986) The moderator-mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations

   Journal of Personality and Social Psychology 51:1173–1182.

   https://doi.org/10.1037/0022-3514.51.6.1173

   • PubMed
   • Google Scholar
5. 1. Bassett DS
   2. Wymbs NF
   3. Porter MA
   4. Mucha PJ
   5. Carlson JM
   6. Grafton ST

   (2011) Dynamic reconfiguration of human brain networks during learning

   PNAS 108:7641–7646.

   https://doi.org/10.1073/pnas.1018985108

   • PubMed
   • Google Scholar
6. 1. Braun U
   2. Schäfer A
   3. Walter H
   4. Erk S
   5. Romanczuk-Seiferth N
   6. Haddad L
   7. Schweiger JI
   8. Grimm O
   9. Heinz A
   10. Tost H
   11. Meyer-Lindenberg A
   12. Bassett DS

   (2015) Dynamic reconfiguration of frontal brain networks during executive cognition in humans

   PNAS 112:11678–11683.

   https://doi.org/10.1073/pnas.1422487112

   • PubMed
   • Google Scholar
7. 1. Burgos N
   2. Cardoso MJ
   3. Thielemans K
   4. Modat M
   5. Pedemonte S
   6. Dickson J
   7. Barnes A
   8. Ahmed R
   9. Mahoney CJ
   10. Schott JM
   11. Duncan JS
   12. Atkinson D
   13. Arridge SR
   14. Hutton BF
   15. Ourselin S

   (2014) Attenuation correction synthesis for hybrid PET-MR scanners: application to brain studies

   IEEE Transactions on Medical Imaging 33:2332–2341.

   https://doi.org/10.1109/TMI.2014.2340135

   • PubMed
   • Google Scholar
8. 1. Carp J

   (2013) Optimizing the order of operations for movement scrubbing: comment on power et al

   NeuroImage 76:436–438.

   https://doi.org/10.1016/j.neuroimage.2011.12.061

   • PubMed
   • Google Scholar
9. 1. Castrillon G
   2. Sollmann N
   3. Kurcyus K
   4. Razi A
   5. Krieg SM
   6. Riedl V

   (2020) The physiological effects of noninvasive brain stimulation fundamentally differ across the human cortex

   Science Advances 6:eaay2739.

   https://doi.org/10.1126/sciadv.aay2739

   • PubMed
   • Google Scholar
10. 1. Cocchi L
    2. Zalesky A
    3. Fornito A
    4. Mattingley JB

    (2013) Dynamic cooperation and competition between brain systems during cognitive control

    Trends in Cognitive Sciences 17:493–501.

    https://doi.org/10.1016/j.tics.2013.08.006

    • PubMed
    • Google Scholar
11. 1. Cocchi L
    2. Halford GS
    3. Zalesky A
    4. Harding IH
    5. Ramm BJ
    6. Cutmore T
    7. Shum DH
    8. Mattingley JB

    (2014) Complexity in relational processing predicts changes in functional brain network dynamics

    Cerebral Cortex 24:2283–2296.

    https://doi.org/10.1093/cercor/bht075

    • PubMed
    • Google Scholar
12. 1. Cocchi L
    2. Sale MV
    3. L Gollo L
    4. Bell PT
    5. Nguyen VT
    6. Zalesky A
    7. Breakspear M
    8. Mattingley JB

    (2016) A hierarchy of timescales explains distinct effects of local inhibition of primary visual cortex and frontal eye fields

    eLife 5:e15252.

    https://doi.org/10.7554/eLife.15252

    • PubMed
    • Google Scholar
13. 1. Cole MW
    2. Reynolds JR
    3. Power JD
    4. Repovs G
    5. Anticevic A
    6. Braver TS

    (2013) Multi-task connectivity reveals flexible hubs for adaptive task control

    Nature Neuroscience 16:1348–1355.

    https://doi.org/10.1038/nn.3470

    • PubMed
    • Google Scholar
14. 1. Cole MW
    2. Ito T
    3. Schultz D
    4. Mill R
    5. Chen R
    6. Cocuzza C

    (2019) Task activations produce spurious but systematic inflation of task functional connectivity estimates

    NeuroImage 189:1–18.

    https://doi.org/10.1016/j.neuroimage.2018.12.054

    • PubMed
    • Google Scholar
15. 1. Corbetta M
    2. Shulman GL

    (2002) Control of goal-directed and stimulus-driven attention in the brain

    Nature Reviews Neuroscience 3:201–215.

    https://doi.org/10.1038/nrn755

    • PubMed
    • Google Scholar
16. 1. Dienel GA

    (2019) Brain glucose metabolism: integration of energetics with function

    Physiological Reviews 99:949–1045.

    https://doi.org/10.1152/physrev.00062.2017

    • PubMed
    • Google Scholar
17. 1. Engell AD
    2. Huettel S
    3. McCarthy G

    (2012) The fMRI BOLD signal tracks electrophysiological spectral perturbations, not event-related potentials

    NeuroImage 59:2600–2606.

    https://doi.org/10.1016/j.neuroimage.2011.08.079

    • PubMed
    • Google Scholar
18. 1. Fox MD
    2. Corbetta M
    3. Snyder AZ
    4. Vincent JL
    5. Raichle ME

    (2006) Spontaneous neuronal activity distinguishes human dorsal and ventral attention systems

    PNAS 103:10046–10051.

    https://doi.org/10.1073/pnas.0604187103

    • PubMed
    • Google Scholar
19. 1. Fox PT
    2. Raichle ME

    (1986) Focal physiological uncoupling of cerebral blood flow and oxidative metabolism during somatosensory stimulation in human subjects

    PNAS 83:1140–1144.

    https://doi.org/10.1073/pnas.83.4.1140

    • PubMed
    • Google Scholar
20. 1. Friston KJ
    2. Harrison L
    3. Penny W

    (2003) Dynamic causal modelling

    NeuroImage 19:1273–1302.

    https://doi.org/10.1016/S1053-8119(03)00202-7

    • PubMed
    • Google Scholar
21. 1. Gusnard DA
    2. Raichle ME
    3. Raichle ME

