Associations between social cognition and inflammatory markers in eating disorders and borderline personality disorder

Abstract

Introduction/objectives

Eating disorders (EDs) and borderline personality disorder (BPD) are severe psychiatric conditions characterized by emotional dysregulation and impairments in social cognition. Growing evidence suggests that inflammatory and oxidative stress pathways may contribute to their pathophysiology. The present study aimed to explore the association between social cognition performance and immune–inflammatory and oxidative stress biomarkers in individuals with BPD, restrictive ED (EDR), and purging ED (EDP) vs healthy controls (HC), using a pathway-based psychoneuroimmunological approach.

Methods

We studied a total of 100 adult women (EDR, EDP, BPD, and HC) using the Movie for the Assessment of Social Cognition (MASC) and a comprehensive clinical battery. Blood samples were drawn to quantify inflammatory, oxidative stress, and antioxidant biomarkers, including intracellular signaling pathways. Group comparisons and group-specific Bayesian Lasso regression analyses were conducted to examine within-group associations between biomarkers and social cognition outcomes, controlling for relevant clinical and demographic covariates.

Results

Participants with BPD and EDP showed greater psychological dysregulation, higher impulsivity, and increased activation of pro-inflammatory and oxidative stress pathways vs HC. The EDR group displayed an intermediate biological and clinical profile. Social cognition impairments were observed across clinical groups, with BPD participants exhibiting the highest hypomentalization. Exploratory regression analyses revealed diagnosis-specific associations between social cognition performance and immune–inflammatory pathways, particularly involving markers related to p38 MAPK, JNK, COX-2, and oxidative stress, with no significant associations observed in healthy controls.

Conclusions

The findings support a psychoneuroimmunological model in which social cognition impairments are associated with diagnosis-specific interactions between immune–inflammatory regulation and clinical phenotype in EDs and BPD. Because of the exploratory and cross-sectional design, these results should be considered hypothesis-generating and warrant replication in longitudinal studies.

Introduction

Eating disorders (ED), such as anorexia nervosa, bulimia nervosa, binge-eating disorder, and other specified feeding or eating disorders (OSFED), are severe psychiatric conditions characterized by a persistent disturbance in eating behaviors – including extreme dietary restriction, binge eating, or purging – and an excessive preoccupation with body weight or shape. These maladaptive behaviours are often accompanied by a distorted body image and result in significant physical and psychosocial impairment.^1,2

Epidemiological data indicate that EDs typically onset during adolescence or early adulthood, with prevalence rates ranging from 2% to 5% in the general population and a higher incidence rate among women, though rates among men are also significant.³ Moreover, EDs are associated with considerable medical comorbidities, including electrolyte imbalances, endocrine dysfunction or other co-occurring psychiatric conditions, and an increased mortality risk.⁴ Besides, individuals with an ED often experience significant emotional dysregulation, interpersonal difficulties, and impairments in social functioning.^5,6 These challenges frequently include heightened sensitivity to rejection, deficits in social cognition, and difficulty managing intense affective states, which can perpetuate disordered eating behaviors.^7,8

Borderline personality disorder (BPD) is a pervasive psychiatric condition characterized by emotional dysregulation, impulsivity, and unstable interpersonal relationships.¹ Individuals with a BPD often experience intense mood swings, chronic feelings of emptiness, fear of abandonment, and self-injurious behaviors, elevated suicidal ideation and a history of multiple suicide attempts.⁹ The disorder typically emerges during adolescence or early adulthood and is associated with significant psychosocial impairment and high comorbidity with mood, anxiety, and substance use disorders.¹⁰ Neurobiological and psychological research suggests that individuals with a BPD have impairments in emotional processing, executive functioning, and social cognition, including deficits in mentalization and theory of mind.^11,12

The comorbidity between EDs and BPD is well established in the literature, with the two conditions frequently co-occurring and sharing overlapping clinical and neurobiological features.¹³ Individuals diagnosed with an ED – particularly bulimia nervosa and binge eating disorder – may exhibit elevated levels of impulsivity, affective instability, and interpersonal dysfunction, traits commonly associated with BPD.¹⁴ Conversely, individuals with BPD may develop disordered eating behaviors as maladaptive strategies to cope with emotional dysregulation and identity disturbances.¹⁵ Shared psychosocial vulnerabilities, such as early trauma, attachment disruptions, and emotion regulation deficits, may contribute to the development of these two disorders. This overlap underscores the need for integrated treatment strategies that address both the core psychopathological features of borderline personality disorder and the symptomatic signs of EDs.²

