LIFESPAN NEURAL DYNAMICS GROUP

Early theory on aging, and in particular, the age-based degradation of cognition, is historically ascribed to some form of heightened “neural noise” that interferes with efficient processing (e.g., Crossman & Szafran, 1956; Cremer and Zeef, 1987; Welford, 1965, 1981). However, these early theories had rarely been tested by examining within-subject brain signals directly. Perhaps counterintuitively, our extant in vivo work consistently shows that older, poorer performers may generally exhibit less moment-to-moment brain signal variability than younger, higher performing adults (e.g., Garrett et al., 2010, 2011, JNeurosci; 2013 Cerebral Cortex; 2013 NBR; 2015 PNAS). Overall, our group seeks to understand why healthy, higher performing brains are highly and robustly dynamic across moments (adaptability, dynamic range, multi-stability?), and how this changes with development and aging.

Our interests in the relations between cognition and brain signal variability have been broad to date, spanning simple and complex perceptual decision-making (e.g., Kloosterman et al., 2019, eLife), internal vs. externally modulated states (e.g., Grady and Garrett, 2018, NeuroImage), working (e.g., Garrett et al., 2015, PNAS) and long-term memory (Kloosterman et al., in prep), and reinforcement learning (Skowron et al., in prep) using a variety of stimulus types, cognitive performance metrics, and model parameterizations. Broadly, we find that better performers have higher brain signal variability, and that within-person signal variability can parametrically tune with changes in task difficulty (Garrett et al., 2013a,b, Cerebral Cortex; 2013, NBR; 2015 PNAS). We also continue to examine whether variability may be an expression of the flexible modulation of brain states that allows a probabilistic search for an optimal state, given either unexpected or anticipated changes in the environment (see Garrett et al., 2013, NBR).

It is well known that functional dynamics occur around a static anatomical (white matter) skeleton (Honey et al., 2009). Computational models suggest that realistic network dynamics, and the influence of noise within them, cannot exist without a healthy, biologically plausible WM structure (Deco et al., 2009). However, it remains unknown to what extent the WM skeleton constrains functional dynamics/variability in humans, and across the lifespan. Conversely, although structural connections form the “core skeleton” of brain function, functional connections also exist broadly in absence of direct structural connections (Deco et al., 2009; Ghosh et al., 2008; Honey et al., 2007). Notably, a theoretically and computationally informed contributing factor to brain signal variability effects is that our neural system can explore a broad repertoire of brain states/networks from moment to moment when variability is present (Deco et al., 2009; 2011; Garrett et al., 2013, NBR). Accordingly, a healthy, developed brain (i.e., in young adults) may be more flexible in the face of changing systemic or environmental demands, as it can reconfigure efficiently as required. Assuming that node-based signal variability reflects a dynamic system that is constantly reconfiguring in strength and direction, it is clear that explicitly measuring network-level dynamics is required to determine the bounds of node-based dynamics. We are examining these issues in detail within the LNDG. A recent attempt to link node-wise temporal dynamics to network-level effects (Garrett et al., 2018, NeuroImage) revealed that the vast majority of local temporal variability may be a reflection of network communication/integration rather than local “noise.” Crucially, the thalamus provided a unique and fundamental variability-based signature of how the brain integrates across scales, providing a key anatomical target for future LNDG work probing the generation and mechanisms of temporal variability in the brain.

We are interested in linking theories of dopamine (DA)-based neural “precision” (e.g., Ballo et al., 2012; Coull et al., 2012; Friston, 2010) and dynamics to brain signal variability effects. It is an interesting paradox that signal variability often appears functional for healthy neural systems, yet optimal DA function is presumed to generally invoke tighter bounds on temporal coding. We are currently collaborating with various research groups in Europe to investigate how DA agonists, antagonists, and PET-based DA binding potential may relate to working memory- and reinforcement-learning-based brain signal variability in young and older adults. We find that DA agonism via amphetamine increases BOLD variability in older adults (Garrett et al., 2015, PNAS; see fig above), DA antagonism decreases BOLD variability in younger adults (Garrett et al., in prep), and DA binding (using PET imaging) correlates with BOLD variability in a regionally-dependent manner (Guitart-Masip et al., 2016, Cerebral Cortex; Garrett et al., in prep). DA thus continues to be a primary neurochemical target within the LNDG, with our interests recently expanding to include the norepinephrine (e.g., systemic explore/exploit trade-offs) and glutamate/GABA systems (e.g., state-configuration via excitation/inhibition).