    (2001) Searching for a baseline: functional imaging and the resting human brain

    Nature Reviews Neuroscience 2:685–694.

    https://doi.org/10.1038/35094500

    • PubMed
    • Google Scholar
22. 1. Hahn A
    2. Gryglewski G
    3. Nics L
    4. Hienert M
    5. Rischka L
    6. Vraka C
    7. Sigurdardottir H
    8. Vanicek T
    9. James GM
    10. Seiger R
    11. Kautzky A
    12. Silberbauer L
    13. Wadsak W
    14. Mitterhauser M
    15. Hacker M
    16. Kasper S
    17. Lanzenberger R

    (2016) Quantification of Task-Specific glucose metabolism with constant infusion of 18F-FDG

    Journal of Nuclear Medicine 57:1933–1940.

    https://doi.org/10.2967/jnumed.116.176156

    • PubMed
    • Google Scholar
23. 1. Hahn A
    2. Gryglewski G
    3. Nics L
    4. Rischka L
    5. Ganger S
    6. Sigurdardottir H
    7. Vraka C
    8. Silberbauer L
    9. Vanicek T
    10. Kautzky A
    11. Wadsak W
    12. Mitterhauser M
    13. Hartenbach M
    14. Hacker M
    15. Kasper S
    16. Lanzenberger R

    (2018) Task-relevant brain networks identified with simultaneous PET/MR imaging of metabolism and connectivity

    Brain Structure & Function 223:1369–1378.

    https://doi.org/10.1007/s00429-017-1558-0

    • PubMed
    • Google Scholar
24. 1. Haier RJ
    2. Karama S
    3. Leyba L
    4. Jung RE

    (2009) MRI assessment of cortical thickness and functional activity changes in adolescent girls following three months of practice on a visual-spatial task

    BMC Research Notes 2:174.

    https://doi.org/10.1186/1756-0500-2-174

    • PubMed
    • Google Scholar
25. 1. Hallquist MN
    2. Hwang K
    3. Luna B

    (2013) The nuisance of nuisance regression: spectral misspecification in a common approach to resting-state fMRI preprocessing reintroduces noise and obscures functional connectivity

    NeuroImage 82:208–225.

    https://doi.org/10.1016/j.neuroimage.2013.05.116

    • PubMed
    • Google Scholar
26. 1. Harris JJ
    2. Jolivet R
    3. Attwell D

    (2012) Synaptic energy use and supply

    Neuron 75:762–777.

    https://doi.org/10.1016/j.neuron.2012.08.019

    • PubMed
    • Google Scholar
27. 1. Hearne LJ
    2. Cocchi L
    3. Zalesky A
    4. Mattingley JB

    (2017) Reconfiguration of brain network architectures between Resting-State and Complexity-Dependent cognitive reasoning

    The Journal of Neuroscience 37:8399–8411.

    https://doi.org/10.1523/JNEUROSCI.0485-17.2017

    • PubMed
    • Google Scholar
28. 1. Hearne LJ
    2. Dean RJ
    3. Robinson GA
    4. Richards LJ
    5. Mattingley JB
    6. Cocchi L

    (2019) Increased cognitive complexity reveals abnormal brain network activity in individuals with corpus callosum dysgenesis

    NeuroImage: Clinical 21:101595.

    https://doi.org/10.1016/j.nicl.2018.11.005

    • PubMed
    • Google Scholar
29. 1. Heeger DJ
    2. Ress D

    (2002) What does fMRI tell us about neuronal activity?

    Nature Reviews Neuroscience 3:142–151.

    https://doi.org/10.1038/nrn730

    • PubMed
    • Google Scholar
30. 1. Horwitz B
    2. McIntosh AR
    3. Haxby JV
    4. Grady CL

    (1995) Network analysis of brain cognitive function using metabolic and blood flow data

    Behavioural Brain Research 66:187–193.

    https://doi.org/10.1016/0166-4328(94)00139-7

    • PubMed
    • Google Scholar
31. 1. Jamadar SD
    2. Ward PG
    3. Li S
    4. Sforazzini F
    5. Baran J
    6. Chen Z
    7. Egan GF

    (2019) Simultaneous task-based BOLD-fMRI and [18-F] FDG functional PET for measurement of neuronal metabolism in the human visual cortex

    NeuroImage 189:258–266.

    https://doi.org/10.1016/j.neuroimage.2019.01.003

    • PubMed
    • Google Scholar
32. 1. Kilroy E
    2. Apostolova L
    3. Liu C
    4. Yan L
    5. Ringman J
    6. Wang DJ

    (2014) Reliability of two-dimensional and three-dimensional pseudo-continuous arterial spin labeling perfusion MRI in elderly populations: comparison with 15O-water positron emission tomography

    Journal of Magnetic Resonance Imaging 39:931–939.

    https://doi.org/10.1002/jmri.24246

    • PubMed
    • Google Scholar
33. 1. Logothetis NK
    2. Pauls J
    3. Augath M
    4. Trinath T
    5. Oeltermann A

    (2001) Neurophysiological investigation of the basis of the fMRI signal

    Nature 412:150–157.

    https://doi.org/10.1038/35084005

    • PubMed
    • Google Scholar
34. 1. Lundgaard I
    2. Li B
    3. Xie L
    4. Kang H
    5. Sanggaard S
    6. Haswell JD
    7. Sun W
    8. Goldman S
    9. Blekot S
    10. Nielsen M
    11. Takano T
    12. Deane R
    13. Nedergaard M

    (2015) Direct neuronal glucose uptake heralds activity-dependent increases in cerebral metabolism

    Nature Communications 6:6807.

    https://doi.org/10.1038/ncomms7807

    • PubMed
    • Google Scholar
35. 1. Magistretti PJ
    2. Allaman I

    (2015) A cellular perspective on brain energy metabolism and functional imaging

    Neuron 86:883–901.

    https://doi.org/10.1016/j.neuron.2015.03.035

    • PubMed
    • Google Scholar
36. 1. McIntosh AR
    2. Grady CL
    3. Ungerleider LG
    4. Haxby JV
    5. Rapoport SI
    6. Horwitz B

    (1994) Network analysis of cortical visual pathways mapped with PET

    The Journal of Neuroscience 14:655–666.

