C01
Improvements in cognition during ageing require neural resource mobilisation through effective training interventions. We hypothesize that neural resource mobilisation can be optimized by adapting the task demands relative to the current ability. We refer to the gap between demand and ability (functional task difficulty) as the Ability Prediction Error (APE). Using computational modelling and longitudinal quantitative MRI (qMRI), we aim to explore how training using experimentally manipulated APEs (underload, optimal load, overload) control sensorimotor performance improvements, behavioural transfer and neural resource mobilization at macro- and micro-structural brain levels in a defined (pre-) frontal brain circuit.
Principal Investigators
Prof. Dr. Marco Taubert
Dr. Gabriel Ziegler
Co-Workers
Norman Aye
Dr. Nico Lehmann
Martin Matke
Nisha Prabhu
Expansion-renormalisation model of plasticity
According to the expansion-renormalisation model, learning-induced structural brain changes often follow a sequence of expansion, selection and renormalisation. Human learning may involve an initial but transient phase of increase in grey matter volume followed by a partial or even complete return to baseline once a behaviourally optimal neural circuitry has been selected (Wenger et al., 2017). The expansion-renormalisation model assumes that transfer effects of training to other related tasks are most likely at or just before volume expansion has reached its peak, motivating closer inspection of the temporal characteristics of plasticity and its relationship to CRC-defined neural resource properties. Depending on the particular phase of plasticity (i.e. grey matter expansion, grey matter renormalisation), the engaged neural circuit may be more or less susceptible to positive or negative transfer to similar learning tasks (Richards & Frankland, 2017).
The match of task demands and ability
It has been hypothesised that training intensity, defined as the mismatch (or gap) between an individual’s current ability and the requested task demand (Lövdén et al., 2010; Lindenberger, 2016), influences the trajectories of brain plasticity and its associated behavioural capacities (specific task-related performance improvements and its potential transfer). It is therefore expected that task demands need to be within the range of functional capacity (underload < optimum > overload) in order to optimally trigger plasticity (Lövdén et al., 2010). Model-based prediction errors are increasingly used in computational neuroscience, reinforcement learning and modelling trial-by-trial decision processes (Dayan, 2005). In this project, we aim to test the above hypothesis by experimentally manipulating Ability Prediction Errors (APEs), i.e. the quantitative mismatch of the required task demands and the subject’s current ability (estimate) in the learning process. APEs will play a central role in our development of a computational model of neural resource mobilisation that predicts multimodal structural MRI patterns of training-induced plasticity in a frontal cortex network.
Quantitative MRI
Quantitative magnetic resonance imaging (qMRI) is a promising tool to study training-induced plasticity in humans due to its sensitivity to (often called “micro-structural”) brain tissue properties such as axon, myelin, and iron at millimetre resolution (Weiskopf et al., 2015, Tardif et al., 2016, Tabelow et al., 2019, Ziegler et al., 2018, 2019). In this particular line of research, “micro-structural” refers to quantitative aspects of the brain tissue (e.g. myelin density per volume) and is independent of the actual image resolution which is used to map it spatially (e.g. meso- or macroscale in this CRC). qMRI goes beyond traditional brain morphometry characterising volume differences or tissue expansion which is traditionally referred to as “macro-structural”. Longitudinal qMRI training studies which aim to analyse micro-structural contributions to macro-structural expansion and renormalisation cycles are still lacking.
The goals of our project
This project develops a computational brain-behavioural model for frontal neurocognitive circuit reorganisation serving performance improvements to match demands of a novel task. We assess brain changes during training that integrates macro- and multiple (micro-) structural imaging modalities (MT, R2* & DTI/NODDI). We study a CRC cohort (Z03-Düzel/Maass/Kreißl) subsample of 60 older participants’ learning trajectories using dynamical system modelling and hierarchical Bayesian inference. We hypothesize that experimentally manipulated APEs predict specific brain-behavioural adaptations. The dynamical model parameters represent (1) the ‘update rules’ (or dynamics) of week-wise individual performance and the (2) ‘microstructural cortical updates’ underlying these performance improvements during learning. Each participant acquiring the sensorimotor ability has an initial brain-behavioural starting point (a given raw ability/resource), adapts according to mechanisms via manipulated APEs and ends at a final state reflecting individual limitations of plasticity. We finaally study the individual differences and the contribution of hidden pathological conditions (such as amyloid) to training-induced brain and performance changes.
