Bridging feeling and motion: Insula–premotor dynamics in the processing of action vitality forms

Significance

Humans perform actions with different kinematic forms, communicating their attitudes to others. These action forms, known as vitality forms (VFs), are encoded in the insula (INS). Here, we investigate brain activity during the feeling and execution phases of action VFs. Results revealed that the INS is already active during the feeling phase that foreshadows action. Moreover, dynamic causal modeling revealed the direction of information flow between the INS and the parieto-frontal network during VFs processing. Specifically, affective information involves the INS and modulates the premotor (PM) cortex. During execution of VFs, motor commands emerge from the PM cortex and influence INS activity, providing information about the VFs of the ongoing action. In this way, a motor act acquires its affective color.

Abstract

Typically, people perform actions in a valenced—positive or negative—way, depending on their attitudes or desires. These forms of action are named vitality forms (VFs). While it is well established that action goals are mediated by a parieto-frontal network, less is known about the processing of VFs. Recent fMRI studies suggest that the insula (INS)—and its connections with the parieto-frontal circuit—plays a crucial role in VFs processing. However, a key question remains: How does our internal affective state shape our motor behavior? To explore this issue, we conducted an fMRI study. Participants were required to perform two sequential tasks: 1) to evoke either a positive (enthusiastic) or a negative (angry) affective state (feeling task); 2) to execute an action while maintaining these affective states (execution task). Univariate analysis revealed activation of the INS and dorsolateral prefrontal cortex (PFC) during the feeling task, which extended to the premotor (PM) and parietal areas during the execution task. To determine the directionality of information flow among these nodes, we employed dynamic causal modeling. Bayesian model comparison showed that, during the feeling task, affect generation involves INS, which, together with the PFC, modulates the activity of PM. In contrast, during execution, motor commands emerge from PM and influence activity in the INS and PFC. These findings indicate that while the internal states crucially imply the INS, their regulation is mediated by PFC. The PM cortex plays a crucial role in the selection of the corresponding action VFs.

There is consistent evidence that the neural correlates of goal-directed actions are formed by a cortical network comprising parietal and frontal areas (1–6). This same network is involved in both the execution and the observation of goal-directed motor actions (1). In addition to the goal, another fundamental aspect of an action is how the action is performed. When interacting with others, the same motor act may be performed in different ways, defined by Stern vitality forms (VFs) (7). VFs can be expressed in gentle, enthusiastic, rude, or other manners. For example, when we meet friends, we greet them warmly, expressing our positive feeling toward them. Similarly, when we observe actions performed by others gently or rudely, we immediately understand their positive or negative intentional stance. Several fMRI studies provided evidence that both the observation and the execution of actions expressed with VFs activate, in addition to the parieto-frontal areas, the dorso-central insula (DCI) (8–10).

However, a key question remains: How do the insula (INS) and the parieto-frontal network interact when we perform actions with VFs? More specifically, how do internal affective states shape our motor actions? More generally, it remains unclear, on one side, where the internal feeling is generated, and on the other, how it is transformed in the appropriate action VFs. To explore these issues, we conducted an fMRI study. Specifically, we required participants to perform two tasks: 1) to induce and maintain a specific affective set or state (feeling task, FEEL) and 2) to execute the corresponding action (execution task, EXE). Regarding the FEEL task, participants observed an image showing a colored heart instructing participants to feel enthusiastic (yellow color) or angry (red color) or to feel neutral (gray color, control condition). To ensure the intended feeling was induced, participants recalled an affective state they had actually experienced in real life, either positive (e.g., receiving an excellent grade) or negative (e.g., being unfairly failed). For further details, see the (SI Appendix, Fig. S1–S3).

Subsequently according to the previously induced affective state, in the EXE state, the participants had to transform the feeling state into VFs performing the action with enthusiasm or anger (vitality-form condition). In the control condition, participants performed the same action without any VFs (control condition).