    https://doi.org/10.1523/JNEUROSCI.14-02-00655.1994

    • PubMed
    • Google Scholar
37. 1. Mergenthaler P
    2. Lindauer U
    3. Dienel GA
    4. Meisel A

    (2013) Sugar for the brain: the role of glucose in physiological and pathological brain function

    Trends in Neurosciences 36:587–597.

    https://doi.org/10.1016/j.tins.2013.07.001

    • PubMed
    • Google Scholar
38. 1. Mishra A
    2. Reynolds JP
    3. Chen Y
    4. Gourine AV
    5. Rusakov DA
    6. Attwell D

    (2016) Astrocytes mediate neurovascular signaling to capillary pericytes but not to arterioles

    Nature Neuroscience 19:1619–1627.

    https://doi.org/10.1038/nn.4428

    • PubMed
    • Google Scholar
39. 1. Park HJ
    2. Friston K

    (2013) Structural and functional brain networks: from connections to cognition

    Science 342:1238411.

    https://doi.org/10.1126/science.1238411

    • PubMed
    • Google Scholar
40. 1. Penny WD
    2. Mattout J
    3. Trujillo-Barreto N

    (2007)

    Statistical Parametric Mapping

    Bayesian model selection and averaging, Statistical Parametric Mapping, London, Academic Press.

    • Google Scholar
41. 1. Power JD
    2. Schlaggar BL
    3. Petersen SE

    (2015) Recent progress and outstanding issues in motion correction in resting state fMRI

    NeuroImage 105:536–551.

    https://doi.org/10.1016/j.neuroimage.2014.10.044

    • PubMed
    • Google Scholar
42. 1. Raichle ME

    (1998) Behind the scenes of functional brain imaging: a historical and physiological perspective

    PNAS 95:765–772.

    https://doi.org/10.1073/pnas.95.3.765

    • Google Scholar
43. 1. Raichle ME
    2. MacLeod AM
    3. Snyder AZ
    4. Powers WJ
    5. Gusnard DA
    6. Shulman GL

    (2001) A default mode of brain function

    PNAS 98:676–682.

    https://doi.org/10.1073/pnas.98.2.676

    • Google Scholar
44. 1. Raichle ME
    2. Mintun MA

    (2006) Brain work and brain imaging

    Annual Review of Neuroscience 29:449–476.

    https://doi.org/10.1146/annurev.neuro.29.051605.112819

    • PubMed
    • Google Scholar
45. 1. Ramsey JD
    2. Hanson SJ
    3. Hanson C
    4. Halchenko YO
    5. Poldrack RA
    6. Glymour C

    (2010) Six problems for causal inference from fMRI

    NeuroImage 49:1545–1558.

    https://doi.org/10.1016/j.neuroimage.2009.08.065

    • PubMed
    • Google Scholar
46. 1. Riedl V
    2. Bienkowska K
    3. Strobel C
    4. Tahmasian M
    5. Grimmer T
    6. Förster S
    7. Friston KJ
    8. Sorg C
    9. Drzezga A

    (2014) Local activity determines functional connectivity in the resting human brain: a simultaneous FDG-PET/fMRI study

    Journal of Neuroscience 34:6260–6266.

    https://doi.org/10.1523/JNEUROSCI.0492-14.2014

    • PubMed
    • Google Scholar
47. 1. Riedl V
    2. Utz L
    3. Castrillón G
    4. Grimmer T
    5. Rauschecker JP
    6. Ploner M
    7. Friston KJ
    8. Drzezga A
    9. Sorg C

    (2016) Metabolic connectivity mapping reveals effective connectivity in the resting human brain

    PNAS 113:428–433.

    https://doi.org/10.1073/pnas.1513752113

    • PubMed
    • Google Scholar
48. 1. Rischka L
    2. Gryglewski G
    3. Pfaff S
    4. Vanicek T
    5. Hienert M
    6. Klöbl M
    7. Hartenbach M
    8. Haug A
    9. Wadsak W
    10. Mitterhauser M
    11. Hacker M
    12. Kasper S
    13. Lanzenberger R
    14. Hahn A

    (2018) Reduced task durations in functional PET imaging with [¹⁸F]FDG approaching that of functional MRI

    NeuroImage 181:323–330.

    https://doi.org/10.1016/j.neuroimage.2018.06.079

    • PubMed
    • Google Scholar
49. 1. Scherr M
    2. Utz L
    3. Tahmasian M
    4. Pasquini L
    5. Grothe MJ
    6. Rauschecker JP
    7. Grimmer T
    8. Drzezga A
    9. Sorg C
    10. Riedl V

    (2019) Effective connectivity in the default mode network is distinctively disrupted in Alzheimer's disease-A simultaneous resting-state FDG-PET/fMRI study

    Human Brain Mapping 31:.

    https://doi.org/10.1002/hbm.24517

    • Google Scholar
50. 1. Siero JC
    2. Hermes D
    3. Hoogduin H
    4. Luijten PR
    5. Petridou N
    6. Ramsey NF

    (2013) BOLD consistently matches electrophysiology in human sensorimotor cortex at increasing movement rates: a combined 7T fMRI and ECoG study on neurovascular coupling

    Journal of Cerebral Blood Flow & Metabolism 33:1448–1456.

    https://doi.org/10.1038/jcbfm.2013.97

    • PubMed
    • Google Scholar
51. 1. Smith SM

    (2002) Fast robust automated brain extraction

    Human Brain Mapping 17:143–155.

    https://doi.org/10.1002/hbm.10062

    • PubMed
    • Google Scholar
52. 1. Smith SM
    2. Fox PT
    3. Miller KL
    4. Glahn DC
    5. Fox PM
    6. Mackay CE
    7. Filippini N
    8. Watkins KE
    9. Toro R
    10. Laird AR
    11. Beckmann CF

    (2009) Correspondence of the brain's functional architecture during activation and rest

    PNAS 106:13040–13045.

    https://doi.org/10.1073/pnas.0905267106

    • PubMed
    • Google Scholar
53. 1. Smith SM
    2. Miller KL
    3. Salimi-Khorshidi G
    4. Webster M
    5. Beckmann CF
    6. Nichols TE
    7. Ramsey JD
    8. Woolrich MW

