A sample of 69 healthy right-handed subjects (28 males: mean age ±SD: 38.21 ± 11.87 years (y), 41 females: 41.88 ± 12.69 y) reporting no history of psychiatric or neurological diseases participated in this study. Educational level ranged from an eighth grade to a post-graduate degree (mean education years ± SD: 15.75 ± 2.99 y). The present investigation was part of a larger project exploring the relations between brain and psychological dimensions (Picerni et al., 2019, 2021, 2022). Only those subjects who agreed to come again to the Santa Lucia Foundation to be tested on RS-10 and other psychological scales were enrolled in the current study. Data were filtered to keep only the participants who could contribute data for all variables included in the final analyses. All participants underwent MRI scanning and completed questionnaires. Inclusion and exclusion criteria are described in detail in the Supplementary materials.

Participants’ psychological profile was assessed through the Italian versions of the following self-report psychological questionnaires and scales: Resilience Scale-10 (RS-10), Attachment Style Questionnaire (ASQ), Beck’s Depression Inventory Scale (BDI), Coping Orientation to Problems and Experiences (COPE), Emotion Regulation Questionnaire (ERQ), General Self-Efficacy Scale (GSES), Hamilton Anxiety Rating Scale (HAM-A), Hamilton Depression Rating Scale (HAM-D), Holmes and Rahe Stressful Event Scale (HR-SS), Interpersonal Reactivity Index (IRI), Positive and Negative Affect Schedule (PANAS), State–Trait Anger Expression Inventory (STAXI), State–Trait Anxiety Inventory-Form Y (STAI-Y), Toronto Alexithymia Scale (TAS-20), Raven’s Progressive Matrices (RPM).

Participants’ personality traits were assessed through the Big Five Questionnaire-2 (BFQ-2) and Temperament and Character Inventory (TCI).

Italian versions of psychological and personality scales were used. Supplementary materials describe tests and questionnaires in detail and report descriptive statistics for socio-demographic and psychological variables (Supplementary Table S1).

Participants underwent a neuroimaging protocol including standard clinical sequences (FLAIR, DP-T2-weighted) and a volumetric whole-brain 3D high-resolution T1-weighted sequence, performed with a 3 T Allegra MRI. Volumetric whole-brain T1-weighted images were obtained in the sagittal plane using a Modified Driven Equilibrium Fourier Transform (MDEFT) sequence (Echo Time/Repetition Time-TE/TR- = 2.4/7.92 ms, flip angle 15, voxel size 1 × 1 × 1 mm³). All planar sequence acquisitions were obtained in the plane of the AC-PC line. The FreeSurfer imaging analysis suite¹ was used for reconstructing volumes and cortical thickness of brain regions (Dale et al., 1999; Fischl and Dale, 2000). Cerebellar parcellation was performed through a freely available patch-based multi-atlas segmentation tool called CERES (CEREbellum Segmentation) able to parcellate the cerebellar lobules automatically. CERES (Romero et al., 2017) is part of a broader software pipeline for volumetric brain analysis, namely volBrain.²

Neuroimaging data acquisition and processing are further detailed in the Supplementary materials.

Before regression modeling, a Spearman correlation filter (Hollander et al., 2013) was used to select independent variables significantly correlated to the dependent variable RS-10 (Supplementary Table S3). The 279 variables taken into account for correlations were: socio-demographic variables (n = 2); psychological (n = 34) and personality (n = 37) variables from the 16 tests detailed in Supplementary Table S3; cortical volume variables (n = 66); sub-cortical volume variables (n = 26); cortical thickness variables (n = 66), cerebellar volume and thickness variables (n = 48). Sex was used as a between-subjects factor in non-parametric statistical analyses. Selected predictors, along with socio-demographics as covariates (namely, age, education, and sex), were used in a regression analysis to predict the resilience score. For the purposes of this exploratory investigation, we considered correlations with p ≤ 0.05 statistically significant. Then, to manage the presence of several predictors and to obtain further statistical reliability, the regression analysis was performed with the Elastic Net method. The Elastic Net modulates regression coefficients to penalize complex models implementing implicit feature selection through a regularization approach that combines ridge and LASSO regression (Kuhn and Johnson, 2013; Zou and Hastie, 2005). The model was trained using repeated k-fold cross-validation, with 5 folds repeated 5 times, to avoid overfitting and ensure the reliability of the results. Repeated k-fold cross-validation was preferred to single-run k-fold cross-validation because it ensures a more accurate estimate of results by reporting the mean result across all folds from all runs. Repeated cross-validation reduces the error in the estimate of mean model performance (Kohavi, 1995). On each run of this cross-validation procedure, data were centered and scaled. Further data transformations were avoided in order to ensure the regression results were directly interpretable. The model was evaluated using R² and Root Mean Squared Error (RMSE) metrics. Feature importance was investigated by evaluating the model’s regression coefficients. All statistical analyses were performed in R.