Given their shared impairments in emotional regulation and interpersonal functioning, both EDs and BPD have been consistently associated with deficits in social cognition (SC). SC refers to the mental processes involved in perceiving, interpreting, and generating responses to the intentions, emotions, and behaviors of others and includes capacities such as theory of mind (ToM), emotion recognition, empathy, and attributional style, which are essential for effective interpersonal functioning. Difficulties in SC may exacerbate the relational instability and maladaptive behaviors characteristic of ED and BPD. In individuals with an ED, research has identified notable impairments in several domains of social cognition. These include reduced ability to accurately recognize facial expressions, interpret social cues, and understand others’ mental states.^5,7 Such deficits may contribute to the development and maintenance of EDs by reinforcing negative self-schemas and promoting social withdrawal or maladaptive coping strategies, such as restrictive eating or purging, especially in contexts involving perceived social evaluation.¹⁶

Similarly, individuals with BPD exhibit significant disturbances in social cognitive functioning, particularly in emotion recognition and mentalization – the ability to understand oneself and others in terms of mental states.^17,18 Hypermentalizing is commonly seen in BPD. This SC error involves over-attributing complex intentions to others, which often leads to interpersonal conflict and emotional dysregulation. Studies using the Movie for the Assessment of Social Cognition (MASC) consistently show that BPD patients have difficulties in accurately interpreting social information, especially under emotional stress situations.¹² Notably, both BPD and EDs may share overlapping deficits in social cognition, suggesting that altered functioning in domains like empathy, affective theory of mind, and reflective functioning may serve as transdiagnostic mechanisms.¹⁹

Inflammatory dysregulation has emerged as a pivotal biological mechanism associated with both EDs and BPD and has been associated with alterations in social cognition. In adults with EDs, elevated levels of pro-inflammatory markers – including NFκB, TNFα, IL-1β, COX-2, and oxidative stress indicators – have been associated with heightened emotional reactivity and impulsivity. For example, increased NFκB activity in peripheral blood mononuclear cells (PBMCs) has been associated with impairments in recognizing others’ emotions and accurately interpreting social cues, which may intensify restrictive or purgative behaviors in response to perceived interpersonal threats or rejection. These findings support a model in which immune system activation may interfere with the ability to build and update accurate mental representations of both oneself and others during social interactions – a process known as mutual mental modeling. When this process is disrupted, individuals may misattribute intentions, misunderstand emotions, and respond in maladaptive ways that reinforce disordered eating behaviors.^20,21

Similar inflammatory patterns have been described in BPD, where patients show increased activation of NFκB, COX-2, iNOS, and decreased anti-inflammatory (IκBα) or antioxidant enzymes (e.g., CAT, SOD), with inflammatory markers correlating with impulsivity and affective instability.^22–24 This systemic inflammatory profile may adversely impact neural circuits – such as the medial prefrontal cortex and limbic regions – that support mentalization, emotion recognition, and Theory of Mind. Consequently, patients with BPD may hyper- or under-mentalize the emotional states of others, leading to severe interpersonal misunderstanding. The convergence of inflammation and social cognitive deficits across EDs and BPD suggests a transdiagnostic pathway: immune-mediated neural alterations may disrupt social processing, fueling emotional dysregulation and behavioral disease.¹⁹

Rather than focusing exclusively on circulating inflammatory markers such as IL-6 or C-reactive protein (CRP), the present study adopted a pathway-based approach by examining intracellular signaling molecules involved in oxidative stress regulation and stress-responsive inflammatory cascades. Keap1 was included due to its central role in regulating the Nrf2 antioxidant response pathway, which is critical for cellular resilience to oxidative stress.^25,26 In addition, the JNK and p38 MAPK pathways were selected given their established involvement in stress-induced inflammatory signaling and their documented relevance for affective regulation, impulsivity, and cognitive processing in psychiatric disorders. This pathway-based strategy allows for a more mechanistic characterization of immune-related processes potentially underlying social cognition impairments in eating disorders and borderline personality disorder.^27,28

Although growing evidence suggests potential links between inflammation and psychiatric conditions such as EDs and BPD, the specific role of social cognition within this relationship remains underexplored. The previous sections highlight hypotheses derived from current findings, but the scientific literature remains limited and lacks targeted research examining social cognition as a core variable. Therefore, this study aims to investigate the association between social cognition impairments and inflammatory markers in individuals with either an ED and BPD.