    (2011) Network modelling methods for FMRI

    NeuroImage 54:875–891.

    https://doi.org/10.1016/j.neuroimage.2010.08.063

    • PubMed
    • Google Scholar
54. 1. Stephan KE
    2. Kasper L
    3. Harrison LM
    4. Daunizeau J
    5. den Ouden HE
    6. Breakspear M
    7. Friston KJ

    (2008) Nonlinear dynamic causal models for fMRI

    NeuroImage 42:649–662.

    https://doi.org/10.1016/j.neuroimage.2008.04.262

    • PubMed
    • Google Scholar
55. 1. Stephan KE
    2. Penny WD
    3. Moran RJ
    4. den Ouden HE
    5. Daunizeau J
    6. Friston KJ

    (2010) Ten simple rules for dynamic causal modeling

    NeuroImage 49:3099–3109.

    https://doi.org/10.1016/j.neuroimage.2009.11.015

    • PubMed
    • Google Scholar
56. 1. Sun FT
    2. Miller LM
    3. D'Esposito M

    (2004) Measuring interregional functional connectivity using coherence and partial coherence analyses of fMRI data

    NeuroImage 21:647–658.

    https://doi.org/10.1016/j.neuroimage.2003.09.056

    • PubMed
    • Google Scholar
57. 1. Vossel S
    2. Geng JJ
    3. Fink GR

    (2014) Dorsal and ventral attention systems: distinct neural circuits but collaborative roles

    The Neuroscientist : A Review Journal Bringing Neurobiology, Neurology and Psychiatry 20:150–159.

    https://doi.org/10.1177/1073858413494269

    • PubMed
    • Google Scholar
58. 1. Wang J
    2. Zhang Y
    3. Wolf RL
    4. Roc AC
    5. Alsop DC
    6. Detre JA

    (2005) Amplitude-modulated continuous arterial spin-labeling 3.0-T perfusion MR imaging with a single coil: feasibility study

    Radiology 235:218–228.

    https://doi.org/10.1148/radiol.2351031663

    • PubMed
    • Google Scholar
59. 1. Zimmer ER
    2. Parent MJ
    3. Souza DG
    4. Leuzy A
    5. Lecrux C
    6. Kim HI
    7. Gauthier S
    8. Pellerin L
    9. Hamel E
    10. Rosa-Neto P

    (2017) [¹⁸F]FDG PET signal is driven by astroglial glutamate transport

    Nature Neuroscience 20:393–395.

    https://doi.org/10.1038/nn.4492

    • PubMed
    • Google Scholar

Acceptance summary:

Hahn and colleagues analyzed data from 22 participants performing a complex cognitive task (Tetris) during concurrent fMRI-PET. The study demonstrated that performing the complex cognitive task yielded significant reconfiguration of brain connectivity (relative to rest). Yet increases in task difficulty yielded relatively small changes (compared with between rest and task). The convergence between two very different approaches – Metabolic Connectivity Mapping (based on PET-MR) and dynamic causal modeling (based on fMRI) – was impressive.

Decision letter after peer review:

Thank you for submitting your article "Reconfiguration of functional brain networks and metabolic cost converge during task performance" for consideration by eLife. Your article has been reviewed by three peer reviewers, one of whom is a member of our Board of Reviewing Editors, and the evaluation has been overseen by Timothy Behrens as the Senior Editor.

The reviewers have discussed the reviews with one another and the Reviewing Editor has drafted this decision to help you prepare a revised submission.

Summary:

In the current manuscript, Hahn and colleagues analyzed data from 22 participants in a concurrent fMRI-PET dataset. Participants were either at rest or asked to play a Tetris game at two different levels of difficulty. While playing, fMRI BOLD or ASL was collected concurrently with FDG PET imaging. In addition, a set of baseline fMRI (BOLD and ASL) resting state scans and a final set of blocked BOLD task scans were collected in the absence of PET imaging. With this data, the authors examined the overlap of glucose metabolism (PET), BOLD, and cerebral blood flow (ASL) during the task. The authors then used a combination of the PET and BOLD data to make estimates of directed connectivity during the Tetris task, a technique called Metabolic Connectivity Mapping (MCM), and compared the results of this MCM analysis to a DCM analysis on the BOLD data.

This is a unique and interesting dataset, providing the authors with concurrent measurements of both fMRI and PET measures. The first analysis (Figure 2) demonstrating substantial overlap between glucose metabolism, BOLD, and ASL appears quite compelling. The convergence between two very different approaches (DCM and MCM; Figure 3) is quite impressive.

Essential revisions:

1) Throughout the manuscript, the authors refer to BOLD measurements as "neural activation" measurements. This is imprecise terminology to use, especially in the current context, where the BOLD measurements are being contrasted with measurements of cerebral blood flow and glucose metabolism. Likely all of these measures are at some level related to neural activity, but all are indirect, and it's not clear why BOLD measurements receive a privileged status on this front in this manuscript. I would strongly recommend shifting this terminology.

2) This terminological distinction is important because it obscured the approach used in the MCM analyses in this paper. The authors argue that glucose metabolism tends to occur largely post-synaptically, and thus a comparison between BOLD functional connectivity and PET glucose metabolism profiles allows for the determination of directed information flow. But there is also substantial evidence that the BOLD signal is better related to neural post-synaptic potentials than action potentials (e.g., Logothetis et al., 2001). Thus this logic is unclear. It would be helpful if the authors could expand further on the motivation and validation for this analysis.

3) Given general issues in the field with interpreting directed connectivity (e.g., Ramsey et al., 2010; Smith et al., 2011), validation with DCM seems like an insufficient standard. The authors should be explicit about the pitfalls of both approaches (DCM and MCM) and whether they cover each other's pitfalls. For example, DCM is probably sensitive to missing nodes, while MCM won't have that issue (as far as we can tell), so the two approaches nicely complement each other in this respect. Are there pitfalls that apply to both DCM and MCM? If so, how serious are these pitfalls and can they be addressed with control analyses?

