Generalizable representations of pain, cognitive control, and negative emotion in medial frontal cortex.

The medial frontal cortex (MFC) contains multiple regions under intense investigation by neuroscientists including the anterior midcingulate cortex (aMCC), an area called the "holy grail" of modern cognitive science because of its critical but poorly understood function. Like many components of the MFC, the aMCC is frequently activated by painful, cognitive, and negative emotion tasks. This high degree of overlap argues against distinct aMCC/MFC processing of these different task domains, and instead supports theories emphasizing integrative processes in the MFC. For instance, it is suggested that the aMCC may detect threats to survival, explaining the overlap of pain, negative emotion, and cognition by a single higher-order process that could be engaged by each. However, part of the overlap may result from the study techniques used and not the underlying neural computations. In support of this, high-resolution and invasive studies in model organisms have provided evidence that regions within the MFC in fact have distinct subpopulations of cells that perform unique tasks (i.e., pain-selective subpopulations) instead of performing integrative processes as proposed by analysis of neuroimaging in humans.

In the paper by Kragel et al., the investigators sought to determine if pain, cognitive control, and negative emotions do indeed have overlapping and highly similar neural representations in the MFC (supporting integrative processes) or are instead uniquely represented (supporting distinct subpopulations of cells). To accomplish this, they performed a meta-analysis of 18 studies, with six pain, six cognition, and six negative emotion studies. The studies within a single domain, like pain, were not identical and often used different methods to elicit pain, such as thermal or electrical stimulation, yielding multiple subdomains within a domain. Using these data, the researchers performed an analysis for the similarity of the neural representations to identify brain representations (called representational similarity analysis or RSA) that were preserved within a domain, but distinct among domains. Additionally, the analysis took advantage of the range of tasks probing a single domain, to identify domain representations that were independent of single study or subdomain idiosyncrasies. This analysis used small volume units called voxels across the whole MFC, or larger a priori regions of interest within the MFC, like the aMCC.

In RSA, maps of activation for every subject (n = 270, 15 from each of 18 studies) were generated and correlated with each other, identifying the similarity and dissimilarity in those maps. By modeling the dissimilarity in activations with a regression model that solved for parameter weights of domain (pain) independent of idiosyncrasies from subdomains (thermal vs. electrical pain) and single studies, the authors found generalizable painful representation in the aMCC, posterior MCC (pMCC), and dorsal medial frontal cortex (dMFC). Additionally, generalizable patterns for negative emotions were found in the ventral medial prefrontal cortex (vmPFC) and dMFC. No generalizable representation was found for the six MFC regions for cognitive control after correcting for multiple comparisons. Notably, the low level of representation overlap among these domains (only for dMFC) argues against MFC function as being integrative instead of having functions specific to multiple domains.

The authors also performed two analyses taking advantage of voxel data, maximizing spatial resolution in the MFC. One approach was to make a prediction of pain, cognitive control, or negative emotion using voxels, which identified qualitatively distinct patterns for each domain across the MFC. Interestingly, while voxels of the aMCC, a focus in the ongoing debate, were heavily weighted to predict the domain for all three domains, the spatial patterns within the aMCC for each were different, again suggesting separable patterns of neural representation. Additionally, using a "searchlight" analysis that functioned similarly to the RSA analysis with a priori regions, but instead used small spheres centered on every voxel of the MFC, the authors found strong statistical evidence arguing against overlapping representation within the MFC.

Together, these data argue for domain-specific neural representation in the MFC, suggesting that subpopulations within the MFC can distinctly encode painful, cognitive, and negative emotions. These findings are consistent with animal literature and suggest that by using complex analyses based on many studies, functionally distinct subpopulations within a given region can be identified in humans. This framework for analysis should be applied to other regions of the brain, like the anterior insula, that have similarly been shown to respond to a bewildering number of distinct tasks, to determine if this region performs a higher-order integrative task, or instead contains multiple subpopulations whose boundaries are typically blurred. These data also inform our understanding of pain and suggest that there are pain-processing subpopulations in the MFC of humans. Such populations are essential to identify, as they are potential therapeutic targets in brain stimulation, and may be used to mitigate pain. This study represents a growing trend in neuroscience to rely on more sophisticated analyses of many studies instead of relying on findings from a single study, and yields a methodology that will improve our knowledge of the brain, our therapies, and our ability to make reproducible science.