Data were analyzed using two procedures: whole-brain univariate analysis and dynamic causal modeling (11–14). Results of the univariate (SPM) analysis showed that the INS and the dorsolateral prefrontal cortex (PFC) were already activated during the feeling task, thus disclosing the crucial role of the INS in the generation of the affective context for the action. During the execution of action VFs, besides these two affect-induced activations, brain responses were also observed in the premotor (PM) and parietal areas. Additionally, by using dynamic causal modeling, we established the causal role played by INS in relation to PFC and PM responses, during the feeling task as well as the execution of actions endowed with VFs. Results of Bayesian model comparison showed that, during the feeling task, generation of affective information involves the INS, which, together with the PFC, modulates the activity of PM. In contrast, during the action execution task, motor commands emerge from PM and influence activity in the INS and PFC. These findings provide evidence on how the affective states generated in the INS, and modulated by PFC, influence our action VFs.

Results

Brain Activations During Feeling and Execution of VFs.

The primary aim of this study was to investigate INS activation during the FEEL and EXE tasks (see also SI Appendix, Fig. S4). Regarding the FEEL task, the contrast Vitality Forms vs. Control revealed consistent activation of the anterior and middle short gyri of the left INS, the left dorsolateral PFC extending to the frontal pole, the left cerebellum, and the occipital cortex (Fig. 1 A, 1 and 2). Regarding the EXE task, the same contrast showed activation of the same INS sector, extending to its dorso-central part, as well as the left cerebellum and bilateral occipital cortex (Fig. 1 B, 1 and 2; see also SI Appendix, Fig. S4).

Fig. 1.

Brain activations during vitality forms (VFs) processing. Sagittal and transverse sections showing the activations of the insula (INS) and other areas in the two hemispheres during the direct contrasts VF FEEL vs. CT FEEL (A) and VF EXE vs. CT EXE, respectively (B). These activations are rendered using a standard Montreal Neurological Institute brain template (PFWE < 0.05 cluster level).

Dynamic Causal Modeling Analysis.

The factorial nature of our experimental design enabled us to model the interactions between vitality (Vitality versus Control) and task (feeling versus execution) in terms of changes in directed connectivity due to the effect of VFs. This allowed us to identify the network of directed connections that were modulated by VFs. Results of a parametric empirical Bayes (PEB) analysis, with Bayesian model comparison, showed that a particular network of connectivity (model 4) best explained our data, with a posterior probability of 98%, in relation to alternative plausible models (Fig. 2 B and C). This model excluded connections between INS and PFC that showed a modulatory effect of VFs over the feeling and execution tasks. By thresholding the Bayesian model average (BMA) of directed connectivity estimates—at >95% posterior probability, we highlighted the connections (i.e., model parameters) that were modulated by VFs (Fig. 2D). These parameters were reported as connectivity matrices. A positive value (depicted in red–yellow) indicates a positive modulation by the vitality effect, whereas a negative value (depicted in teal–blue) indicates a negative modulation. Self-connections (the diagonal elements) are, by definition, inhibitory: Positive self-connections indicate increased inhibition, whereas negative self-connections indicate disinhibition (15, 16).

Fig. 2.

The “full” DCM model. (A) Black arrows represent endogenous connections (A matrix). Pink and green arrows represent driving inputs entering in INS (feeling task) and PM (execution task). Red dots represent modulatory inputs related to VFs (VF FEEL and VF EXE, B matrix) useful to modulate all connections, including self-inhibitory connections (circular black arrows). Model space including the “full” model (1), four reduced models (2 to 5) in which modulatory inputs to specific connections were switched off, and a “null” model (6) with no modulation serving as a baseline. (B) The best model was model 4 with a posterior probability of 98%. (C) Bayesian model average (BMA) of the model parameters that survived a posterior probability threshold of >95%. Effective connectivity matrices are shown in panel. (D) Off-diagonal values represent changes in connectivity strength, with increases depicted on a scale from yellow to red and decreases from teal to blue. For diagonal values, which reflect self-connectivity and are inhibitory by definition, the color scale is inverted: More positive values indicate stronger self-inhibition, while more negative values indicate disinhibition, relative to the default of –0.5 Hz.

