The scaling gap: from synapses to network neuroscience

A primary challenge for the authors’ framework is the “scaling gap” between microscopic metabolic estimates and macroscale neural signals. Haueis & Colaço’s thesis relies heavily on bottom-up costs, such as the estimated 24,000 ATP/bit required for synaptic transmission. While these values are foundational, it remains unclear how they translate to metabolism measured in the human brain in vivo at the macroscale, such as regional cerebral metabolic rates of glucose (CMR_GLC), which reflect the aggregate activity of millions of neurones (Sokoloff, Reference Sokoloff1960).

There is a critical need to bridge local metabolic signals to the large-scale functional organisation of the brain. Cognition, affect, and behaviour emerge from the coherent activity of spatially distributed populations of neurones (Raichle et al., Reference Raichle, MacLeod, Snyder, Powers, Gusnard and Shulman2001; Yeo et al., Reference Yeo, Krienen, Sepulcre, Sabuncu, Lashkari, Hollinshead, Roffman, Smoller, Zöllei, Polimeni, Fischl, Liu and Buckner2011). This coherent and spatially distinct activity has primarily been modelled as a complex network of units (nodes or brain regions) and their interactions (edges or paths) using fMRI and M/EEG. Graph theory is often used to model how information is transferred throughout the system – the mechanism by which cognition, behaviour, and affect are thought to emerge (van den Heuvel & Sporns, Reference van den Heuvel and Sporns2011). Great strides have been made in mapping the haemodynamic (fMRI) and electromagnetic (M/EEG) characteristics of the functional brain network, but until recently, technical limitations have stymied the capacity to map the metabolic characteristics of this network within individuals (Sala et al., Reference Sala, Lizarraga, Caminiti, Calhoun, Eickhoff, Habeck, Jamadar, Perani, Pereira, Veronese and Yakushev2023). Advances in [¹⁸F]-FDG-PET acquisition and modelling (Hahn et al., Reference Hahn, Gryglewski, Nics, Hienert, Rischka, Vraka, Sigurdardottir, Vanicek, James, Seiger, Kautzky, Silberbauer, Wadsak, Mitterhauser, Hacker, Kasper and Lanzenberger2016; Jamadar et al., Reference Jamadar, Ward, Li, Sforazzini, Baran, Chen and Egan2019; Villien et al., Reference Villien, Wey, Mandeville, Catana, Polimeni, Sander, Zürcher, Chonde, Fowler, Rosen and Hooker2014; Volpi et al., Reference Volpi, Silvestri, Corbetta and Bertoldo2021) now make it possible to characterise the macroscale metabolic network of the brain (Jamadar et al., Reference Jamadar, Zhong, Carey, McIntyre, Ward, Fornito, Premaratne, Shah, O’Brien, Stäb, Chen and Egan2021; Reed et al., Reference Reed, Ponce de León, Vraka, Rausch, Godbersen, Popper, Geist, Komorowski, Nics, Schmidt, Klug, Langsteger, Karanikas, Traub-Weidinger, Hahn, Lanzenberger and Hacker2023; Volpi et al., Reference Volpi, Vallini, Silvestri, Francisci, Durbin, Corbetta, Lee, Vlassenko, Goyal and Bertoldo2023). These methods provide the necessary “bridge” between micro- and macro-scale metabolic function, revealing how local glucose metabolism constrains macroscale organisation and how topological properties, such as network efficiency, directly relate to cognition (Deery et al., Reference Deery, Liang, Siddiqui, Murray, Voigt, Paolo, Moran, Egan and Jamadar2024; Deery et al., Reference Deery, Liang, Moran, Egan and Jamadar2026).