Methods and materials

Sample and assessment tools

The sample consisted of a total of 100 individuals divided into three clinical groups and a control group. In relation to the clinical groups, patients from the Psychiatry Unit of three Spanish hospitals were recruited: participants were recruited from Hospital Clínico San Carlos de Madrid, Hospital Universitario Marqués de Valdecilla, and Hospital General Universitario de Ciudad Real. Eligible patients were required to have a diagnosis of BPD or an ED according to the Diagnostic and Statistical Manual of Mental Disorders, Fourth Edition (DSM-IV).²⁹ ED patients with either EDR or EDP subtype were considered. Participants were diagnosed according to DSM-IV criteria, as recruitment took place in 2016, prior to the widespread adoption of DSM-V¹ in the participating clinical settings. Although diagnoses were established according to DSM-IV criteria, current DSM-V¹ terminology is used throughout the manuscript for conceptual clarity and consistency with contemporary literature.

Participants were ≥18 years. Patients were excluded when the following conditions were present: a previous history of a schizophreniform, bipolar or neuropsychiatric disorder or any medical illness that may interfere with the study. In addition, patients with an intellectual quotient below 85 were not included. Healthy controls did not show any current diagnostic condition and they did not take any type of psychoactive medication.

To evaluate social cognition, the MASC³⁰ was used. MASC is an instrument that evaluates cognition using a video in which a social interaction takes place in a group of four people. Participants have to make a series of deductions about the mental states of the video protagonists. In addition to the total score (e.g., the total number of correct responses), the MASC allows the evaluation of two main types of ToM errors: error due to excessive ToM (hypermentalization index), which refers to the attribution of a mental state when it does not exist; the error due to ToM defect (hypomentalization index), which refers to the attribution of an erroneous mental state to the existing one.

To evaluate the psychopathological profile of patients, the following instruments were used: the Barrat Impulsiveness Scale (BIS-11),³¹ the Interpersonal Reactivity Index (IRI),³² the Emotional Intimacy Rejection Sensitivity Questionnaire.³³ The Trauma Questionnaire (TQ)³⁴ was used to evaluate trauma events.

Blood sample

A venous blood sample was collected in tubes with anticoagulant after overnight fasting (approx. 10 mL) on an empty stomach, in tubes with anticoagulant. The rest of the blood sample was processed to analyze the soluble and intracellular components in the inflammatory and nitrosative stress response of patients and controls as previously described.^20–22

To study the pro-inflammatory path, we focused on the induced isoforms of the enzymes involved in the inflammatory response: induced oxide synthase (iNOS), cyclooxygenase-2 (COX-2) and Kelch-like ECH-associated protein (Keap1), as well as the analysis of the levels of two of the most important pro-inflammatory cytokines, IL-1β and TNFα; and the determination of possible intracellular mechanisms involve in the expression of iNOS and COX-2: the nuclear transcription factors NFκB and the expression in PBMCs (peripheral blood mononuclear cells) of the components of the MAPK (mitogen-activated protein kinases) pathway, ERKs, JNKs, added to the analysis of p38 as elements that regulate the transcription factors by using the Western Blot technique. The oxidation degree of the cell membranes lipid components (lipid peroxidation) was analyzed by using the commercial test for TBARS (thiobarbituric acid reactive substances).

Keap1 was selected due to its central role in regulating the Nrf2 antioxidant response pathway, which is critically involved in cellular resilience to oxidative stress. JNK and p38 MAPK pathways were included given their established involvement in stress-related inflammatory signaling and their documented influence on affective regulation, impulsivity, and cognitive processing in psychiatric disorders.^27,28,35

Furthermore, the anti-inflammatory pathway components were analyzed using the Western Blot technique, as well as the nuclear extract of PBMCs of the peroxisome proliferator-activated receptor gamma (PPAR) – this is a transcription factor that inhibits gene expression and the synthesis of inflammatory mediators, on which 15d-PGJ2 acts as an endogenous ligand and oxytocin levels in blood.