4) Given the large amount of spatial smoothing that was performed across all modalities (8 mm), how informative is a spatial correlation calculated within ROIs for MCM? One way to test for this more robustly is to perform permutation testing by comparing a BOLD connectivity map from one subject to a glucose metabolism map of a different subject (and build a null distribution through many such permutations), rather than performing a basic t-test against zero (and similarly for the comparison against rest). This will determine whether spatial similarity is truly indicative of connectivity-metabolic coupling, or whether it is observed as a function of the large smoothing kernel (and relatively small ROIs).

5) Why were DCM ROIs different from the MCM ROIs (Figure 4—figure supplement 1)? If DCM is used for validation, it seems to make sense to use the same regions in both analyses. Furthermore, how does the interaction between the size of the ROIs (5 mm) and the smoothing kernel (8mm) affect the results?

6) Some of the statistical tests are unclear. The authors wrote that "the significance of each connection was tested separately using one-sample t-tests against zero". So let's consider one connection: FEF to IPS. The MCM analysis yields one spatial correlation for each subject. So can I confirm the t-test involves 22 numbers (corresponding to the 22 subjects)? Because a p value of 10e-12 is quite impressive with only 22 subjects; one would need a t-score of around 14?

7) For Figure 3, the authors say that the thickness of arrows depend on r values. More details will be useful. For example, when looking at Occ -> FEF, the two box plots for easy and hard conditions look highly overlapping, but the green arrow was much thicker in the hard condition than easy condition. And from what I gather from the plot, the IPS->FEF during hard condition has higher MCM correlation than Occ->FEF during hard condition. So why is the green arrow thickness than the yellow arrow during the hard condition?

8) It would be important to report the pure functional connectivity among the 3 regions. More specifically, are there region pairs with strong functional connectivity, but the MCM connectivity is weak in both directions. If so, how can this be interpreted? Similarly, are there regions with weak functional connectivity but the MCM connectivity is strong? If so, how can this be interpreted?

9) The DCM modeling needs to be further elaborated. Maybe the current details are sufficient for a practiced DCM user, but as someone who is somewhat familiar with DCM (but do not actively use it), I found it hard to know how task modulation and intrinsic connection were modeled in the generative model.

10) "A putative mediation effect of FEF was suggested by the observation that Occ -> IPS MCM values decreased in the easy condition (from 0.17 to 0.14, pBonf-Holm = 0.017) when controlling for the indirect pathway (Occ -> FEF -> IPS)." – Please elaborate on how this is done.

11) Literature review and discussion.

A) Given the nature of the paper, the authors should include a discussion of how their findings compare to a large and long-standing literature on cerebral blood flow and metabolism, which explicitly contrasts cerebral blood flow, oxygen consumption, and glucose metabolism to one another in and outside of tasks contexts, e.g., from Marc Raichle and others (see for example: Gusnard and Raichle, 2001). The current work would have substantially greater impact if it were to engage more deeply with this older literature. For example, do the authors mean to suggest (as might appear from their statement in the Abstract that "dynamic regulation of neural-metabolic coupling is essential to support human cognition") that different mechanisms exist to couple neural activity with glucose consumption depending on the task at hand? This claim seems quite large and far reaching given the evidence, and if true, would have profound implications in invalidating the current use and interpretation of many neuroimaging methods.

B) The authors should also be more precise about differences with previous work (e.g., Jamadar et al., 2019; Rischka et al., 2018). For example, "Here we extend upon previous work by providing a comprehensive assessment of the neural and metabolic processes supporting performance in a complex cognitive visuo-motor task" can be briefly expanded to mention that Rischka investigated a finger-tapping task and Jamadar investigated a visual checkerboard task.

12) The Discussion states: "Our results extend this knowledge by highlighting that the convergence between metabolic factors and functional network reconfiguration is pivotal to support cognition". However, the relationship with cognition or performance was not tested here (e.g. MCM was not shown to be stronger for those subjects who performed better). As such, I would recommend adjusting this interpretation.

Essential revisions:

1) Throughout the manuscript, the authors refer to BOLD measurements as "neural activation" measurements. This is imprecise terminology to use, especially in the current context, where the BOLD measurements are being contrasted with measurements of cerebral blood flow and glucose metabolism. Likely all of these measures are at some level related to neural activity, but all are indirect, and it's not clear why BOLD measurements receive a privileged status on this front in this manuscript. I would strongly recommend shifting this terminology.

We agree with the editors/reviewers that all three imaging modalities reflect different aspects of neural activation. Thus, when referring to BOLD measurements the term “neural activation” was changed to best match the specific context (see examples below).

Results:

“We first integrated information regarding task-related changes in local BOLD signals, cerebral blood flow (CBF) and glucose consumption to identify functionally segregated brain regions involved in task performance (Figure 1C). This analysis leveraged the recently introduced approach of functional PET (fPET) imaging to determine task-specific changes in the cerebral metabolic rate of glucose (CMRGlu) (Hahn et al., 2016; Rischka et al., 2018) as well as canonical analyses for BOLD and ASL signal changes.”

Results:

“Relative to resting-state, task performance elicited significant increases in CMRGlu, CBF and BOLD-signals […]”

Materials and methods:

“Data from this BOLD block design allowed inference of task-specific changes in neural activation and dynamics using DCM.”

Materials and methods:

“BOLD changes

Task-induced changes in BOLD signal were estimated using data from the block design acquired in the final part of the experiment (Figure 1A, red box). […] To estimate task-specific effects the general linear model was used with one regressor for each condition (easy, hard, control) and nuisance regressors for movement, white matter and cerebrospinal fluid.”

Materials and methods:

“To provide a robust identification of the network involved in task performance, the different metabolic demands reflected by regional CMRGlu, CBF and BOLD signal changes were combined.”

Figure 1:

“In the final part of the experiment, participants performed 12 task blocks (easy, hard and control, 30 s for each block, red) during BOLD acquisition to estimate neural task effects and to compute effective connectivity using dynamic causal modelling. […]

First, brain regions involved in task processing were identified as conjunction of task-specific changes in glucose metabolism (fPET), blood oxygenation (BOLD) and blood flow (ASL).”