Accounting for spatial and temporal variability in glucose metabolism

Models must further account for the fact that cerebral metabolism is not uniform across space nor time. Glucose is a rate-limiting substrate that sets biophysical limits on when and where neural activity occurs (Jamadar et al., Reference Jamadar, Behler, Deery and Breakspear2025). As a result of these limits, the brain must operate as a highly responsive and scalable system that supports rapid neuronal activity. Glucose metabolism itself fluctuates as a dynamic signal – which we have termed “glucodynamics” – reflecting the time-variant energetic demands of underlying neuronal activity (Deery et al., Reference Deery, Liang, Moran, Egan and Jamadar2026; Hahn et al., Reference Hahn, Reed, Vraka, Godbersen, Klug, Komorowski, Falb, Nics, Traub-Weidinger, Hacker and Lanzenberger2024; Hahn et al., Reference Hahn, Breakspear, Rischka, Wadsak, Godbersen, Pichler, Michenthaler, Vanicek, Hacker, Kasper, Lanzenberger and Cocchi2020; Jamadar et al., Reference Jamadar, Behler, Deery and Breakspear2025). Metabolic rates and glucodynamic signals vary across the cortex, shaped by the underlying cytoarchitecture, gene expression, structural constraints and neural functions (Deery et al., Reference Deery, Moran, Liang, Egan and Jamadar2026; Jamadar et al., Reference Jamadar, Behler, Deery and Breakspear2025). Greater glucodynamic variability and complexity determine the efficiency of information flow across the metabolic network of the brain, enable it to transition between a rich repertoire of network states, and are associated with more adaptive cognition (Deery et al., Reference Deery, Liang, Moran, Egan and Jamadar2025). Cognitive models must therefore integrate these spatial and temporal glucodynamics to understand how the brain dynamically allocates resources to meet shifting internal and external demands.

Scaling information: from bits to complex cognitive processes

Haueis and Colaço define information rate in terms of bits per second: “one bit is the information needed to decide between two equally likely alternatives.” While precise for signalling budgets, to be generative, models must quantify the cost or efficiency of information flow for complex operations, such as prediction error, executive control or memory binding. These complex operations emerge from parallel processing across distributed brain networks (Edelman, Reference Edelman1993; Goldman and Rakic, Reference Goldman-Rakic1988; Mesulam, Reference Mesulam1990) and cannot easily be reduced to their component parts (Miller et al., Reference Miller, Brincat and Roy2024), including bits per information unit processed. To truly understand the energetic demands of cognition, we must find a way to bridge the microscale energetic demands of single-unit firing to the macroscale coordinated dynamics from which complex mental operations emerge. One way to do this could be to develop biophysically informed models of the relationship between microscale and macroscale energetic demands. Such models, which provide a conceptual framework for bridging the micro to macroscale dynamics, exist for fMRI and M/EEG (Friston, Harrison, & Penny, Reference Friston, Harrison and Penny2003). While causal models do not currently exist for FDG-PET data, the field is well-positioned to develop these, given the elegant kinetic models that exist for FDG data (Hahn et al., Reference Hahn, Matej Murgas, Klug, Reed, Godbersen, de Leon, Milz, Hacker, Rujescu and Lanzenberger2025; Reivich et al., Reference Reivich, Alavi, Wolf, Fowler, Russell, Arnett, MacGregor, Shiue, Atkins and Anand1985; Sokoloff et al., Reference Sokoloff, Reivich, Kennedy, Des Rosiers, Patlak, Pettigrew, Sakurada and Shinohara1977). In the meantime, some insight can be gained from non-biophysically informed causal models, such as Granger causality (Deery et al., Reference Deery, Liang, Moran, Egan and Jamada2026). This shifts the focus from abstract energy-per-bit to the relative energetic capacity required to sustain and dynamically reconfigure neural systems within and across cognitive systems.

Energy as an enabler: optimisation, trade-offs, and alterations in health and disease

Haueis and Colaço argue that metabolism should not be treated merely as a background constraint but as a generative principle shaping brain structure and cognitive processes. We agree and emphasise that metabolism should be viewed as an enabler that actively structures large-scale brain network organisation and cognition. The brain is organised as a cost-efficient network that balances the metabolic expense of long-range integration against the economy of local segregation (Bullmore and Sporns, Reference Bullmore and Sporns2012). Evolution prioritises “high-cost, high-return” infrastructure, such as metabolic hubs and long-range connections, to facilitate the global integration required for complex behaviour (Barron, Halina, & Klein, Reference Barron, Halina and Klein2023). This balance is an evolutionary “design choice”: inter-species differences in brain architecture reflect different optimisations of the energy-information trade-off. For instance, axon diameter and myelin thickness affect the conduction velocity of action potentials and the “rate-efficiency” of information transfer and vary across species (Gong et al., Reference Gong, Maffei, Dann, Lee, Lee, Augustinack, Huang, Haber and Yendiki2025; Perge et al., Reference Perge, Koch, Miller, Sterling and Balasubramanian2009). From this perspective, states of altered brain metabolism and metabolic network topology (e.g., in ageing and neurodegeneration) are not merely “energy shortages” but perturbations that impair the brain’s capacity to maintain these optimal, high-cost network states. Energetic cognitive models must be flexible to account for how these design trade-offs are made or break down and to establish states in which altered energetic supply, use, and dynamics lead to diminished network flexibility and poor functional outcomes.