Lipid peroxidation was determined by thiobarbituric acid reactive substances (TBARS) assay (ref. 10009055, Cayman Chemical Europe, Estonia), based on the reaction of malondialdehyde (MDA) and thiobarbituric acid under high temperature (95 ̊C) and acidic conditions.

Glutathione peroxidase (GPx) antioxidant enzyme activity levels were quantified using commercial kit (ref. 703102, Cayman Chemical Europe, Estonia), catalase (CAT) and superoxide dismutase (SOD) plasma levels were measured using a colorimetric kit (ref. K033-H1 and K028-H1, Arbor Assays, USA) following the manufacturer's instructions for use.

To control the possible effects of exogenous nicotine (for example, through tobacco consumption), cotinine quantitative tests (the major product resulting from nicotine metabolism) were carried out in patients and controls, using commercial EIA kits.

Data analysis

Descriptive statistics were conducted for the sociodemographic and clinical factors, as well as for the inflammatory biomarkers and the three MASC indexes (eg, total number of correct responses, hypermentalization and hypomentalization index). Chi-square-based (for categorical data) and F-based tests (analysis of variance for continuous variables) were used to compare the study groups in terms of the aforementioned variables. Effect size estimates (e.g., Cramer's V for categorical data and the η² for continuous data) were calculated to visualize the magnitude of differences. Post hoc measures under Bonferroni correction were then calculated to find pairwise differences.

Group-specific Bayesian Lasso regression with Monte Carlo Markov Chain (MCMC) was conducted to study the relationship between the biomarker levels with the MASC outcomes, controlling for relevant sociodemographic and clinical covariates (age, body mass index, traumatic experiences, measured by the TQ; concern about rejection sensitivity, using the CSR, concern score; ED symptoms, measured by the EAT-40; purging symptoms, measured with the total BITE score; and impulsivity, derived from the BIS-11 total score). The B coefficient and the 95% high density interval (HDI) was used for relationship estimate. To explore multicollinearity between predictors, the variance inflation factor (VIF) was calculated. As a result, the VIF values arranged between 1.09 and 1.98, indicating low multicollinearity between predictors.

The Lasso regression method involves estimating regression models including a regularization factor to remove noise derived from spurious relationships from predictors.³⁶ To control for potential model overfitting, the prior B (regression loading) distribution was modelled under a double exponential function assuming that many predictors may be relevant. Under this exponential distribution (also called, a Laplace distribution), a shrinkage of weak loadings is done. This allows noise (eg, weak regression effects) to zero out. A horseshoe prior was set on the sigma component for the prior B distribution.³⁷ To calculate this robust shrinkage factor, we assumed that at least 5 predictors would have a real relationship with the outcome. As a result, a λ₁ = .038 was set. By fixing this value, the model therefore forces most of the B to stay very close to zero, allowing only the variables with a strong signal to be detected as having a real non-zero effect.

Bayesian regularization approaches with shrinkage priors are particularly appropriate in settings with relatively small sample sizes and a large number of potentially correlated predictors, as they allow for stable estimation by shrinking weak or spurious effects toward zero holes preserving signals with higher posterior support. In this context, the use of a Bayesian Lasso along with a global–local shrinkage prior (horseshoe) reduces the risk of overfitting and improves model interpretability. As discussed by Piironen and Vehtari,³⁷ such priors are especially suitable for exploratory and hypothesis-generating analyses in high-dimensional models, where conventional regression techniques may yield unstable or inflated estimates.

The R² was used as a model effect size estimate. Data were analyzed using Rstudio software Version 2023.03.0+386 with the RStan packages for Bayesian inference using MCMC sampling.

Although the number of predictors included relative to subgroup sample sizes may raise concerns regarding statistical power, the present analyses were designed with an explicitly exploratory and hypothesis-generating aim. The inclusion of multiple biomarkers and clinical covariates reflects a pathway-based and mechanistic approach intended to capture the complex and multidimensional nature of immune-inflammatory regulation in ED and BPD.

Results

Clinical and demographic characteristics

The clinical and demographic characteristics of the sample (Table 1) revealed significant group differences in body mass index (BMI) and illness chronicity, although age was comparable across groups. Individuals with EDP displayed the lowest BMI values, while patients with BPD had the highest BMI, often within the overweight range. BMI category was associated with TNFα and keap1 levels (Supplementary Table S1). However, the post hoc analysis revealed the lack of pairwise differences. BPD participants also reported the longest illness duration, suggesting greater chronicity. In contrast, HC showed consistently lower levels of trauma exposure, rejection sensitivity, and eating disorder symptoms.