Our initial studies in this domain focus on directly invoking stochastic resonance, which is classically observed when the presence of additive noise allows input signals to be better detected in the brain. Historically, “noise” has been experimentally applied to the nervous system in a number of ways via noisy peripheral visual/tactile/auditory stimuli (e.g., Collins et al., 1996; Wells et al., 2005). However, good evidence for whether/how these peripherally applied “noise” sources reach the brain has been lacking. Alternatively, transcranial noise stimulation methods (tRNS) provide a more direct route to the cortex of the brain, thus piquing our interest in simultaneous tRNS-neuroimaging to examine links between stimulation and brain signal dynamics. Overall, we seek to use tRNS as a “causal” technique for restoring impoverished signal variability levels in older adults.

Measures

We continue to both develop and apply brain signal variability methods, moving beyond general variance measures to ask more explicit questions about the nature of that variance (e.g., spectral content and structure, time delay embedding). We are also inherently interested in discrete patterns in brain signals (e.g., entropy), and how they relate to overall variance (*Grandy, *Garrett et al., 2016). For example, we continue to optimize the estimation of multi-scale entropy that effectively eliminates variance-based biases from computation (Kosciessa et al., in prep; Kloosterman et al., in prep), permitting entropy to be directly compared with oscillatory dynamics within-person (e.g., Kosciessa et al., 2019).

Statistical and computational modeling

We have a keen interest in multivariate models (e.g., singular value decomposition (SVD)-based techniques) that best enable various correlates of signal dynamics to be simultaneously examined. We are currently exploring how SVD models can be optimized within a mixed-modeling framework in an attempt to link latent-level parametric cognitive performance to parametric task-based brain dynamics (e.g., Garrett et al., 2013 Cereb Cortex; Garrett et al., 2015 PNAS). We also continue to develop whole-to-whole brain multivariate models linking different neuroimaging modalities (e.g., ASL and BOLD; dopamine PET and BOLD; EEG and BOLD; DWI and BOLD; e.g., Garrett et al., 2017, Sci Rep; *Burzynska, *Garrett, et al., 2013, JNeurosci).

We also utilize and develop various computational models of brain and behaviour. For example, we have recently leveraged the HMAX model of the ventral visual cortex to estimate how the complexity of visual input may drive neural variability during perception (Garrett et al., 2018). With regard to behaviour, we have also modelled within-person, reward-driven perceptual decision-making via hierarchical drift diffusion modelling (Kloosterman et al., 2019, eLife), and now deploy various reinforcement-learning models across a host of probabilistic reward-based paradigms.

Software Development

An important goal of our proposed work is to support the future examination of brain signal variability in neuroimaging through open-source software. At present, no analysis package provides for a comprehensive brain signal variability analysis. As such, we continue to develop an event-related and block-design fMRI-based Variability Toolbox (VarTbx) for SPM. Mimicking a Level-1 analysis, our toolbox allows SPM users (or more generally, Matlab users) to import any preprocessed fMRI data, choose from a number of different variability measures, and to output resulting individual maps for further analysis within SPM’s standard Level-2 module (or elsewhere). The first version of VarTbx is now available; see our software page for details. We further support open science by making all code/algorithms freely available on our LNDG Github repository (https://github.com/LNDG).

Postdoctoral positions

We continue to recruit excellent postdoctoral fellows, with no fixed deadline for applications or expression of interest.

Ph.D. positions

Since 2016, an International Max Planck Research School (Ph.D. level) on computational methods (IMPRS COMP2PSYCH) has been funded, in association with the Max Planck UCL Centre for Computational Psychiatry and Ageing Research. The annual application deadline for positions is typically in late Winter each year.

Student assistant/Master's thesis student positions

Several student assistant positions may be available in LNDG at any one time, depending on the studies we are currently running. Many student assistants also formulate and complete their master's thesis while employed.

Internships

Three-month paid internships are also possible within LNDG. There is no specific structure to such an internship, but is typified by students acquiring a specific technique and/or helping with a specific study.

Regarding all available positions, please email us directly at lndg -at- mpib-berlin.mpg.de.