A series of connectional studies (Gerbella et al., 2017) highlighted the presence of a modular organization in the monkey ventrolateral prefrontal cortex (VLPF). Indeed, VLPF is arranged into three vertical strips, rostro-caudally organized: the posterior and intermediate sectors take part to oculomotor and skeletomotor circuits, respectively, and the anterior part has a strong intraprefrontal connectivity and a possible involvement in more abstract operations. Such connectional evidence is also in agreement with functional observations demonstrating that lateral PFC has a rostro-caudal hierarchical organization (Koechlin et al., 2003; Riley et al., 2017). Rozzi and coworkers (Rozzi et al., 2023) recently studied VLPF by using a Visuomotor (Go/NoGo) task, investigating how behavioral rules and different phases of the behavioral tasks are encoded in the whole region. As they did not analyze the differential representation of the various functional aspects in the distinct sectors constituting VLPF, in the present work we re-analyzed the populations of recorded neurons based on an architectonical and connectional parcellation of VLPF, by employing Demixed Principal Components (dPCA), decoding, and hierarchical clustering analyses. Monkeys performed two conditions, Inaction and Action, instructed by two differently colored cues (red and green, respectively) projected onto a closed box door placed in front of them. Monkeys had to mantain the fixation onto the cue and, at the box door opening, to observe one of three presented objects (cube, cylinder, sphere). When cues switched off, monkeys had to either maintain their gaze on the object for 600 ms (Inaction condition) or to reach for the object, grasp it and hold it for at least 600 ms (Action condition). The location and extent of VLPFC areas were based on architectonical and connectional maps already present in the literature (Gerbella et al., 2007, 2010; 2013; Borra et al., 2011). These maps were superimposed on the histological reconstructions of the recorded animals using a non-linear transformation procedure. Neural activity was recorded by means of a multielectrode system (Alpha- Omega Nazareth, Israel). The rostral sector was considered as a single region due to the small number of recording sites located in it. Microstimulation sessions and recordings of saccade-related activity allowed us to define a pre-arcuate oculomotor region that includes areas 45B and 8, that were excluded from further analyses. To evaluate how the populations of task-related neurons encode the different factors involved in the task, which are the Condition (Action vs Inaction) and the Object (sphere vs cube vs cylinder), we adopted the demixed principal components analysis (dPCA, Kobak et al., 2016), which allows to quantify the percentage of variance explained by each of them. Decoding analysis (Meyers, E., 2013) was used to evaluate whether the same factors are encoded in dynamic or static patterns of activity. In order to evaluate from a functional perspective the possible similarities and differences among areas, we described the pattern of responses of each area, through a hierarchical clustering of the firing rates recorded in the Cue, Presentation, and Go/NoGo-Behavioral response periods of the Action and Inaction conditions of the task. Our data highlight a differential involvement of the various VLPF areas in coding contextual information relevant for the task execution, consistently with connectional data. Furthermore, the information coding follows a caudo-rostral direction through task unfolding, with a greater involvement of caudal areas in the initial task phases, and a progressively higher involvement of the intermediate sectors as the initial information is converted into action. Specifically, the stronger neural activity in Caudal 12r and Caudal 46v during the initial task phases is related to visual information regarding instructive cues and objects presentation. The activity in area Middle 12r mainly encodes the condition and the object during the whole task, as well as the different task phases. The activity in area Middle 46v is the one that showed the strongest condition coding during the second half of the task, from the object presentation to the action execution/witholding.