Table 1.

Descriptive features of the sample.

	EDR	EDP	BPD	HC	χ^2/F	ES	Post hoc
N	26	23	30	21
Age (median)	25.5	25	30.5	22	97.13	0.57
18–25	50	52.2	33.3	90.5
26–40	46.2	43.5	53.3	9.5
40–older	3.8	4.3	13.4
BMI (median)	16.33	20.28	23.39	19.38	28.34**	0.38	EDR < EDP or BPD
Underweight (<20)	84.6	43.5	20	57.1
Normative	7.7	47.8	56.7	42.9
Overweight (>25)	7.7	8.7	23.3
Duration of disease (in months)^a	111.69 (96.82)	106.17 (64.81)	195.4 (158.35)		5.01**	0.31	BPD > EDR; EDP
Types of psychoactive drugs^a,b	0.27 (0.78)[range, 0–3]	0.26 (0.75)[range, 0–3]	1.6 (1.52)[range, 0–4]		13.51**	0.26	BPD > EDR; EDP
Trauma experiences	7.08 (6.81)	11.22 (6.42)	11.1 (6.39)	4.29 (6.12)	6.37**	0.17	HC > EDP; BPD
Interpersonal reactivity	91.62 (12.94)	87.39 (24.25)	79.83 (35.27)	98.52 (14.35)	2.59	0.07
Rejection sensitivity (concern)	45.96 (7.82)	44.04 (7.67)	39.3 (10.76)	39 (8.02)	3.91*	0.11	EDR > HC; BPD
ED symptoms	57.62 (26.74)	50.17 (21.46)	35.63 (24.96)	17 (10.36)	14.65**	0.31	HC < all groups; EDR > BPD
Purging symptoms	18.08 (8.39)	33.65 (18.35)	24.37 (11.69)	8.43 (8.37)	16.59**	0.34	HC < EDP or BPD; EDP > EDR or BPD
Impulsivity	38.54 (16.92)	50.43 (19.56)	54.77 (27.14)	37.29 (14.16)	4.53**	0.12	BPD > HC; EDR
TBARS	12.97 (7.05)	12.03 (4.78)	13.24 (5.88)	17.07 (18.72)	0.32	0.01
TNFα	10.79 (6.29)	10.55 (7.01)	18.29 (40.34)	10.29 (15.76)	0.82	0.02
Catalase	236.46 (27.35)	238.53 (27.33)	252.21 (30.31)	266.17 (36.09)	3.75*	0.1	HC > EDR
GPx	152.81 (34.59)	149.74 (52.69)	142.75 (34.05)	148.33 (31.37)	0.38	0.01
SOD	1.6 (0.29)	1.56 (0.42)	1.5 (0.54)	1.82 (0.62)	1.7	0.05
iNOS	185.88 (110.92)	177.28 (86.59)	160.81 (85.46)	100.99 (20.9)	5.20**	0.14	HC < all groups
COX2	108.71 (39.44)	100.59 (47.7)	113.03 (40.55)	89.94 (27.48)	2	0.06
p38	2.71 (1.41)	2.18 (1.28)	2.86 (1.51)	1.75 (0.7)	3.51*	0.1	HC < BPD
ERK	1.63 (0.82)	1.35 (0.55)	1.52 (0.5)	1.03 (0.36)	5.5**	0.15	HC < EDR or BPD
Keap1	93.66 (41.94)	101.62 (47.38)	78.86 (49.37)	106.98 (24.93)	4.31**	0.12	HC > BPD
JNK	0.89 (0.45)	0.98 (0.41)	1.39 (0.66)	0.95 (0.38)	6.34**	0.17	BPD > all groups

Note. Percentage of cases are displayed for dichotomous and categorical variables. Means and standard deviation (between brackets) are displayed for continuous variables.

The F-based statistic was used for continuous variable contrast test. The chi-square-based test was used for dichotomous/categorical variable contrast test. Post hoc tests under Bonferroni correction were used.

