Altered Expectations in Depression: A Predictive Processing Model

Post by Stephanie Williams

What's the science?

The cognitive model of depression suggests that individuals with depression experience the world in a more negative way -- likely as a result of abnormal information processing and cognitive function. Although this model has been valuable in understanding depression, some researchers have argued that altered expectations, specifically, might underlie Major Depressive Disorder (MDD). Mismatches between predictions and expectation (also known as ‘prediction errors’) are used to update current belief models. Individuals with depression are known to generate predictions and process mismatches between predictions and expectations differently than people who are considered mentally healthy. This week in Biological Psychiatry, Kube and colleagues propose a new model that incorporates aspects of predictive processing to explain how learning from predictions can be affected in MDD. 

How did they do it?

The authors review a collection of recent findings on cognitive processing in depression to propose a new model of depression focusing on expectations and predictive processing. They summarize key findings related to the neurobiology of prediction, behavioral studies and neurophysiological studies that support their framework for altered expectation and prediction error processing in MDD.

What did they find?

The authors propose that individuals with MDD predict negative experiences, discard positive information, and then find confirmation of those negative predictions. This process of discarding positive feedback is referred to as “cognitive immunization.” This framework suggests a biased learning process in individuals with depression, resulting in sustained negative predictions about their environment (i.e. a “negative feedback loop”). Interestingly, the authors point out that healthy individuals employ the same cognitive immunization strategy --- when faced with disconfirming negative information, healthy individuals discard the negative information and subsequently sustain positive expectations, which is related to optimism bias. Individuals with MDD rarely predict positive experiences, and consequently, miss the opportunity to perceive them. Specifically, the authors propose that individuals with MDD attend to negative experiences with greater precision, reducing the weight of positive experiences. The authors use their model to draw conclusions about learning rates in healthy and depressed individuals, which is dependent on the valence of the experience. They show that individuals with depression tend to maintain negative predictions longer than healthy people before updating them. The authors also suggest several brain regions involved in the maintenance of MDD, including the prefrontal cortex and the ventral striatal regions. The cognitive immunization strategy that individuals with MDD employ in the face of positive information may be supported by the suppression of prediction error processing in the ventral striatum by the prefrontal cortex. 

What's the impact?

The authors proposed a novel model focused on expectation that helps advance our knowledge of distorted cognition in depression. Their proposed framework could inform treatment strategies, such as attempts to reduce individual use of cognitive immunization against positive information. The authors also propose future research involving prediction errors to better understand which types of prediction errors are critical in perpetuating depressive symptoms.

Kube et al. Distorted Cognitive Processes in Major Depression- A Predictive Processing Perspective. Biological Psychiatry (2019). Access the original scientific publication here.

Brain Connectivity and the Spread of α-Synuclein Pathology

Post by Shireen Parimoo

What's the science?

Misfolding of α-synuclein protein is thought to underlie neurological disorders like dementia with Lewy bodies and Parkinson’s disease. As Parkinson’s disease progresses, Lewy bodies composed of misfolded α-synuclein are found in an increasing number of brain regions. However, little is known about how α-synuclein pathology spreads through the brain over time. Research in rodents has found that injecting misfolded α-synuclein directly into the brain can induce the aggregation of the rodent's own α-synuclein into Lewy body-like aggregates, eventually leading to neuronal loss. Additionally, people with a G2019S mutation in the LRRK2 gene have an increased risk of getting Parkinson’s disease, but not everyone with this mutation goes on to develop Parkinson’s disease. Thus, it is unclear how this genetic risk factor interacts with α-synuclein pathology in the development of disease. This week in Nature Neuroscience, Henderson and colleagues quantified the spatiotemporal spread of α-synuclein pathology in non-transgenic mice and examined the effect of the G2019S LRRK2 mutation in transgenic “at-risk” mice. 

How did they do it?

The authors injected pre-formed α-synuclein fibrils into the dorsal striatum of non-transgenic mice and mice with the G2019S LRRK2 mutation. α-Synuclein pathology was then quantified using immunohistochemistry one, three, or six months after the injection in 172 regions throughout the brain. A separate quantitative algorithm allowed the authors to develop a measure of neuron loss based on Lewy body loss over time, which was validated by counting the number of dopamine neurons lost in mice (detected by tyrosine hydroxylase antibodies). This experiment generated the first quantitative map of α-synuclein pathology in mice.

To further understand factors which impact the spread of α-synuclein pathology, the authors generated a network diffusion model that makes predictions about how pathology would spread along the axonal connections between regions. As a means of validation, they also tested several other models which predict spread based on 1) how close regions are to each other, 2) only anterograde (forward from neuron cell bodies to presynaptic terminals) connections, 3) a mixed-up connection map or 4) simulated injections from other regions. Finally, they measured relative vulnerability of brain regions to pathology by comparing the data to the model’s predictions. The reasoning behind this approach is that if there is more pathology in a given region than the model predicts, then that region is more vulnerable. Conversely, if there is less pathology than predicted by the model, then those regions are more resilient to α-synuclein pathology.