Regarding extrinsic connectivity (between-region) during the VFs feeling task, results revealed a positive modulation, i.e., an increase of the connections from INS to PM (0.10) and from PFC to PM (0.14). During the VFs execution task, results revealed a positive modulation from PM to INS (0.20) and from PM to PFC (0.17) (Figs. 2D and 3B). For extrinsic connections, these changes have units of hertz (per second) and are large in relation to the average connectivity. Moreover, results showed a strong “inhibition effect” via the self-connections of PM (0.65) during the feeling of VFs (Fig. 2 D, left panel) and a strong “disinhibition effect” of PM (−0.86), INS (−1.04), and PFC (−0.29) during the execution of VFs (Fig. 2 D, right panel). For intrinsic connections, these changes are expressed in terms of the log ratio of effective connectivity. An overview of modulatory connection strengths in the winning model relative to the FEEL and EXE tasks is shown in Fig. 3.

Fig. 3.

Modulatory connection strengths in the winning models relative to the FEEL and EXE tasks surviving at BMA thresholding at >95% posterior probability (strong evidence). Red dots represent positive modulation effects during the VFs feeling (A) and VFs execution (B). Yellow arrows indicate driving inputs where the affective and motor information of action originate.

Discussion

When individuals perform a goal-directed action—like grasping an object or an intransitive action like hand waving—the behavior comprises two fundamental components: the action’s content and its form. While there is substantial evidence regarding the neural substrates of action content (1, 2, 17), much less is known about the neural mechanisms underlying action VFs.

Recent fMRI studies have shown that both observing and executing actions characterized by positive or negative VFs activate not only the parieto-frontal network but also the DCI (8, 9), and the middle cingulate cortex (MCC) (10). These findings suggest that while the parieto-frontal circuit processes the goal of an action, the DCI and the MCC are involved in processing its forms.

Although these studies highlight the existence of two distinct circuits it is less clear how they interact during the execution of VFs. In a previous study (16), using Dynamic causal modeling, we investigated the relation between the INS and the parieto-frontal network, during the processing of actions VFs. Results showed that, during action observation, two streams arose from the superior temporal sulcus (pSTS): one toward inferior parietal lobe (IPL) (encoding action goal) and one toward DCI (encoding action VFs). During action execution, two streams arose from PM: one toward IPL (goal) and one toward DCI (VFs). This last finding suggests that when VFs are triggered by cognitively following the instruction and not by a real internal feeling, VFs start from PM and reach the DCI (18). However, during social interactions, there is another way to elicit action VFs, i.e. affectively. To demonstrate the role of the INS in the generation of the affective component of action, we asked participants to maintain a positive or negative affective state for 18 seconds (feeling task) and then to perform an action accordingly (execution task). Results of a univariate analysis showed that during the feeling task, the INS and the dorsolateral PFC were activated. Furthermore, during the execution of action VFs, these activations extended to other brain regions like the PM and parietal areas.

Dynamic causal modelling (DCM) (11–14) analysis allowed us to also quantify the direction of information flow among INS, PFC, and PM nodes during the feeling and execution of VFs. Results of this analysis revealed that during the feeling phase, affective information implies the INS, which, together with the PFC, modulates the activity of the PM cortex. In contrast, during the execution phase, action conveying associated VFs emerges from the PM and, in turn, influence both the INS and PFC nodes.

This result suggests an interpretation of the generation and execution of action VFs. Regarding affect generation: When performing an action that conveys a genuine affective state, this information is encoded in the INS and transmitted to the PM cortex, which contains a repertoire of motor acts characterized by specific kinematic profiles. It is plausible that in the PM area, information from the INS helps the selection of the most appropriate kinematic patterns, enabling the agent to act in a way that aligns with their affective state. It is important to note that the regulation of the affective state could be modulated by PFC (19).

The presence of kinematic motor repertoire in PM is corroborated by studies carried out in both humans and monkeys. Specifically, Di Dio and colleagues (20) showed that, in humans, the PM and parietal areas encode kinematic information related to reaching movements performed by a biological effector. This work also demonstrated increased parieto-frontal activity during observation of different movement velocities. These findings are in line with monkey data showing that in the PM cortex are located neurons that encode various aspects of hand movement (21–24), including speed (22–25), distance (26), and reaction time for reaching (27, 28).

Regarding the execution of VFs, our findings showed that during this phase, the PM cortex sends information on the ongoing action to the INS as well as other structures such as the parietal area. It is plausible that this information allows the agent to associate the ongoing action with the corresponding affective state.

