Learning induces coordinated neuronal plasticity of metabolic demands and functional brain networks

Competitive plasticity to reduce the energetic costs of learning

The brain is not only constrained by energy needed to fuel computation, but it is also constrained by energy needed to form memories. Experiments have shown that learning simple conditioning tasks which might require only a few synaptic updates, already carries a significant metabolic cost. Yet, learning a task like MNIST to 95% accuracy appears to require at least 108 synaptic updates. Therefore the brain has likely evolved to be able to learn using as little energy as possible. We explored the energy required for learning in feedforward neural networks. Based on a parsimonious energy model, we propose two plasticity restricting algorithms that save energy: 1) only modify synapses with large updates, and 2) restrict plasticity to subsets of synapses that form a path through the network. In biology networks are often much larger than the task requires, yet vanilla backprop prescribes to update all synapses. In particular in this case, large savings can be achieved while only incurring a slightly worse learning time. Thus competitively restricting plasticity helps to save metabolic energy associated to synaptic plasticity. The results might lead to a better understanding of biological plasticity and a better match between artificial and biological learning. Moreover, the algorithms might benefit hardware because also electronic memory storage is energetically costly.

Synaptic signaling modeled by functional connectivity predicts metabolic demands of the human brain

PURPOSE

The human brain is characterized by interacting large-scale functional networks fueled by glucose metabolism. Since former studies could not sufficiently clarify how these functional connections shape glucose metabolism, we aimed to provide a neurophysiologically-based approach.

METHODS

51 healthy volunteers underwent simultaneous PET/MRI to obtain BOLD functional connectivity and [18F]FDG glucose metabolism. These multimodal imaging proxies of fMRI and PET were combined in a whole-brain extension of metabolic connectivity mapping. Specifically, functional connectivity of all brain regions were used as input to explain glucose metabolism of a given target region. This enabled the modeling of postsynaptic energy demands by incoming signals from distinct brain regions.

RESULTS

Functional connectivity input explained a substantial part of metabolic demands but with pronounced regional variations (34 - 76%). During cognitive task performance this multimodal association revealed a shift to higher network integration compared to resting state. In healthy aging, a dedifferentiation (decreased segregated/modular structure of the brain) of brain networks during rest was observed. Furthermore, by including data from mRNA maps, [11C]UCB-J synaptic density and aerobic glycolysis (oxygen-to-glucose index from PET data), we show that whole-brain functional input reflects non-oxidative, on-demand metabolism of synaptic signaling. The metabolically-derived directionality of functional inputs further marked them as top-down predictions. In addition, the approach uncovered formerly hidden networks with superior efficiency through metabolically informed network partitioning.

CONCLUSIONS

Applying multimodal imaging, we decipher a crucial part of the metabolic and neurophysiological basis of functional connections in the brain as interregional on-demand synaptic signaling fueled by anaerobic metabolism. The observed task- and age-related effects indicate promising future applications to characterize human brain function and clinical alterations.

Validation of cardiac image-derived input functions for functional PET quantification

PURPOSE

Functional PET (fPET) is a novel technique for studying dynamic changes in brain metabolism and neurotransmitter signaling. Accurate quantification of fPET relies on measuring the arterial input function (AIF), traditionally achieved through invasive arterial blood sampling. While non-invasive image-derived input functions (IDIF) offer an alternative, they suffer from limited spatial resolution and field of view. To overcome these issues, we developed and validated a scan protocol for brain fPET utilizing cardiac IDIF, aiming to mitigate known IDIF limitations.

METHODS

Twenty healthy individuals underwent fPET/MR scans using [18F]FDG or 6-[18F]FDOPA, utilizing bed motion shuttling to capture cardiac IDIF and brain task-induced changes. Arterial and venous blood sampling was used to validate IDIFs. Participants performed a monetary incentive delay task. IDIFs from various blood pools and composites estimated from a linear fit over all IDIF blood pools (3VOI) and further supplemented with venous blood samples (3VOIVB) were compared to the AIF. Quantitative task-specific images from both tracers were compared to assess the performance of each input function to the gold standard.

RESULTS

For both radiotracer cohorts, moderate to high agreement (r: 0.60-0.89) between IDIFs and AIF for both radiotracer cohorts was observed, with further improvement (r: 0.87-0.93) for composite IDIFs (3VOI and 3VOIVB). Both methods showed equivalent quantitative values and high agreement (r: 0.975-0.998) with AIF-derived measurements.

CONCLUSION

Our proposed protocol enables accurate non-invasive estimation of the input function with full quantification of task-specific changes, addressing the limitations of IDIF for brain imaging by sampling larger blood pools over the thorax. These advancements increase applicability to any PET scanner and clinical research setting by reducing experimental complexity and increasing patient comfort.

