In this study, we aimed to characterize how structural and effective connectivity within fronto-striato-motor circuits relate to individual differences in apathy in the general population. Specifically, we asked whether distinct circuits are differentially associated with apathy levels, as measured through questionnaires17,18, and the valuation of efforts and rewards, as measured through computational phenotyping of behavior in an effort-based decision-making task6,10,11,59. While previous studies have linked apathy to altered connectivity within fronto-BG networks, findings have been mixed in terms of directionality - reporting both increases and decreases in connectivity - and largely focused on non-motor circuits. In light of these mixed findings and the paucity of data on motor circuits17, the present study hypothesized altered connectivity within these circuits, without making directional assumptions regarding connectivity-behavior relationships. To address these questions, we combined DWI and dual-coil ppTMS to probe both structural and effective connectivity in key fronto-M1 and fronto-BG-M1 circuits originating from the SMA and OFC. In addition, single-site ppTMS was used to assess the physiological properties of intra-M1 circuits influencing corticospinal output. This multimodal framework thus allowed us to capture connectivity patterns across both anatomical and effective levels, from frontal input to motor cortex, and extending to the final corticospinal output stage. Among the most robust findings, we observed that both OFC-BG structural connectivity and OFC-M1 long-latency effective connectivity were positively associated with apathy scores, thus replicating this relationship across our two connectivity measures. Moreover, SMA-M1 structural connectivity was related to both apathy and effort valuation, consistent with SMA's known role in effort processing29,30. Altogether, these findings suggest that partially distinct fronto-motor circuits - originating from both OFC and SMA - contribute to apathy and effort valuation, highlighting complementary circuits as potential neuromodulatory targets for specific mechanisms of disrupted motivated behavior.
The study involved 45 healthy right-handed participants (25.1 ± 0.8 years old, 31 females, 14 males). We conducted a comprehensive neuropsychological assessment that included apathy scores (presented in Fig. 1A), as well as control variables related to apathy, such as depression and anhedonia scores (presented in Supplementary Fig. 1). Apathy scores ranged between 0.54 and 1.92 (Fig. 1A), consistent with previous studies using the same scale in healthy individuals, namely the extended version of the Lille Apathy Rating Scale (LARS-e). Higher scores indicated stronger apathy.
A possible limitation of apathy scales is that they rely on subjective psychological constructs. To address this, we employed an effort-based decision-making task, where participants decided whether to exert varying levels of effort with their right arm (biceps contraction) to obtain monetary rewards (see Fig. 1B and Methods, Effort-based decision-making task). Briefly, on each trial, participants were presented with one of 16 possible combinations of effort and reward levels, defined by 4 effort levels (20%, 40%, 60%, or 80% of their maximal voluntary contraction, MVC) and 4 reward amounts (1, 5, 10, or 20 euro cents). Using a keyboard, they had 5 seconds to decide whether to accept or reject the offer. If they accepted, they performed an isometric contraction of the right biceps, monitored via electromyography (EMG), and received the reward if the contraction met both intensity and duration criteria (see Methods for full details). A key behavioral measure in this task is the acceptance rate, which reflects participants' willingness to exert effort for reward. As expected, we replicated the classical pattern: acceptance rates decreased with increasing effort and increased with increasing reward (Fig. 1B). Interestingly, different components of behavior in this task have previously been linked to apathy-related deficits in multiple clinical populations. We replicated this finding and confirmed that higher apathy scores were negatively correlated with acceptance rates in this task, specifically in high-reward trials (e.g., 20-cent rewards at 80% of maximal effort; R = -0.29, p = .046; see Supplementary Fig. 2A). Supporting this, participants in the highest apathy quartile accepted fewer of the high-reward trials than those in the lowest quartile (0.68 ± 0.11 vs. 0.92 ± 0.04), reflecting a blunted behavioral drive to obtain high rewards. In addition, we analyzed a metric of effort exertion, defined as the excess contraction produced beyond the required level (e.g., generating 35% of MVC in a trial with a 20% MVC target corresponds to 15% excess contraction), which provides an index of voluntary drive or overcommitment during effort execution. We found that higher apathy scores were significantly associated with lower excess force (e.g., R = -0.39, p = 0.007 in 20% MVC/10 cents trials; see Supplementary Fig. 2.B), indicating reduced spontaneous effort investment. Together with the acceptance behavior results, these findings support the value of effort-based decision-making tasks for capturing inter-individual differences in motivational drive across both the decision and execution phases of goal-directed behavior.