Figure 2:

“Dot plots depict within subjects changes in glucose metabolism (CMRGlu), BOLD signal and cerebral blood flow (CBF) as a function of increasing task load in the frontal eye field (FEF, from the conjunction analysis; ^(*)p = 0.07, ***p < 0.001, corrected; large circles indicate group mean values).”

2) This terminological distinction is important because it obscured the approach used in the MCM analyses in this paper. The authors argue that glucose metabolism tends to occur largely post-synaptically, and thus a comparison between BOLD functional connectivity and PET glucose metabolism profiles allows for the determination of directed information flow. But there is also substantial evidence that the BOLD signal is better related to neural post-synaptic potentials than action potentials (e.g., Logothetis et al., 2001). Thus this logic is unclear. It would be helpful if the authors could expand further on the motivation and validation for this analysis.

While it is true that post-synaptic responses make a strong contribution to both glucose metabolism and BOLD responses, considerable work since the seminal paper by Logothetis et al. has shown that BOLD reflects more of a composite of local activity which includes post-synaptic activity arising from distant inputs, but also local recurrent feedback and output-related activity. The latter can be seen in studies that show a link between BOLD responses in M1 and motor movement rates. We have expanded on the explication of MCM and also disambiguated between fPET and BOLD signal changes):

“The main assumption of the approach is that the majority of energy demands arise post-synaptically, i.e., in the target region (Attwell and Laughlin, 2001; Harris et al., 2012; Mergenthaler et al., 2013; Riedl et al., 2016) which are tightly linked to post-synaptic glutamate-mediated processes (Attwell et al., 2010; Mishra et al., 2016). […] MCM exploits this asymmetry across data modalities: If the BOLD-derived pattern of functional connectivity in region B is indeed caused by the influence region A exerts on region B (i.e., testing directionality of pairwise interactions), this will yield a corresponding pattern of the underlying energy consumption in region B, given the tight coupling between BOLD signal changes and glucose metabolism.”

We have also significantly extended the Discussion on the underlying physiological mechanisms of BOLD and [¹⁸F]FDG PET imaging (response to main issue 11A) and added the following text:

“The relationship between blood flow and metabolic factors has been demonstrated to be in balance at rest, whereas task-induced neural activations induce a mismatch (Gusnard and Raichle, 2001; Raichle et al., 2001). […] Our results extend this knowledge by highlighting that external task engagement is accompanied by the convergence between metabolic factors and functional brain network reconfigurations.”

3) Given general issues in the field with interpreting directed connectivity (e.g., Ramsey et al., 2010; Smith et al., 2011), validation with DCM seems like an insufficient standard. The authors should be explicit about the pitfalls of both approaches (DCM and MCM) and whether they cover each other's pitfalls. For example, DCM is probably sensitive to missing nodes, while MCM won't have that issue (as far as we can tell), so the two approaches nicely complement each other in this respect. Are there pitfalls that apply to both DCM and MCM? If so, how serious are these pitfalls and can they be addressed with control analyses?

Considering that MCM and DCM use different underlying data and employ distinct models, the two approaches indeed complement each other. As such, we did not attempt to use DCM as a validation tool, but rather to assess if a comparable network structure was obtained by two independent, yet complementary approaches. As suggested by the reviewers we have addressed this by adding the following to the Discussion:

“Estimating measures of directed functional connectivity is a considerable challenge, particularly when those measures depend purely upon conditional likelihoods in the data (Smith et al., 2011), or in networks with closed cycles (Ramsey et al., 2010). […] By using a generative model and an observation (HRF) equation, DCM also avoids reliance upon making direct inference on conditional dependences in data, a challenge for many measures of directed functional connectivity (Smith et al., 2011).”

4) Given the large amount of spatial smoothing that was performed across all modalities (8 mm), how informative is a spatial correlation calculated within ROIs for MCM? One way to test for this more robustly is to perform permutation testing by comparing a BOLD connectivity map from one subject to a glucose metabolism map of a different subject (and build a null distribution through many such permutations), rather than performing a basic t-test against zero (and similarly for the comparison against rest). This will determine whether spatial similarity is truly indicative of connectivity-metabolic coupling, or whether it is observed as a function of the large smoothing kernel (and relatively small ROIs).

It is in principle possible that spatial correlations induced by smoothing may have biased MCM estimates. We directly tested this hypothesis by assessing if an original correlation of zero was artificially inflated by different levels of smoothing. Compared to the suggested permutation approach, the adopted test unequivocally addresses the potential problem. We have added the following to demonstrate that smoothing does not significantly impact the MCM findings:

Materials and methods:

“To assess the influence of spatial smoothing on MCM estimates, a control analysis was carried out testing whether a spatial correlation of zero is artificially inflated by smoothing. […] To generate a null distribution representing the influence of smoothing in these finite sampled data, this procedure was repeated 500 times for each of the 22 subjects.”

Results:

“A control analysis was performed to assess the impact of spatial smoothing on the MCM estimates (Figure 3—figure supplement 3). […] This suggests that smoothing does not systematically inflate the MCM estimates, which is also in line with our observation that half of the MCM values were approximately zero despite using smoothed data (Figure 3).”

5) Why were DCM ROIs different from the MCM ROIs (Figure 4—figure supplement 1)? If DCM is used for validation, it seems to make sense to use the same regions in both analyses. Furthermore, how does the interaction between the size of the ROIs (5 mm) and the smoothing kernel (8mm) affect the results?

As suggested by the reviewers we have re-calculated the DCM with a ROIs size of 8 mm (i.e., matching the smoothing kernel), yielding similar results. We also added an explanation about the different ROI definition for the two approaches.

BOLD methods:

“[…] the data were normalized to MNI-space and smoothed with an 8 mm Gaussian kernel to match the kernel used for the PET and ASL data.”

DCM methods:

“The BOLD signal was extracted as the eigenvariate within a sphere of 5 mm radius, adjusted for the effect of interest. […] Thus, the region employed for DCM represents a subregion of that used for MCM, with a focus on the individual location of task-specific changes in the BOLD signal.”

DCM results:

“Notably, the DCM results did not change when using a larger ROI size of 8 mm (Figure 4—figure supplement 3).”