The factor with the most predictive value was linked to self-efficacy by GSES, defined as individuals’ confidence in their ability to succeed in particular situations or to accomplish a task. The very definition of self-efficacy emphasizes how much it is related to resilience. By determining the beliefs people hold regarding their power to affect situations, self-efficacy strongly influences both the power to competently face challenges and the most likely made choices. People with high self-efficacy values perceive challenges as opportunities to be mastered instead of threats to be avoided (Bandura, 2006). Self-efficacy not only impacts our lives during highly stressful situations but also enhances our motivation and the capacity to pursue ambitious goals.

Although self-efficacy and resilience are distinct psychological resources, independent from each other, they are highly related since both of them share the ability to persevere in the face of difficulty and have a positive self-concept. By activating affective, motivational, and behavioral mechanisms in demanding situations, self-efficacy beliefs can promote resilience so much that sometimes self-efficacy has been conceptualized as one component of resilience (Werner, 1997). Remarkably, high self-efficacy levels have been linked to low levels of anxiety and low vulnerability to depression (Hinz et al., 2023). Note that within the predictors of resilience found in the present research, the values of anxiety and depression scales were the factors with the highest negative coefficients associated with resilience.

It is worth noting that within personality traits, Conscientiousness and Openness – factors of the Big Five model – and Resourcefulness and Enlightened second nature – subscales of TCI character dimension Self-directedness – predicted resilience values with positive coefficients, while Fear of uncertainty – subscale of the TCI Harm Avoidance temperamental dimension – predicted resilience values with negative coefficient. These outcomes fully agree with literature findings on healthy individuals of different ages (Graham et al., 2021; Linnemann et al., 2022; Nieto et al., 2023), although such associations have been more frequently described in patients with various pathologies. Briefly, Conscientiousness is the personality trait that implies being careful and diligent, efficient and organized. Although with less predictive impact, resilience values were also predicted by the trait Openness, which assesses how open-minded, imaginative, creative, and insightful a person is. Those who are more broadminded tend more willing to listen to multiple viewpoints or try new things (high cognitive flexibility). Not very differently, the most distinctive characteristics of self-directed individuals are that they are effective and able to adapt their behavior according to voluntary goals. Finally, we found that resilience displayed a small negative relationship with Fear of uncertainty, which is associated with behavioral inhibition, avoidance of aversive situations, poor coping, proneness to negative emotions, and the tendency to be sensitive to punishment signals. Furthermore, the TCI subscales Resourcefulness and Enlightened second nature were associated with RS-10 scores with positive coefficients. As a final note, it has to be underlined that the further subscales of TCI (Purposefulness, Extravagance, Anticipatory worry) and BFQ-2 (Energy/Extraversion, Emotional Stability) were significantly correlated to RS-10, even if not resulting as predictors in the Elastic Net model.

This ensemble of personality-related predictors of resilience reflects the benefits of having a hard-working and positive affective style as well as abilities of interpersonal closeness and social interaction. On one hand, the meticulous approach of conscientious individuals may lend itself well to effectively coping with negative life experiences, resulting in a sense of self-efficacy. On the other hand, positive emotions and close social interactions contribute to resilience because they broaden the “thought-action repertoires.”