As mentioned above, to further characterize the latent components of decision-making, we used computational modeling to estimate two key parameters: β, reflecting sensitivity to effort, and β, reflecting sensitivity to reward (see Fig. 1B as well as Supplementary Fig. 3A-C for model selection and fitting). High β and low β values indicate increased effort cost valuation and reduced reward valuation -- both features may be associated with reduced goal-directed behavior. Nonetheless, neither β nor β significantly correlated with apathy scores in our sample. Bayesian regression analyses provided moderate evidence in favor of the null hypothesis (BF = 2.97 for β; BF = 2.24 for β; see Supplementary Fig. 3D). One possible explanation is that the effort and reward manipulations in this controlled laboratory task, although widely validated and commonly used to probe motivational processes, may not fully capture the complexity or ecological relevance of the cost-benefit trade-offs assessed with apathy questionnaires. As such, the latter relate more directly to complex, real-world goal-directed behaviors than to more simple decisions probed in laboratory tasks. Another explanation is that the model fitting process, by design, emphasizes latent constructs across trial types, averaging out trial-to-trial variability in acceptance rates that might reflect subtle motivational differences linked to apathy, particularly in conditions like high-reward trials (as shown in Supplementary Fig. 2A). This may limit the detection of significant correlations between apathy scores and model parameters. Still, model-derived parameters offer a robust framework to describe general effort and reward valuation processes, and are thus widely used in the field to capture inter-individual differences in these processes.
For structural connectivity, we acquired whole-brain DWI and performed streamline tractography (Fig. 1C). We examined all tracts connecting the SMA and OFC to M1, both directly and through the BG structures identifiable using our atlas, all within the left hemisphere. The BG structures included the dorsal and ventral Caudate (d- and vCaudate), dorsolateral and ventromedial Putamen (dl- and vmPutamen), and the Nucleus Accumbens (NAcc). We also included the motor part of the thalamus to ensure the completeness of the circuits projecting to M1. We extracted the number of streamlines in each subject for 19 pairs of structures (detailed in Fig. 2). A higher number of streamlines between two structures in one subject compared to another, reflects a greater probability that the fibers linking these two regions are more numerous, occupy a larger volume, or are in better condition and thus indicate a greater structural connectivity.
For effective connectivity, we used ppTMS (Fig. 1D), which involved administering a test stimulation over the left M1, preceded by a conditioning stimulation (CS) in some trials. We recorded the motor evoked potentials (MEPs) induced by the test stimulation (TS) with EMG. We calculated a ratio of test MEPs for trials with both conditioning and test stimulations to trials with test stimulations only. A larger MEP ratio (> 1) indicates a stronger facilitatory drive from the pre-activated circuit to M1. We used an individualized, neuronavigated dual-coil approach to investigate circuits from SMA and from the frontopolar portion of OFC to M1 in the left hemisphere. We used short (7 ms) and long (15 ms) inter-stimulation intervals to probe short-latency and long-latency circuits, involving preferentially fronto-M1 and fronto-BG-M1 circuits, respectively. We also used a single-coil ppTMS approach to examine intra-M1 inhibitory and excitatory circuits. This technique relies on systematic variations of the inter-stimulation interval (ISI) and the intensity of the CS to preferentially probe different circuits with one coil positioned over M1. Specifically, short ISIs (e.g., 3 ms) with subthreshold CS target fast, low-threshold GABA-mediated inhibition, as supported by pharmacological and electrophysiological studies (see ref. , for review). Then, longer ISIs (e.g., 100 ms) with suprathreshold CS engage slow, high-threshold GABAb-mediated inhibition, while intermediate ISIs (e.g., 10 ms) with subthreshold CS recruit glutamatergic neurons. The functional strength of these specific intra-M1 circuits is quantified by considering changes in the amplitude of MEPs elicited by the test stimulus (TS) when this TS is preceded by a CS, compared to when it is applied alone, providing a measure of intra-M1 effective connectivity.