Results:

“To allow for a focus on cross-modal changes, the regions included in MCM and DCM analyses were Occ, FEF and IPS as obtained from the conjunction analysis (Figure 2). […] The average distance between any voxel of one region to the closest voxel of another region was 46 ± 23 mm, mitigating potential cross-talk between the regions introduced through spatial smoothing.”

6) Some of the statistical tests are unclear. The authors wrote that "the significance of each connection was tested separately using one-sample t-tests against zero". So let's consider one connection: FEF to IPS. The MCM analysis yields one spatial correlation for each subject. So can I confirm the t-test involves 22 numbers (corresponding to the 22 subjects)? Because a p value of 10e-12 is quite impressive with only 22 subjects; one would need a t-score of around 14?

Correct, the t-tests of the MCM against zero comprised 22 numbers, one for each subject. The t-value of the connection FEF to IPS (hard condition) was indeed t = 14.4 and the corresponding p-value was p_uncorrected = 2*10^-12 and p_Bonf-Holm = 4*10^-11.

To address the comment, we have also provided the data used to generate Figure 3, enabling the reader to assess the validity of the statistical tests. We have also amended the manuscript as follows:

“For the MCM analysis, the significance of each connection was tested separately using one-sample t-tests against zero (6 connections * 3 conditions). As MCM yields one spatial correlation value for each subject, the test for each connection and condition is pooled across 22 correlations, corresponding to n = 22 subjects.”

7) For Figure 3, the authors say that the thickness of arrows depend on r values. More details will be useful. For example, when looking at Occ -> FEF, the two box plots for easy and hard conditions look highly overlapping, but the green arrow was much thicker in the hard condition than easy condition. And from what I gather from the plot, the IPS->FEF during hard condition has higher MCM correlation than Occ->FEF during hard condition. So why is the green arrow thickness than the yellow arrow during the hard condition?

We have now revised the figure as suggested. Furthermore, r-values were changed to z-scores. We also changed boxplots to dot plots with lines connecting the individual subjects to better reflect the within-subject nature of the design and the statistics.

8) It would be important to report the pure functional connectivity among the 3 regions. More specifically, are there region pairs with strong functional connectivity, but the MCM connectivity is weak in both directions. If so, how can this be interpreted? Similarly, are there regions with weak functional connectivity but the MCM connectivity is strong? If so, how can this be interpreted?

We have now also computed the pure functional connectivity between the three regions. The results are now included in the manuscript as follows.

Materials and methods:

“For MCM, functional connectivity was computed as seed-to-voxel correlations between two brain regions followed by z-transformation. […] For direct comparison between MCM and functional connectivity, we also report the functional connectivity as the Pearson’s correlation between time courses of the three brain regions.”

Results:

“Notably, functional connectivity was also sensitive to task performance with significant increases from rest to task for all three connections (Figure 3—figure supplement 1). Taking into account the underlying metabolic demands using MCM allowed us to infer directionality on these task-induced changes in functional connectivity, i.e., the influence one region exerts on another (Occ -> IPS, Occ -> FEF), as well as a separation of task effects (FEF -> IPS) from intrinsic connectivity (IPS -> FEF).”

9) The DCM modeling needs to be further elaborated. Maybe the current details are sufficient for a practiced DCM user, but as someone who is somewhat familiar with DCM (but do not actively use it), I found it hard to know how task modulation and intrinsic connection were modeled in the generative model.

As requested, we have improved the description of the DCM approach:

“Dynamic causal modelling (DCM): The directed functional connectivity model obtained from the MCM analysis was benchmarked against directed effective connectivity inferred from DCM. […] Each effective connection was also individually tested for the presence of task modulation using model family inference (i.e., all models comprising task modulation of a given connection against those without).”

10) "A putative mediation effect of FEF was suggested by the observation that Occ -> IPS MCM values decreased in the easy condition (from 0.17 to 0.14, pBonf-Holm = 0.017) when controlling for the indirect pathway (Occ -> FEF -> IPS)." – Please elaborate on how this is done.

We apologize for the missing information and have added the following details to the Materials and methods section:

“We tested for a potential mediation effect of FEF on the connection Occ -> IPS by adapting an established statistical approach (Baron and Kenny, 1986) to the MCM framework. […] The resulting MCM values were then compared to the original MCM values of Occ -> IPS using a paired t-test.”

11) Literature review and discussion.

A) Given the nature of the paper, the authors should include a discussion of how their findings compare to a large and long-standing literature on cerebral blood flow and metabolism, which explicitly contrasts cerebral blood flow, oxygen consumption, and glucose metabolism to one another in and outside of tasks contexts, e.g., from Marc Raichle and others (see for example: Gusnard and Raichle., 2001). The current work would have substantially greater impact if it were to engage more deeply with this older literature. For example, do the authors mean to suggest (as might appear from their statement in the Abstract that "dynamic regulation of neural-metabolic coupling is essential to support human cognition") that different mechanisms exist to couple neural activity with glucose consumption depending on the task at hand? This claim seems quite large and far reaching given the evidence, and if true, would have profound implications in invalidating the current use and interpretation of many neuroimaging methods.

We agree that the paper could have been more clearly contextualized and now include a more comprehensive discussion of the physiological mechanisms underpinning CBF, oxygen consumption, and glucose metabolism (see also, our revisions in response to main issue 2, above). Notably, we did not mean to imply that the physiological mechanisms coupling neural activity to glucose metabolism (and BOLD) depend on the task at hand. As rightly pointed out, this claim goes well beyond what we can rightly infer from our data. Rather, the time-dependent, multi-step and nonlinear nature of these relationships become more apparent in the setting of a dynamic task of increasing complexity, as does the role of varying signal and noise on the statistical effects evident across different imaging modalities.

We revised the Abstract:

“The ability to solve cognitive tasks depends upon adaptive changes in the organization of whole-brain functional networks. […] Our findings show how the upregulation of key computational mechanisms to support cognitive performance unveils the complex, interdependent changes in neural metabolism and neuro-vascular responses.”

… and the impact statement:

“Cognitive performance is supported by symbiotic metabolic and neuro-vascular responses in task-specific brain networks.”