The present findings clearly indicate that VFs generated by the PM cortex are driven by affective states arising from INS activation. Beyond this affective elicitation of VFs in the PM, it is plausible that in everyday life, VFs can also be voluntarily generated either in response to social contexts or, as in experimental settings, based on explicit instructions which could be modulated by PFC. Future research should clarify this point, specifically the possibility that VFs could be generated affectively or cognitively. Interestingly, the dichotomy we proposed is also evident in the context of laughter production, revealing two relatively distinct neural networks: one linked to emotional laughter and the other to non-emotional, conversational laughter (29).

In conclusion, considering the findings of this study alongside those of the previous one (18), which included an action observation condition, the reversal of directed influences—from a top–down, centrifugal influence on the INS to a bottom-up, centripetal influence on the PM cortex—provides insight into the interpretation of action observation within the framework of VFs. In other words, the top–down influences (PM -> INS) enable intentional priors to select appropriate kinematic forms during motor preparation, while bottom-up influences (INS -> PM) enable the formation of posterior beliefs about the intentional stance conveyed by the movement. In one sense, this generalizes the notion of the mirror neuron system to encompass the inferred intentions of the self and other (30–33).

Methods

Participants.

Twenty-two healthy right-handed volunteers (10 females, average age = 24.1, SD = 3.2) participated in the fMRI experiment. All had normal or corrected-to-normal visual acuity. None reported a history of psychiatric or neurological disorders or current use of psychoactive medications. Written informed consent was obtained from all participants, and the study was approved by the Local Ethics Committee of Parma (552/2020/SPER/UNIPR) in accordance with the Declaration of Helsinki.

Paradigm and Task.

The fMRI experiment comprised three functional runs. In each run, participants were presented with images regarding two tasks (FEEL and EXE) and two different conditions (VFs; control, CT). In total, four conditions were presented in independent randomized blocks (VFs FEEL, VFs EXE, CT FEEL, CT EXE) (see also SI). Notably, two days before the fMRI experiment, participants were invited to familiarize themselves with the experimental procedures. Specifically, they were trained to recall a positive feeling by thinking about when they received an excellent grade during their academic career or a negative feeling by thinking about when they were unfairly graded (for details, see SI Appendix, Fig. S1). The day of the experiment, before starting the fMRI session, participants were required to perform the same training task. This time, after each trial, they rated their performance by answering “How well did you succeed?” on a scale from 0 to 100. The fMRI experiment began only after when the participant indicated that they were able to recall the positive, neutral, or negative feelings (score ≥ 70). This procedure facilitated the re-experiencing of the affective state, ensuring that it occurred not merely at a cognitive level but involved entailed genuine affective participants’ affective experience.

During the fMRI experiment, before starting the FEEL task, the instruction, prepare to be positive/negative, remind participants to focus on the affective state to enter during the FEEL task (VFs FEEL). Subsequently, during the EXE task, participants transformed that feeling into action VFs, by simply rotating a little box with enthusiasm or anger (VFs EXE, Fig. 4). A cue presented in the center of the screen indicated when to execute the action. During the Control condition, participants received the instruction, prepare to be neutral, felt in that way (CT FEEL), and then executed the same actions with that neutral feeling (Fig. 4). The control stimuli were designed to enable participants to perform the same tasks without conveying any affective information.

Fig. 4.

Experimental Paradigm. Separate blocks presented VFs and control conditions. Each condition started with the instruction, reminding participants to enter in a specific affective state. Subsequently, they were asked either to feel enthusiastic/angry (VFs condition) or to feel neutral (control condition). Finally, in the execution task, participants had to perform an action with that specific feeling. After the action execution, a block interval lasting 18s was presented (Rest).

fMRI Data Acquisition and Analysis.