EDR, eating disorder, restrictive subtype; EDP, eating disorder, binging-purging subtype; BPD, borderline personality disorder; HC, healthy controls; ES, effect size estimate (ηpartial2 for continuous outcomes and Cramer's V for non-continuous outcomes). BMI, body mass index.

a

Contrast test only conducted for the clinical groups (e.g., EDR, EDP, BPD).

b

Types considered: antidepressant, benzodiazepine, mood stabilizer, antipsychotic, stimulant, antiepileptic.

*

P < .05.

**

P < .01.

From a psychological perspective, both the EDP and BPD groups reported markedly elevated clinical symptoms, including impulsivity and purging behaviors. Impulsivity was more pronounced in the BPD group, consistent with its diagnostic profile. The EDR group, on the other hand, exhibited higher levels of rejection sensitivity, despite scoring lower on impulsivity-related measures. Finally, the patients from the BPD group showed significantly higher levels of different psychoactive medications, in comparison to the rest of clinical groups. In this group, the number of different medications was significantly associated with the TNFα levels (P = .47; P < .05) and the p38 levels (P = .46; P < .05). No correlations were found between the biomarkers and the number of different medications for the other clinical groups.

Inflammatory and oxidative stress markers

Biologically, significant group differences were observed in inflammatory and oxidative stress markers (e.g., catalase, iNOS, p38, ERK, Keap1, JNK). Regarding oxidative stress markers, no significant between-group differences were observed for TBARS or SOD at the descriptive level (Table 1). Notably, the HC group showed higher mean TBARS values, accompanied by greater variability vs the clinical group. Both the EDP and BPD groups showed heightened activation of pro-inflammatory signaling pathways (ERK, JNK, p38). The EDR participants demonstrated biomarker levels that were in between those observed in the clinical groups (BPD and EDP), who showed the highest levels of inflammation and oxidative stress, and the HC group, who exhibited the lowest levels.

Despite the absence of significant between-group differences in TBARS levels, this marker showed meaningful within-group associations with social cognition in subsequent regression analyses.

Social cognition performance (MASC)

From a cognitive standpoint, deficits in social cognition were evident across all clinical groups (see Fig. 1). On the MASC task, BPD participants had the highest levels of hypomentalization (M, 7.77; SD, 6.3), significantly exceeding those of HC (M, 4.05; SD, 2.09), EDR (M, 4.69; SD, 3.25), and EDP (M, 4.74; SD, 1.76), F(3, 96) = 4.67; P < .01, η² = .13. While HC participants showed the most accurate performance, both ED groups demonstrated social cognitive impairments, albeit less pronounced than in BPD. The performance of the EDR patients on the MASC was slightly better than that of the EDP group, but still showed signs of hypomentalization, suggesting persistent difficulty in accurately inferring others’ mental states.

Fig. 1.

MASC scores according to the study groups. EDR, eating disorder, restrictive subtype; EDP, eating disorder, binging-purging subtype; BPD, borderline personality disorder; HC, healthy controls; ES, effect size estimate (ηpartial2 for continuous outcomes and Cramer's V for non-continuous outcomes).

Associations between biomarkers and social cognition

Lasso regression analysis suggested distinct biomarker-social cognition association patterns across diagnostic groups, as illustrated in Figs. 2–4. Exact coefficients and related odds ratio (OR) estimate is displayed in the Supplementary material Tables S2–S4.

In the EDR group, two biomarkers were negatively associated with MASC accuracy: TBARS (B = −0.14; OR, 0.87; HDI95, −0.33, −0.01), and SOD (B = −0.15; OR, 0.86; HDI95, −0.33, −0.01). For the EDP group, positive associations were found between the number of correct responses and the biomarkers JNK (B = 0.18; OR, 1.20; HDI95, 0.01, 0.37) and p38 (B = 0.19; OR, 1.21; HDI95, 0.02, 0.39). In the BPD group, a positive association between the MASC accuracy and the TBARS levels (B = 0.48; OR, 1.62; HDI95, 0.31, 0.65); and negative with SOD (B = −0.23; OR, 0.79; HDI95, −0.39, −0.08). No significant associations were found between biomarkers and the outcome for healthy control participants. Relationship loadings for the MASC accuracy are displayed in Fig. 2.

Fig. 2.