Our study aimed to identify which independent variables quantifying structural connectivity (number of streamlines in 19 tracts) and effective connectivity (MEP ratios) are associated with three dependent variables: apathy scores, β and β. We employed a conservative two-step statistical approach. First, we used Least Absolute Shrinkage and Selection Operator (LASSO) regression, a penalized least squares method that selects relevant independent variables (see also Supplementary Fig. 4 for a replication of the main findings using a less conservative Elastic Net regression). LASSO can yield some regression coefficients (β coefficients) as zero. Only independent variables with non-zero regression coefficients are considered as associated with the dependent variable and thus selected. Second, we performed partial correlations between the dependent variables and the independent variables selected by the LASSO regression. This was done to control for multiple potentially confounding variables that may covary with apathy and effort-based decision-making processes (i.e., age, gender, depression, and anhedonia), ensuring that the associations we identified between apathy scores, β, and β, on the one hand, and structural or effective connectivity, on the other, were truly specific to the dependent variable considered.
First, for structural connectivity, we performed a LASSO regression with the number of streamlines for the 19 tracts as independent variables and apathy scores as the dependent variable. The regression identified three tracts, namely SMA-M1, SMA-dlPutamen, and OFC-dCaudate, with non-zero coefficients, suggesting potential correlation with apathy scores (LASSO's β coefficients = 0.001, 0.00005, and 0.016, respectively; Fig. 2A). Partial correlation analysis confirmed significant positive correlations between apathy scores and the number of streamlines in these circuits, even when controlling for the potentially confounding variables (R = 0.34, p = 0.028; R = 0.33, p = 0.033; and R = 0.39 p = 0.009, respectively; Fig. 2B), indicating that higher apathy scores are associated with stronger structural connectivity in these circuits. Interestingly, a second-level analysis of apathy subscores revealed that streamline count in the OFC-dCaudate tract was preferentially associated with the behavioral dimension of apathy (i.e., assessed through the action initiation subscore of the apathy scale; R = 0.31, p = 0.015; Supplementary Fig. 5A), but not with emotional or cognitive dimensions of apathy (R = -0.002, p = 0.998, and R = 0.13, p = 0.405, respectively), whereas connectivity in the SMA-dlPutamen tract covaried more specifically with the emotional component (R = 0.32, p = 0.045). These findings suggest that the observed associations with global apathy scores may reflect distinct contributions of fronto-striatal circuits to specific apathy dimensions.
As highlighted above, LASSO regression is a penalized least squares method that eliminates irrelevant independent variables by reducing their regression coefficients to zero. To ensure no false negatives in LASSO's selection, we performed additional correlations with BF computation on non-selected tracts. This involved all other SMA-BG and OFC-BG circuits. The analysis confirmed no significant positive correlations between apathy scores and the number of streamlines in these alternative circuits (p-values range = [0.437-0.872] and [0.199-0.826] for SMA-BG and OFC-BG circuits, respectively). The BF for these correlations averaged 2.47 ± 0.05 (BF range = [1.78-3.24]), indicating a higher likelihood of no correlation for these circuits. This analysis confirms the absence of false negatives in LASSO's selection and emphasizes the circuit-specificity of the significant correlations.
Second, for effective connectivity, we applied LASSO regression with all MEP ratios as independent variables and apathy scores as the dependent variable. This included ratios for short- and long-latency SMA- and OFC-M1 circuits, as well as intra-M1 GABAa, GABAb, and glutamatergic circuits. The regression revealed two circuits (short- and long-latency OFC-M1) with non-zero coefficients (LASSO's β coefficients = -0.23 and 0.61, respectively; Fig. 3A). Importantly, partial correlation analyses confirmed a significant positive correlation for the long-latency OFC-M1 circuit (R = 0.38, p = 0.014; Fig. 3B), indicating that higher apathy scores are associated with a stronger facilitatory influence of OFC on M1 (i.e., MEP ratios > 1) in this long-latency circuit. This effect aligns with the positive relationship observed between apathy scores and OFC-dCaudate structural connectivity (Fig. 2), making it one of the most robust findings of the study, as it emerges across two independent modalities of connectivity quantification. Besides, apathy scores were not associated with connectivity in the short-latency OFC-M1 circuit (R = -0.09, p = 0.571; Supplementary Fig. 6), showing that this effect was specific to the long-latency circuit potentially involving the BG. Interestingly, the significant positive correlation for the long-latency OFC-M1 circuit was specific to the behavioral dimension of apathy (R = 0.37, p = 0.015), but not to the emotional or cognitive components (R = 0.22, p = 0.152, and R = 0.19, p = 0.222, respectively; Supplementary Fig. 5B). This mirrors the similar association found for OFC-dCaudate structural connectivity, suggesting that disruptions in OFC-NG-M1 circuits may selectively contribute to reduced action initiation.