Anatomical T1-weighted and functional T2*-weighted MR images were collected using a 3 Tesla General Electric scanner (see details in the SI Appendix). Following standard preprocessing procedure, the data were analyzed with a random-effects model consisting of a two-level procedure. At the first level, each participant’s fMRI BOLD signal was modeled using a general linear model (GLM) with a design matrix containing the onsets and durations of each event, specified according to the experimental task for each functional run. The GLM comprised the following regressors: Vitality Feeling (VFs FEEL), Control Feeling (CT FEEL), Vitality Execution (VFs EXE), Control Execution (CT EXE), and Instruction. Within each block, the videos were modeled as a single event with a duration of 18 s. The instruction was modeled with a duration of 6s. In the second-level (group) analysis, the contrast images of the each participant computed in the first level, were analyzed using a flexible ANOVA with sphericity correction for repeated measures. This model was composed of 4 regressors (VFs FEEL, CT FEEL, VFs EXE, and CT EXE) and considered the activation pattern obtained for different tasks (FEEL, EXE) in two different conditions (VFs and control). Within this model, we assessed activations associated with each task versus implicit baseline (fixation cross (SI Appendix, Fig. S4) and activations resulting from the direct contrast between conditions (VFs FEEL vs. CT FEEL, VFs EXE vs. CT EXE; Fig. 1). To identify the overall activity patterns involved in the feeling and execution of VFs, a global analysis was carried out (VFs FEEL & VFs EXE; SI Appendix, Fig. S4). After the identification of this activity pattern, the following regions were identified in the left hemisphere: dorsolateral PFC, PM, and the central sector of INS (SI Appendix, Fig. S4).

Dynamic Causal Modeling.

Selection of volume of interest and BOLD time-series extraction.

The first step of the DCM analysis involved identifying regions of interest (ROIs) and extracting their corresponding time series. In this study, as in our previous work (18), we selected three brain regions implicated in both the perception and execution of actions conveying VFs: PFC, PM, and INS. For each participant, three spherical ROIs were created around the coordinates identified at the group level [(PM: x = −46, y = 4, z = 44 (average: x = −45.2, y = −3.6, z = 44, SD: x = 2.5, y = 1, z = 1.05; INS: z = −38, y = 10, z = −4 (average: x = −38.1, y = 10.2, z = 3.9, SD: x = 0.58, y = 1.27, z = 0.42; PFC x = −42, y = 32, z = 32 (average: x = −42, y = 31.7, z = 31.8, SD: x = 0, y = 1.27, z = 1.5)]. If a subject showed no activation at one of the specific group-level coordinates, activated voxels were identified near the expected location. Then, for each ROI, time series data were extracted for four conditions (VFs FEEL, CT FEEL, VFs EXE, CT EXE) using the eigenvariate function (5 mm radius sphere).

First-level DCM analysis.

The second step of the DCM analysis involved the specification of a full model for each participant (Fig. 2A). In this model, matrix A modeled recurrent connections between all ROIs and self-connections for each region. From the GLM regressors, we define FEEL (feeling) as driving input entering in INS and EXE (execution) as driving input entering in PM (matrix C). To test the effect of VFs on the effective connectivity of the circuit (both during feeling and execution), the GLM regressors VFs FEEL and VFs EXE were used as modulatory inputs and were allowed to modulate all connections, including self-connections (matrix B).

Second-level DCM analysis with PEB.

After the estimation of each subject’s full DCM, as in our previous work (18), the estimated connectivity parameters from each full model were carried forward to the group level for PEB analysis. The PEB analysis captures commonalities and differences across participants and provides a free energy (F) score reflecting the quality of the group-level model (12). This (variational) free energy quantifies the trade-off between model accuracy and complexity, with higher values indicating better model evidence (cf., a log-evidence lower bound). By comparing free energy values from different PEB models with various parameters switched on or off, one can select the model with the greatest free energy, thereby identifying the best explanation for the dataset. In this study, we aimed to determine the best model accounting for commonalities across subjects in effective connectivity changes, driven by the modulatory effect of VFs during both the feeling of an affective state and execution of the corresponding action.

Starting from the full model, we defined a model space consisting of reduced models with different configurations of the B-matrix. Specifically, in each reduced model, the modulatory effect of VFs on the connection between two regions was switched off. The model space included six models in total: the full model, four reduced models, and a “null” model with no modulation, which served as a baseline (Fig. 2B). Bayesian model comparison was used to evaluate the evidence for each model. In addition, to aggregate parameters across all models and obtain an estimated value for each of them, a BMA was estimated (34, 35).

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

Anonymized dataset data have been deposited in ZENODO (https://doi.org/10.5281/zenodo.15656070) (36).