The attachment style is an acknowledged descriptor of interpersonal patterns beginning with early interactions with the primary caregiver (Bowlby, 2003). The initial experiences have a great impact over time and influence the way later relationships are processed, the ability to cope with hardships, and the overall functioning and mental health. Attachment styles have been classically categorized as secure or insecure. It has been shown that secure attachment can foster resilience by implementing effective problem-focused coping strategies, which may in turn shape resilience (Rasmussen, 2019; Villasana et al., 2016). Secure attachment and resilience are complementary concepts, which share similar developmental circumstances, stemming from a healthy childhood, and leading to the emergence of adaptive self-esteem and empathy, through positive relations with others. Actually, individuals with a secure attachment exhibit high levels of resilience, and both these variables, in turn, correlate positively with proactive coping strategies, and negatively with avoidant coping strategies (Marriner et al., 2014). Intriguingly, we found that among the predictors of resilience, there were the positively predicting ASQ dimension Confidence and the negatively predicting ASQ dimension Need for Approval. The other ASQ dimensions, Relationships as Secondary and Preoccupation with Relationships as well as the coping strategy Problem Solving assessed by COPE inventory were positively correlated with the RS-10, although not resulting as predictors in the Elastic Net model. Specifically, active coping strategies are intentional efforts aimed at minimizing the physical, psychological, or social harm of a stressor. They are associated with actual or perceived control over the stressor, leading to changes facilitating resilient responses (Wood and Bhatnagar, 2015).

Research has repeatedly demonstrated that experiencing positive emotions in the face of adversity is one of the most important processes involved in resilience (Fredrickson, 2001; Paquette et al., 2023). In accordance, we found that resilience values were associated positively with positive mood scores and negatively with negative mood scores, as assessed by PANAS. Notably, positive emotions resulted as predictors in the present Elastic Net model for resilience. This result is not surprising, given that positive emotions play the role of a buffer between the distressing situation and the emotional elicitation and appraisal of that situation (Paquette et al., 2023; Philippe et al., 2009). The role of positive emotions in resilient behaviors has been explained by the “Broaden-and-Build Theory” (Fredrickson, 2001), positing that positive emotions facilitate resilience by broadening one’s attention and effective coping strategies (Tugade et al., 2004). Repeated experiences of positive emotions would render this broadened mindset habitual and result in increased personal resources that can be drawn on in times of need and facilitate resilient behaviors and adaptive coping strategies (Tugade et al., 2004). In addition, positive emotions would have an undoing effect, given that they counteract the deleterious after-effects of negative emotions and stress (Cohn et al., 2009).

Identifying the profile that characterizes resilient people and has predictive value as to whether or not anxiety and depression symptoms will be present is an important issue in the resilience literature (Wermelinger Ávila et al., 2017). In agreement with previous reports (Lyu et al., 2022; To et al., 2022), we found that state and trait anxiety, as well as depression, showed the highest negative coefficients associated with resilience. Predictably, as resilience reflects the ability to adaptively act against psychological distress, we found significant inverse relationships between resilience values and anxiety and depression scores. These negative predictors suggest that resilience resources may turn the triggers for anxiety and depression into opportunities to improve performances and overcome difficulties, accordingly with the inverse associations between resilience and psychological distress described in patients with chronic diseases (To et al., 2022), and older people (Wermelinger Ávila et al., 2017).

In addition to the psychological and personality factors, we inserted several brain morphological features within the present regression model to achieve an even more multifaceted profile of the resilient phenotype.

Human cross-sectional studies have focused on neural structures and neuroendocrine markers of resilience, and the animal models provided data on the behavioral, genetic, molecular, and hormonal bases of resilience, showing that susceptible subjects exhibit specific molecular abnormalities and distinct epigenetic and cellular adaptations lacking in resilient individuals (Laricchiuta et al., 2023a,b; Nasca et al., 2019; Russo et al., 2012). In brief, the neuronal architecture of resilience largely overlaps with the neuronal structures related to cognitive and emotional regulation, as the executive control network (including prefrontal, frontal, and parietal regions) and the emotional arousal network (including cingulate cortex subregions, amygdala, hippocampus, and insula). However, up to now, these brain structures have been mainly implicated in the vulnerability, rather than in the resilience, to stress or trauma (Laricchiuta et al., 2023a,b). In fact, since the brain is continuously adapting to the perturbations in bodily homeostasis, most information regards the neurobiology of resilience as a response to disease or traumatic adversities and not to trait resilience, which we were conversely mainly interested in. However, since the maladaptive responses to stress/trauma are the flipside of resilience, it is appropriate to take into account even the literature data analyzing the brain morphological responses to stress/trauma. Interestingly, in the present model of resilience, the volumes of the amygdala and hippocampus, as well as the cortical thickness of the temporal pole, predicted the resilience values with positive coefficients.

Let us analyze the single neuronal predictors of resilience in detail.