Regression loadings relating clinical factors and biomarkers with the correct responses in the MASC task by groups. Note. The posterior mean of the loading is displayed. Whiskers correspond to the 95% high density interval. The dashed vertical line represents the no-effect point (zero loading). EDR, eating disorder, restrictive subtype; EDP, eating disorder, binging-purging subtype; BPD, borderline personality disorder; HC, healthy controls.

Regarding hypermentalization (see Fig. 3), TNFα was the biomarker negatively associated in the EDR group (B = −0.25; OR, 0.78; HDI95, −0.43, −0.06). A positive association was found between hypermentalization score and JNK (B = 0.31; OR, 1.36; HDI95, 0.10, 0.51) in the EDP group. Finally, two positive relationships with this outcome were found for the BPD group: keap1 (B = 0.13; OR, 1.14; HDI95, 0.01, 0.39) and TBARS (B = 0.21; OR, 1.23, CI₉₅ = 0.03, 0.39). No significant associations were found between biomarkers and the outcome for healthy control participants. For hypomentalization (Fig. 4), a positive association was found for the EDR group between the outcome and TBARS levels (B = 0.22; OR, 1.25; HDI95, 0.03, 0.42). Negative associations were found for the BPD group in terms of SOD levels (B = −0.14; OR, 0.87; HDI95, −0.30, −0.01); and between the outcome and iNOS levels (B = −0.15; OR, 0.86; HDI95, −0.34, −0.01) for healthy control group. No significant associations were found between biomarkers and the outcome for EDP participants.

Fig. 3.

Regression loadings relating clinical factors and biomarkers with the hypermentalization index in the MASC task by groups. Note. The posterior mean of the loading is displayed. Whiskers correspond to the 95% high density interval. The dashed vertical line represents the no-effect point (zero loading). EDR, eating disorder, restrictive subtype; EDP, eating disorder, binging-purging subtype; BPD, borderline personality disorder; HC, health controls.

Fig. 4.

Regression loadings relating clinical factors and biomarkers with the hypomentalization index in the MASC task by groups. Note. The posterior mean of the loading is displayed. Whiskers correspond to the 95% high density interval. The dashed vertical line represents the no-effect point (zero loading). EDR, eating disorder, restrictive subtype; EDP, eating disorder, binging-purging subtype; BPD, borderline personality disorder; HC, health controls.

Overall, these results indicate diagnosis-specific patterns of association between inflammatory markers and social cognition outcomes. In the EDR, social cognition performance showed a pattern characterized by negative associations with markers of oxidative stress and inflammatory activation, including TBARS, COX-2, and SOD, alongside a positive association with ERK signaling. This pattern suggests that increased inflammatory burden may be associated with reduced accuracy in mental state attribution, even in the absence of overt between-group biomarker differences. In contrast, the purging ED group (EDP) exhibited a more heterogeneous association profile, with both positive and negative relationships between inflammatory markers and MASC outcomes. This variability may reflect greater clinical and biological instability in this subgroup. In the BPD group, social cognition performance was primarily associated with markers related to oxidative stress and pro-inflammatory signaling, particularly TBARS and COX-2, consistent with a heightened inflammatory state associated with social cognitive dysfunction.

Overall, these patterns suggest that immune-inflammatory markers may differentially relate to social cognitive functioning across diagnostic groups, warranting cautious interpretation within an exploratory framework.

Discussion

This study investigated the associations between social cognition performance and inflammatory and oxidative stress markers in individuals with EDs and BPD, adopting a pathway-based psychoneuroimmunological framework. The findings reveal diagnosis-specific patterns linking distinct immune–oxidative processes with different dimensions of mentalization, underscoring the importance of biological regulation for social cognitive functioning in these clinical populations.^16,38

Across diagnostic groups, the results support the notion that inflammatory and oxidative stress pathways are not uniformly associated with social cognition deficits, but instead relate to specific types of mentalization errors depending on diagnostic profile. Importantly, the present findings emphasize within-group associations rather than between-group biomarker differences, suggesting that individual variability in immune regulation may be particularly relevant for understanding social cognitive dysfunction.³⁹

Participants with BPD exhibited the most complex and dysregulated pattern of associations between biological markers and social cognition. Hypomentalization was negatively associated with superoxide dismutase (SOD), suggesting that reduced antioxidant capacity may impair the ability to accurately infer others’ mental states. In parallel, hypermentalization was positively associated with both thiobarbituric acid reactive substances (TBARS) and Kelch-like ECH-associated protein 1 (Keap1), indicating that oxidative stress and dysregulated antioxidant signaling may contribute to excessive or inaccurate mental state attribution.

