Is the Dorsolateral Prefrontal Cortex Actually Several Different Brain Areas?

The human dorsolateral prefrontal cortex (DLPFC) has long captured the attention of neuroscientists because of the diversity of the cognitive functions it supports. Its involvement has been shown in a variety of executive functions such as abstract reasoning (Bernardi et al., 2020), response inhibition (Blasi et al., 2006), planning (Crescentini et al., 2012), cognitive flexibility (Badre et al., 2021), and working memory (Warden and Miller, 2010). How does one brain region accomplish so many different feats?

In truth, treating the DLPFC as a singular region likely amounts to a gross oversimplification. Even at a purely cytoarchitectonic level, the DLPFC comprises at least two subregions—Brodmann area 9 (BA9) and BA46 (Petrides, 2005). Subdivisions within the DLPFC receive inputs from distinct regions throughout the brain, providing further evidence for a more parcellated anatomic organization (Petrides and Pandya, 1984; Thiebaut de Schotten et al., 2012). Therefore, it is not unlikely that just as there are multiple inputs and anatomic distinctions, there are also functionally distinct subregions within the DLPFC. Despite these anatomic differences, functional magnetic resonance imaging (fMRI) research has not been very successful at delineating such subregions, in part because of the high variability in the location and size of reported fMRI activation sites within the DLPFC from one study to the next (Nee et al., 2007; Rottschy et al., 2012; Kohn et al., 2014). Moreover, some electrophysiology studies suggest that DLPFC neurons cannot be neatly organized into obvious functional subdivisions, because each neuron participates in multiple cognitive functions (Lin et al., 2020; Dang et al., 2022). Other studies on working memory and cognitive control have suggested that there are systematic functional differences along the rostral–caudal and dorsal–ventral axes of the entire frontal lobe (Hoshi, 2006; Badre, 2008; O'Reilly, 2010). It is thus possible that there are similarly organized additional subregions within individual frontal lobe areas (e.g., the DLPFC) that simply have not been studied in sufficient detail and with appropriate nuance.

In a new study, Jung et al. (2022) tackle the question of potential DLPFC functional subdivisions using a different approach—connectivity analysis using in vivo probabilistic tractography, as well as resting-state fMRI. Drawing on the hypothesis that functional subdivisions fall along the rostral–caudal and dorsal–ventral axes, they defined seven anatomic seed regions of interest (ROIs) consisting of 6-mm-radius spheres, spread out in each dimension across the DLPFC (Jung et al., 2022, their Fig. 1A). Examples of such seed ROIs include 9a (the anterior portion of BA9), 9p (the posterior portion of BA9), and BA46. They also defined 63 target regions spanning almost the entire frontal, temporal, parietal, and limbic cortices. Probabilistic tractography was then used to determine the structural connectivity between seed and target regions. Jung et al. (2022) found that across the DLPFC, almost all seed regions had a high degree of connectivity with the frontal pole, inferior frontal gyrus (IFG), and subcortical limbic regions such as the basal ganglia. Seed ROIs also had notable intra-DLPFC connectivity, with the strongest connections falling along the gyri on the dorsal–ventral axis. The most ventral and caudal seed regions showed high connectivity with primary and supplementary motor regions, while none of the seed ROIs appeared connected to the temporal and inferior parietal cortices. Overall, the results of the tractography analysis showed that seed ROIs along the ventral–caudal axis of the DLPFC had more connections outside the DLPFC, while those along the dorsal–ventral axis had more intra-DLPFC connectivity.

Next, Jung et al. (2022) attempted to determine the functional connectivity of their seven chosen seed ROIs using resting-state fMRI data. The seed ROIs appeared differentially connected to two distinct functional networks—the default mode network (DMN) and the multiple demand network (MDN). The DMN is classically thought of as a network of brain regions (including the medial prefrontal cortex and orbitofrontal cortex, among others) that tend to be active when an individual is not actively engaged in a specific task, and instead might be daydreaming or mind wandering (Raichle et al., 2001). The MDN, on the other hand, is thought of as a network of common brain regions (including the IFG and supplementary motor area, among others) that are active when an individual actively engages in and deploys cognitive control, regardless of specific features of the exact task at hand (Duncan, 2010; Wen et al., 2020). In other words, the DMN and MDN can be seen as complementary networks with distinct functions. Jung et al. (2022) found that the two most dorsal seed ROIs had strong resting-state functional connectivity with the DMN, whereas the remaining, more ventral ROIs had strong resting-state functional connectivity with the MDN (Jung et al., 2022, their Fig. 3). This difference provides a valuable clue as to how functional distinctions within the DLPFC might operate.