Imaging pilot study
In a test-retest reliability pilot study, we assessed multi-parameter maps (MPM) for qMRI as well as diffusion-tensor imaging (DTI) plus neurite orientation dispersion and density imaging (NODDI) parameters for N=30 subjects at two measurement occasions one month apart and derived intraclass correlation coefficients (ICC). Specifically, voxel-based reliability for advanced imaging markers (MPMs and NODDI parameters) was promising (Lehmann et al., 2021, Aye et al., under review), supporting the application of this imaging protocol to map frontal circuit changes under training in the proposed project.
Behavioral piloting
We aim to extend previous balance training paradigm (Sehm et al., 2014) and develop an ability testing protocol for APE manipulation during training. First, we set the stage towards a fine-grained assessment of the subject’s current ability during each learning session using mechanical sensors. Task difficulty is then adaptively regulated via additional board weights across training and clamped to manipulate functional task difficulty according to each participant’s current balance ability from a testing session. Using this stabilometer paradigm and response-optimisation strategy, we will manipulate task difficulty over six training sessions and across individuals (N = 30 older participants between 60 – 75 years of age). We assess motor task improvements and transfer task performance using neuropsychological testing.
Longitudinal neuroimaging study
We will perform an intervention study with 60 healthy older participants (60-75 years of age) from CRC’s central cohort (Z03). Participants will be randomly allocated in main experimental groups with response-optimized vs. suboptimal APEs during training. Recent developments of MRI technology allow us to go beyond morphometric assessments to realize Multi-Parameter Mapping (MPM) of (MT, R1, R2*, PD) and combine the strength of qMRI with advanced diffusion MRI (NODDI). We are involved in the software development of state-of-the-art processing routines for quantitative imaging data in an international team (www.hmri.info; see Tabelow et al., 2019). Based on 6 weekly observations we will test for neural resource mobilisation in a frontal brain network comprising primary motor cortex (Taubert et al., 2016), pre-SMA/SMA, dorsolateral prefrontal cortex, anterior prefrontal cortex and hippocampus as well as adjacent white matter fibre tracts (Taubert et al., 2010, 2011; Sehm et al., 2014; Lehmann et al., 2019).
Computational model of brain-behavioral plasticity
We will extend previous work on dynamical modelling (Ziegler et al. 2017, Johnson/Ziegler et al.. 2021) to build a quantitative model of macro- and microstructural brain plasticity induced by learning the new sensorimotor (balance) task. We will focus on a hypothesised neurocognitive circuit (regions of interest) including the motor, pre-motor and anterior prefrontal cortex areas. A dynamical system with (macro- and microstructural states) captures multipametric MRI measurement and parameters of interest are estimated on a single subject and group level using Bayesian inference. Evidence-based model comparison will be performed to compare the dynamic model to more traditional polynomial models. The training-induced non-linear brain changes and cross-modal interactions of different processes are captured by model parameters, allowing us to study the interplay of myelination/iron changes and volume expansion-renormalisation during training for the first time.
Long-term perspective
In the long run, the project aims at (1) developing a behavioural intervention strategy to promote neural resource mobilisation at an optimal cost-benefit relationship (costs = training time; benefits = mobilised neural resources of cognition). Furthermore, the project develops (2) an increased understanding of the key determinants underlying structural brain plasticity studied with state-of-the-art neuroimaging. Our research agenda focusses on overcoming the descriptive nature of previous research on training-induced plasticity. In contrast to previous work where brain changes were induced and ‘just’ observed, we will follow the vision of building a new generation of generative models that incorporate a quantitative representation of the hypothesised drivers of neural changes on various levels, e.g. the behavioural deficit causing tissue property changes (resulting from a demand-ability mismatch during an interaction with the environment).