What did they find?

In non-transgenic mice, α-synuclein fibril injection induced reproducible spatiotemporal patterns of α-synuclein pathology in many regions of the brain, including the vulnerable substantia nigra. Neuronal death was observed primarily in the ipsilateral substantia nigra, which contains the dopamine neurons most affected in Parkinson’s disease. The network model based on retrograde connectivity (backwards from the synapse to neuron cell bodies) showed the best fit to actual pathology data, suggesting that these connections form the major pathway for pathology spread. Using this model, the authors found that thalamic nuclei were among the most resilient to pathology, and the amygdala and the piriform cortex were among the most vulnerable. These vulnerability measures correlated with α-synuclein gene expression, meaning that both the anatomical connectivity of brain areas and their gene expression profiles are important determinants of α-synuclein pathology. Interestingly, a different pattern was observed in G2019S LRRK2 transgenic mice, and pathology increased selectively in regions that were resilient in non-transgenic mice, indicating that the spread of α-synuclein pathology varies in the presence of genetic risk factors. 

What's the impact?

This study is the first to quantify the spread of misfolded α-synuclein throughout the brains of non-transgenic and transgenic mouse models of Parkinson’s disease, and to generate a network model that can predict how α-synuclein pathology spreads. The final model, which predicts how pathology will spread through the brain based on anatomical connectivity and α-synuclein gene expression, provides a valuable tool for investigating the impact of genetic risk factors, understanding regional vulnerability and estimating the efficacy of therapeutic interventions.

Henderson et al. Spread of α-synuclein pathology throughout the brain connectome is modulated by selective vulnerability and predicted by network analysis. Nature Neuroscience (2019). Access the original scientific publication here.

Generating New Neural Patterns With Learning

Post by Anastasia Sares

What's the science?

Plasticity is a common buzzword in the neuroscientific community nowadays. It refers to the brain’s ability to re-organize itself in order to learn new skills or accommodate new information. But is it possible to induce plasticity in the brain and simulate real-world learning? This week in Proceedings of the National Academy of Sciences (PNAS), Oby and colleagues used a brain-computer interface to answer this question.

How did they do it?

The authors implanted a set of electrodes in the motor cortex of monkeys. The monkeys first observed passively as a random target appeared on the edge of the screen, and a cursor moved towards it. Based on the activity of the neurons recorded during this initial phase, the researchers created a rough mapping of which patterns of neurons were associated with different aspects of the movement. Then, they allowed the monkeys to take control of the cursor by feeding their neural activity directly into the computer. This is known as a brain-computer interface. Throughout, they gave the monkeys rewards when they successfully moved the cursor to the correct target and continued to refine the mapping. This “intuitive mapping,” corresponded to the monkey’s natural neural patterns for this task.

After the intuitive mapping was established, the researchers created new neural mappings that the monkeys would have to learn in order to perform the same task. There were two kinds of remapping. First, there was a simple transformation of the intuitive mapping, which would essentially keep the same neural patterns but re-assign the way they moved the cursor. Think of a video game controller that goes right when you press the “left” button. Second, they used a complex transformation, which forced the monkeys to produce completely new neural patterns, with different groups of neurons working in synchrony—a new kind of controller. The monkeys were then trained to complete the same task with these new mappings, either introducing them immediately or incrementally. Again, they were given rewards throughout, and their learning was tracked by measuring how many times they could move the cursor to the correct target in under 7.5 seconds.

What did they find?

The simple new mappings were easily learnable within a day and generally did not result in new patterns of neural activity. The complex mappings, on the other hand, were best learned over a number of days, with incremental training (gradually going from the intuitive mapping to the new mapping). The monkeys’ progress with the complex mappings over time resembled the way we learn other new, complex skills. The speed of their movements during the late stages of learning was faster than what would have been possible with the intuitive mapping, meaning that new neural activity patterns had been established in the monkey’s brains. Analyses of the neural activity for these complex mappings revealed changes in both the amount of firing for different neurons, as well as the correlation patterns between neurons.

What's the impact?

This work shows a causal link between the reorganization in a group of neurons and learning of a new skill. Though the physical connections between neurons were too small to be visible, the brains of these monkeys did develop different functional connections in order to improve on the task. Allowing neural activity to directly control the cursor eliminated many possible intermediary mechanisms. This also shows that fast, simple learning happens through a different mechanism than slow, complex learning.

Oby et al. New neural activity patterns emerge with long-term learning. PNAS (2019). Access the original scientific publication here.