Hence, our findings establish a link between apathy scores and connectivity in different SMA and OFC circuits part of which connect to M1. Notably, apathy scores showed a positive correlation with both structural and effective connectivity in OFC-BG-M1 circuits. However, the association observed with structural connectivity in SMA-M1 and SMA-dlPutamen circuits did not extend to effective connectivity. All observed correlations were positive, indicating that higher levels of apathy are associated with a stronger structural connectivity in these circuits and a stronger facilitatory influence from OFC to M1. This aligns with previous reports linking high subclinical apathy levels to increased activation and hyperconnectivity in frontal structures.
We aimed to characterize the brain circuits associated with increasing β values. High β values are of particular interest because they reflect a stronger reduction in acceptance rates as effort levels increase. In other words, high β values indicate a greater sensitivity to effort costs, which is linked to reduced engagement in effortful tasks.
The relationship between connectivity and effort valuation was analyzed using the same statistical procedure as for apathy scores. First, we performed a LASSO regression, with the number of streamlines for the 19 tracts as independent variables and β as the dependent variable. The regression identified seven tracts with non-zero coefficients: SMA-M1 (LASSO's β = 0.21), SMA-dlPutamen (β = -0.02), SMA-vmPutamen (β = -0.31), SMA-NAcc (β = -0.25), vCaudate-GP (β = -0.04), NAcc-GP (β = -0.11) and GP-Thalamus (β = 0.02; Fig. 4A). Partial correlation analyses confirmed significant correlations between β and the number of streamlines in three of these seven tracts, namely SMA-M1 (R = 0.42, p = 0.006), SMA-NAcc (R = -0.32, p = 0.042) and NAcc-GP circuits (R = -0.41, p = 0.007; Fig. 4B). As indicated by the sign of the R-values, the correlation with the number of streamlines in SMA-M1 circuit was positive while the correlations with SMA-NAcc and NAcc-GP circuits were negative, indicating that higher levels of effort valuation are associated with stronger cortico-cortical connectivity and weaker cortico-subcortical connectivity in SMA-related circuits. Other partial correlations were not significant and are presented in Supplementary Fig. 7 (p-values range = [0.198-0.784]).
Second, we applied LASSO regression with MEP ratios as independent variables and β as the dependent variable. The regression yielded zero coefficients for all circuits (i.e., all β = 0), signifying a lack of association between β and effective connectivity in the investigated circuits. All graphs depicting the results of the LASSO regression as well as the absence of significant partial correlation between all effective connectivity data and β are provided in Supplementary Fig. 8 (p-values range = [0.385-0.887]).
Hence, our findings establish a link between effort valuation and structural connectivity in different SMA-related - and not OFC-related - circuits, in line with the well-known role of SMA in effort valuation. However, as for apathy scores, the associations observed with structural connectivity in SMA-related circuits did not extend to effective connectivity alterations, in line with the idea that structural and effective connectivity measures provide complementary insights into neural circuits and do not necessarily co-vary with one another. Further, the correlation was positive for the SMA-M1 circuit and negative for SMA-NAcc and NAcc-GP circuits, indicating that effort valuation might be associated with different alterations in distinct fronto-motor circuits.
We aimed to characterize the brain circuits associated with decreasing β values. Low β values are of particular interest because they reflect a lower increase in acceptance rates as reward levels increase. In other words, low β values indicate a lower sensitivity to rewards, which is linked to reduced engagement in effortful tasks.