Neural mechanisms underlying improved new-word learning with high-density transcranial direct current stimulation

Neurobehavioral studies have provided evidence for the effectiveness of anodal tDCS on language production, by stimulation of the left Inferior Frontal Gyrus (IFG) or of left Temporo-Parietal Junction (TPJ). However, tDCS is currently not used in clinical practice outside of trials, because behavioral effects have been inconsistent and underlying neural effects unclear. Here, we propose to elucidate the neural correlates of verb and noun learning and to determine if they can be modulated with anodal high-definition (HD) tDCS stimulation. Thirty-six neurotypical participants were randomly allocated to anodal HD-tDCS over either the left IFG, the left TPJ, or sham stimulation. On day one, participants performed a naming task (pre-test). On day two, participants underwent a new-word learning task with rare nouns and verbs concurrently to HD-tDCS for 20 min. The third day consisted of a post-test of naming performance. EEG was recorded at rest and during naming on each day. Verb learning was significantly facilitated by left IFG stimulation. HD-tDCS over the left IFG enhanced functional connectivity between the left IFG and TPJ and this correlated with improved learning. HD-tDCS over the left TPJ enabled stronger local activation of the stimulated area (as indexed by greater alpha and beta-band power decrease) during naming, but this did not translate into better learning. Thus, tDCS can induce local activation or modulation of network interactions. Only the enhancement of network interactions, but not the increase in local activation, leads to robust improvement of word learning. This emphasizes the need to develop new neuromodulation methods influencing network interactions. Our study suggests that this may be achieved through behavioral activation of one area and concomitant activation of another area with HD-tDCS.

Non-invasive assessment of stimulation-specific changes in cerebral glucose metabolism with functional PET

PURPOSE

Functional positron emission tomography (fPET) with [18F]FDG allows quantification of stimulation-induced changes in glucose metabolism independent of neurovascular coupling. However, the gold standard for quantification requires invasive arterial blood sampling, limiting its widespread use. Here, we introduce a novel fPET method without the need for an input function.

METHODS

We validated the approach using two datasets (DS). For DS1, 52 volunteers (23.2 ± 3.3 years, 24 females) performed Tetris® during a [18F]FDG fPET scan (bolus + constant infusion). For DS2, 18 participants (24.2 ± 4.3 years, 8 females) performed an eyes-open/finger tapping task (constant infusion). Task-specific changes in metabolism were assessed with the general linear model (GLM) and cerebral metabolic rate of glucose (CMRGlu) was quantified with the Patlak plot as reference. We then estimated simplified outcome parameters, including GLM beta values and percent signal change (%SC), and compared them, region and whole-brain-wise.

RESULTS

We observed higher agreement with the reference for DS1 than DS2. Both DS resulted in strong correlations between regional task-specific beta estimates and CMRGlu (r = 0.763…0.912). %SC of beta values exhibited strong agreement with %SC of CMRGlu (r = 0.909…0.999). Average activation maps showed a high spatial similarity between CMRGlu and beta estimates (Dice = 0.870…0.979) as well as %SC (Dice = 0.932…0.997), respectively.

CONCLUSION

The non-invasive method reliably estimates task-specific changes in glucose metabolism without blood sampling. This streamlines fPET, albeit with the trade-off of being unable to quantify baseline metabolism. The simplification enhances its applicability in research and clinical settings.

High-temporal resolution functional PET/MRI reveals coupling between human metabolic and hemodynamic brain response

PURPOSE

Positron emission tomography (PET) provides precise molecular information on physiological processes, but its low temporal resolution is a major obstacle. Consequently, we characterized the metabolic response of the human brain to working memory performance using an optimized functional PET (fPET) framework at a temporal resolution of 3 s.

METHODS

Thirty-five healthy volunteers underwent fPET with [18F]FDG bolus plus constant infusion, 19 of those at a hybrid PET/MRI scanner. During the scan, an n-back working memory paradigm was completed. fPET data were reconstructed to 3 s temporal resolution and processed with a novel sliding window filter to increase signal to noise ratio. BOLD fMRI signals were acquired at 2 s.

RESULTS

Consistent with simulated kinetic modeling, we observed a constant increase in the [18F]FDG signal during task execution, followed by a rapid return to baseline after stimulation ceased. These task-specific changes were robustly observed in brain regions involved in working memory processing. The simultaneous acquisition of BOLD fMRI revealed that the temporal coupling between hemodynamic and metabolic signals in the primary motor cortex was related to individual behavioral performance during working memory. Furthermore, task-induced BOLD deactivations in the posteromedial default mode network were accompanied by distinct temporal patterns in glucose metabolism, which were dependent on the metabolic demands of the corresponding task-positive networks.

CONCLUSIONS

In sum, the proposed approach enables the advancement from parallel to truly synchronized investigation of metabolic and hemodynamic responses during cognitive processing. This allows to capture unique information in the temporal domain, which is not accessible to conventional PET imaging.