The researchers followed up on this finding by carrying out pairwise comparisons in functional connectivity between seed ROIs along the dorsal–ventral and rostral–caudal axes. First, they looked at pairs of ROIs that were matched in their position along the dorsal–ventral axis, but varied along the rostral–caudal axis (e.g., 9a vs 9p). In general, they found that the more rostral ROIs showed stronger connectivity, with brain areas falling within the MDN whereas the more caudal ROIs showed stronger connectivity with areas falling within the DMN (Jung et al., 2022, their Fig. 4). Next, they conducted similar pairwise comparisons, this time between groups of ROIs that were matched along the rostral–caudal axis but varied along the dorsal–ventral axis. Here too they found a graded axial effect, with more dorsal regions showing stronger connectivity with the DMN, and more ventral regions showing stronger connectivity with the MDN (Jung et al., 2022, their Fig. 5).

The findings from the study by Jung et al. (2022) offer both structural and functional evidence that the DLPFC is not a homogeneous chunk of cortex. Instead, the concrete differences in connectivity among the seven analyzed seed ROIs suggest that the DLPFC is composed of functionally specialized subregions. These subregions project to distinct networks of brain areas that support separate sets of cognitive abilities. Further, the nature of projections from DLPFC subregions appears to vary systematically along the rostral–caudal and dorsal–ventral axes. The rostral–caudal differences dovetail well with existing theories of the PFC that propose a hierarchical, anterior–posterior organization of brain regions, depending on the complexity and abstractness of the cognitive phenomenon they support (Koechlin et al., 2003; Badre, 2008). On the other hand, the dorsal–ventral differences align with the underlying cytoarchitectonics, since the two Brodmann areas that comprise the DLPFC, BA9 and BA46, are arranged along this axis. Overall, these findings are compatible with various existing paradigms of PFC organization and highlight the functional heterogeneity within the DLPFC.

The significance of this research is clear because our rudimentary understanding of the organization of the DLPFC limits our ability to understand its role in various cognitive processes. This is most apparent in primate electrophysiology studies of the PFC, where the translatability of findings to humans is not always clear without a good understanding of functional subdivisions within the human and primate brain (including homologies between the species). Studies like that by Jung et al. (2022) can improve this understanding (at least on the human side) and increase the likelihood that findings from primate research will end up having clinical relevance for treating psychiatric conditions.

That said, there are some limitations of this study that might be addressed in future research. For instance, the extent to which the present findings depend on the precise seed ROIs chosen is unclear Given that our current understanding of the microstructure of the DLPFC is not particularly refined, it is possible that the results of connectivity analyses might paint a different picture were one to use 3-mm-radius spheres instead of the 6-mm-radius spheres used in this article. As our understanding of the molecular composition and tissue microstructure of DLPFC gets better, it might be worthwhile to revisit such connectivity analyses. Similarly, it might prove revealing to select seed regions based on anatomic boundaries (e.g., sulci and gyri), as opposed to spheres. Future research that takes a more systematic approach to seed region selection may help answer questions about functional subdivisions in the DLPFC. In addition, while connectivity analyses can provide useful blueprints, they are not by themselves sufficient for defining clear subregions. Follow-up fMRI studies that actively test various cognitive abilities will be crucial for assigning functional specialization to the seed regions used in this article. While past active fMRI experiments have struggled with questions of subdivision (in part because of the variability in findings between studies), the ROIs outlined in the study by Jung et al. (2022) could serve as a roadmap for specific areas to examine.

Despite these limitations, the findings from the study by Jung et al. (2022) argue against treating the DLPFC as a homogeneous catch-all region for any and all cognitive faculties. While this area does support a diverse array of abilities, the equally diverse connectivity of ROIs within the DLPFC strongly suggests that the DLPFC may not be a single brain area, but rather several distinct regions.