The relationship between connectivity and reward valuation was analyzed using the same statistical procedure as for the other dependent variables. First, we performed the LASSO regression, with the number of streamlines for the 19 tracts as independent variables and β as the dependent variable, and identified one tract with a non-zero coefficient (NAcc-GP, LASSO's β coefficient = -0.012; Fig. 5A). The partial correlation analysis confirmed the presence of a significant negative correlation between β and the number of streamlines in the NAcc-GP circuits (R = -0.42, p = 0.006; Fig. 5B), consistent with the well-identified role of NAcc in value computation. As such, the negative correlation indicates that lower β parameters - i.e., lower levels of reward valuation - are associated with stronger structural connectivity in this circuit.
We then applied the LASSO regression with all MEP ratios as independent variables and β as the dependent variable and identified five circuits with non-zero coefficients (Fig. 6A): intra-M1 GABAa (LASSO's β = -0.015), intra-M1 GABAb (β = -0.006), intra-M1 glutamatergic (β = 0.002), short-latency OFC-M1 (β = 0.005) and long-latency OFC-M1 circuits (β = -0.0004). Partial correlation analyses confirmed the presence of a significant correlation between β and two of the five circuits, namely intra-M1 GABAa (R = -0.50, p = 0.0007) and intra-M1 GABAb circuits (R = -0.31, p = 0.047; Fig. 6B), two circuits that are known to be modulated during reward-based tasks. As indicated by the sign of the R-values, the correlations between β and effective connectivity in intra-M1 GABAa and GABAb circuits was negative. The negative correlations indicate that participants with lower β, reflecting lower sensitivity to rewards (i.e., those at the bottom of the y-axis in Fig. 6B), had reduced effective connectivity in intra-M1 GABAa and GABAb circuits. Reduced effective connectivity suggests less inhibitory influence from these GABAergic circuits on corticospinal neurons (i.e., a reduction in the suppressive effect of GABAa and GABAb circuits, seen on the right side of the x-axis in Fig. 6B). In simpler terms, individuals who placed a lower value on rewards exhibited less GABA-mediated inhibition within M1. The other partial correlations were not significant and are presented in Supplementary Fig. 9 (p-values range = [0.101-0.907]).
Hence, our findings establish a link between reward valuation and structural connectivity in an isolated tract of the BG circuitry - i.e., the NAcc-GP tract. Interestingly, this tract's connectivity is also tied to effort valuation, supporting the NAcc's role in integrating reward and effort values to compute the net value of actions. We also found a negative link between reward valuation and effective connectivity in intra-M1 GABAa and GABAb circuits. This means lower reward valuation is linked to less inhibition of corticospinal neurons by these circuits, aligning with literature showing GABAergic circuits' role in reward-based behaviors. A summary of the correlations we uncovered is presented in Fig. 7.
The data presented above show correlations between apathy scores, β and β, on the one hand, and both structural and effective connectivity, on the other, including circuits projecting to and within M1. Given this, it is possible that variability in connectivity towards M1 could lead to differences in M1's net neural output across apathy scores, β and β values. If that were the case, M1 net corticospinal output would be, by itself, linked to apathy scores, β, and β.
To test this idea, we performed partial correlations with apathy scores, β, and β as dependent variables and test MEP amplitudes (obtained with a single test stimulation over M1), a proxy for M1 net output, as the independent variable. Control variables included age, gender, depression, and anhedonia. Partial correlations showed no significant link between apathy scores (R = -0.11, p = 0.468), β (R = 0.05, p = 0.737), β (R = -0.12, p = 0.446) and test MEP amplitudes (Supplementary Fig. 10). BF computation indicated a higher likelihood of no correlation (BF averaged 2.59 ± 0.34, range = [2.12-3.25]).
Thus, despite correlations between apathy scores, effort valuation, reward valuation, and connectivity in circuits projecting to M1, these dependent variables are not linked to M1 net corticospinal output alone. This analysis is also an important methodological control as it shows that the relationships between apathy scores, β, β, and MEP ratios are not due to associations with the test MEP amplitudes exploited to compute these ratios.