We have significantly extended the Discussion on the underlying physiological mechanisms of BOLD and [¹⁸F]FDG PET imaging and included the following text (see also our response to main issue 2):

“The relationship between blood flow and metabolic factors has been demonstrated to be in balance at rest, whereas task-induced neural activations induce a mismatch (Gusnard and Raichle, 2001; Raichle et al., 2001). […] Our results extend this knowledge by highlighting that external task engagement is accompanied by the convergence between metabolic factors and functional brain network reconfigurations.”

B) The authors should also be more precise about differences with previous work (e.g., Jamadar et al., 2019; Rischka et al., 2018). For example, "Here we extend upon previous work by providing a comprehensive assessment of the neural and metabolic processes supporting performance in a complex cognitive visuo-motor task" can be briefly expanded to mention that Rischka investigated a finger-tapping task and Jamadar investigated a visual checkerboard task.

As requested, we have added this information:

“To date, only a few studies using [¹⁸F]FDG fPET imaging have assessed the possible functional relationships between these diverse parameters of brain activity (Horwitz et al., 1995). […]. Moreover, our results directly support previous findings suggesting that regional neural and metabolic changes converge in spatially circumscribed and functionally specialized brain regions (Jamadar et al., 2019; Riedl et al., 2014; Rischka et al., 2018).”

12) The Discussion states: "Our results extend this knowledge by highlighting that the convergence between metabolic factors and functional network reconfiguration is pivotal to support cognition". However, the relationship with cognition or performance was not tested here (e.g. MCM was not shown to be stronger for those subjects who performed better). As such, I would recommend adjusting this interpretation.

This statement has been amended as follows: “Our results extend this knowledge by highlighting that external task engagement is accompanied by the convergence between metabolic factors and functional brain network reconfigurations.”

Author details

1. Andreas Hahn

   Department of Psychiatry and Psychotherapy, Medical University of Vienna, Vienna, Austria

   Contribution

   Conceptualization, Data curation, Formal analysis, Supervision, Funding acquisition, Investigation, Visualization, Methodology, Project administration

   For correspondence

   andreas.hahn@meduniwien.ac.at

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0001-9727-7580

2. Michael Breakspear

   1. QIMR Berghofer Medical Research Institute, Brisbane, Australia
   2. School of Psychology, University of Newcastle, Newcastle, Australia

   Contribution

   Conceptualization, Supervision, Investigation, Methodology

   Competing interests

   No competing interests declared

   "This ORCID iD identifies the author of this article:" 0000-0003-4943-3969

3. Lucas Rischka

   Department of Psychiatry and Psychotherapy, Medical University of Vienna, Vienna, Austria

   Contribution

   Data curation, Formal analysis, Methodology

   Competing interests

   No competing interests declared

4. Wolfgang Wadsak

   1. Department of Biomedical Imaging and Image-guided Therapy, Division of Nuclear Medicine, Medical University of Vienna, Vienna, Austria
   2. Center for Biomarker Research in Medicine (CBmed), Graz, Austria

   Contribution

   Conceptualization, Data curation, Supervision

   Competing interests

   WW declares to having received speaker honoraria from the GE Healthcare and research grants from Ipsen Pharma, Eckert-Ziegler AG, Scintomics, and ITG; and working as a part time employee of CBmed Ltd. (Center for Biomarker Research in Medicine, Graz, Austria).

5. Godber M Godbersen

   Department of Psychiatry and Psychotherapy, Medical University of Vienna, Vienna, Austria

   Contribution

   Data curation

   Competing interests

   No competing interests declared

6. Verena Pichler

   Department of Biomedical Imaging and Image-guided Therapy, Division of Nuclear Medicine, Medical University of Vienna, Vienna, Austria

   Contribution

   Data curation

   Competing interests

   No competing interests declared

7. Paul Michenthaler

   Department of Psychiatry and Psychotherapy, Medical University of Vienna, Vienna, Austria

   Contribution

   Data curation

   Competing interests

   No competing interests declared

8. Thomas Vanicek

   Department of Psychiatry and Psychotherapy, Medical University of Vienna, Vienna, Austria

   Contribution

   Data curation

   Competing interests

   No competing interests declared

9. Marcus Hacker

   Department of Biomedical Imaging and Image-guided Therapy, Division of Nuclear Medicine, Medical University of Vienna, Vienna, Austria

   Contribution

   Conceptualization, Supervision

   Competing interests

   MH received consulting fees and/or honoraria from Bayer Healthcare BMS, Eli Lilly, EZAG, GE Healthcare, Ipsen, ITM, Janssen, Roche, and Siemens Healthineers.

10. Siegfried Kasper

    Department of Psychiatry and Psychotherapy, Medical University of Vienna, Vienna, Austria

    Contribution

    Conceptualization, Supervision

    Competing interests

    SK received grants/research support, consulting fees and/or honoraria within the last three years from Angelini, AOP Orphan Pharmaceuticals AG, Celgene GmbH, Eli Lilly, Janssen-Cilag Pharma GmbH, KRKA-Pharma, Lundbeck A/S, Mundipharma, Neuraxpharm, Pfizer, Sage, Sanofi, Schwabe, Servier, Shire, Sumitomo Dainippon Pharma Co. Ltd., Sun Pharmaceutical Industries Ltd. and Takeda.

11. Rupert Lanzenberger

    Department of Psychiatry and Psychotherapy, Medical University of Vienna, Vienna, Austria

    Contribution

    Conceptualization, Supervision

    Competing interests

    RL received conference speaker honorarium within the last three years from Shire and support from Siemens Healthcare regarding clinical research using PET/MR. He is shareholder of BM Health GmbH since 2019.

12. Luca Cocchi

    QIMR Berghofer Medical Research Institute, Brisbane, Australia

    Contribution

    Conceptualization, Formal analysis, Supervision, Investigation, Visualization, Methodology

    Competing interests

    No competing interests declared

    "This ORCID iD identifies the author of this article:" 0000-0003-3651-2676

• 2,236

  Page views

• 284

  Downloads

• 26

  Citations

Article citation count generated by polling the highest count across the following sources: Scopus, Crossref, PubMed Central.