Research – Miriam Klein-Flügge

Knowledge resulting from multiple learning mechanisms

In some of my recent work, I looked at different mechanisms of learning and contrasted their neural mechanisms. Rather than examining what happens during learning, we focused on the knowledge that results from learning. In particular, we contrasted knowledge that results from reinforcement learning, statistical learning and a noncontingent form of learning.

This showed that partially overlapping and partially specialized anatomical regions carried knowledge from these three different learning mechanisms. We also found that the stability versus flexibility of the acquired knowledge depended on how the information was learnt, with RL knowledge leading to more stable but less flexible neural codes compared to statistically learnt knowledge.

Read more in pdf

Decision making and mental health

In ongoing work, I am interested in the changes in learning and decision making present in mental health disorders. There is a large discrepancy between the impairments reported anecdotally by patients and their families, and those observed in laboratory settings.

In a recent review, together with my colleague Dr Jacquie Scholl, I suggested that using more ecological tasks, together with computational models, might help understand impairments. This is what we are doing in ongoing work.

Read more in pdf

In a separate line of work, I am also examining the connections in subcortical circuits to see if specific anatomical pathways might relate to specific problems in mental well-being. This is ongoing work, currently focusing on amygdala and hypothalamus pathways.

Read more in pdf

Transcranial focused ultrasound (deep-brain) stimulation

Many of the regions altered in mood disorders such as depression are buried deep within the brain (including e.g., amygdala, hypothalamus or subgenual/pregenual anterior cingulate cortex). In humans, it has thus far not been possible to directly manipulate these regions using the available neurostimulation techniques. In collaborative work, we have recently shown that transcranial ultrasound stimulation (TUS) can reach deep areas and modulate connectivity and function. This is something I hope to translate to humans in the next few years.

This study published in Nature, led by Alessandro Bongioanni and to which I contributed as a shared senior author, showed TUS can disrupt value formation when value relies on an integration across multiple value dimensions (e.g. reward probability and reward size).

Read more in pdf

In this study I helped with, my colleague Lennart Verhagen showed the effects of TUS on brain connectivity.

Read more in pdf

Decisions involving physical effort

I am interested in decisions that involve effort costs because they are ubiquitous in daily life, impaired in several common disorders (e.g., depression) and much more closely tied to motor output than e.g., decisions involving a delay to reward. To date, the neural mechanisms underlying such effort-based decisions remain incompletely understood.

I have developed a formal model to measure how effort costs decrease the value of an expected reward.

Read more in pdf

We then used this model to look at the brain networks computing decisions entailing a trade-off between physical effort and reward, highlighting areas in cingulate cortex and SMA.

Read more in pdf

Check out this lay video summary

In collaboration with Tanja Müller and Matt Apps, we recently extended the computational model and neural analyses to dissociate mechanisms of recoverable and unrecoverable fatigue in medial PFC during effort-based decision making.

Read more in pdf

Decision implementation in motor regions

In two studies, I examined whether (and how) the excitability of neurons in human primary motor cortex (M1) reflects on-going decision and motor preparatory processes. In both studies, I used transcranial magnetic stimulation (TMS) to measure the average cortico-spinal excitability (CSE) of a large pool of neurons in M1.

In the first study I found that the decision process leaks to (and thus influences) M1 while the decision is on-going, rather than in a temporally serial way.

Read more in pdf

In the second study I showed that the variability of M1-CSE, and thus the state of a motor output region, tracks the readiness to execute a prepared movement.

Read more in pdf

Decision outcomes in VTA, VS and orbitofrontal regions

In two studies, I have looked at how the brain learns the relevant dimensions of an outcome, and how it represents an outcome's identity. These studies involved functional magnetic resonance imaging (fMRI) to measure changes in brain activity using the BOLD signal.

In the first study I found that temporally precise predictions of the size of reward are present in brainstem structures (VTA) even when reward size is irrelevant for behaviour. The ventral striatum (VS), however, a structure previously shown to encode such predictions about reward size, flexibly switches to encoding the relevant outcome dimension - in our case reward timing.

Read more in pdf

In the second study, performed together with Helen Barron, we used repetition suppression to show that the identity of an outcome is represented in a posterior region of orbito-frontal cortex.

Read more in pdf