This dual pattern supports a model of mentalization instability in BPD, rather than the predominance of a single type of error. Although hypermentalization has traditionally been considered a hallmark of BPD,¹⁸ the present findings suggest that individuals with BPD may oscillate between over- and under-mentalizing tendencies, potentially depending on contextual stress and underlying physiological states. Higher TBARS levels may reflect lipid peroxidation processes that are associated with individual differences in neural functioning relevant to cognition within the BPD group, whereas altered Keap1 signaling may indicate a compensatory yet insufficient activation of antioxidant defenses via the Nrf2 pathway.⁴⁰

Of note, HC displayed significantly higher Keap1 levels than participants with BPD, suggesting that the Keap1–Nrf2 system may be functionally compromised in this disorder. Within this context, the positive association between Keap1 and hypermentalization in BPD may reflect a maladaptive or dysregulated antioxidant response rather than a protective mechanism. These findings are consistent with evidence indicating that chronic immune activation in BPD has been associated with alterations in fronto-limbic networks involved in emotion regulation and theory of mind, thereby contributing to unstable interpersonal interpretations.⁴¹

The EDR group showed a distinct pattern primarily driven by oxidative stress markers. Better social cognition accuracy was associated with lower TBARS and SOD levels, whereas hypomentalization was positively associated with TBARS. This pattern suggests that lipid peroxidation processes may selectively impair mental state attribution in restrictive ED, even in the absence of pronounced inflammatory activation. Interestingly, hypermentalization in this group was negatively associated with tumor necrosis factor alpha (TNFα), indicating that lower inflammatory signaling may protect against excessive mental state attribution.

These findings support the characterization of restrictive ED as a condition marked by subtle yet persistent social cognitive vulnerabilities, potentially associated with chronic metabolic stress, emotional overcontrol, and cognitive rigidity.² Despite fewer externalizing symptoms, individuals with restrictive ED may experience difficulties accurately integrating social cues, particularly under conditions of oxidative imbalance.

In contrast, the EDP group demonstrated a more heterogeneous profile. Hypermentalization was positively associated with c-Jun N-terminal kinase (JNK) signaling, consistent with heightened stress-responsive inflammatory activation. This pattern may reflect greater emotional and biological instability in the EDP group, whereby fluctuating inflammatory states contribute to inconsistent social cognitive processing. The absence of significant associations with hypomentalization suggests that excessive, rather than deficient, mental state attribution may be more salient in purging presentations.

HC exhibited the most adaptive psychobiological profile. Notably, lower inducible nitric oxide synthase (iNOS) levels were associated with reduced hypomentalization, highlighting the potential role of nitrosative stress in modulating baseline social cognition. This finding suggests that even within non-clinical populations, immune regulation contributes to individual differences in mentalization accuracy. Reduced nitric oxide-related stress may support more stable and efficient social cognitive processing.

Overall, these findings support a psychoneuroimmunological framework in which oxidative stress, inflammatory signaling, and antioxidant regulation differentially may be involved in specific mentalization errors across diagnostic groups. Rather than reflecting uniform deficits, social cognition impairments appear to emerge from dynamic interactions between biological vulnerability and clinical phenotype.

Clinically, these results underscore the potential value of integrating immune–inflammatory markers into the assessment of social cognitive functioning. Interventions aimed at reducing oxidative stress or modulating inflammatory pathways – through pharmacological, nutritional, or lifestyle-based strategies – may enhance the effectiveness of psychotherapeutic treatments targeting interpersonal functioning in ED and BPD populations.

Finally, because of the cross-sectional and exploratory nature of the study, these associations should be interpreted with caution. Longitudinal and experimental designs will be required to clarify causal pathways and to determine whether immune modulation can directly improve social cognition outcomes.

This study presents several limitations that should be acknowledged. First, its cross-sectional design limits the ability to draw causal inferences between inflammatory markers and social cognitive performance. Longitudinal data would be necessary to determine the directionality of these associations and whether inflammatory dysregulation precedes or results from social cognitive impairment. Second, although the study includes well-characterized clinical groups, sample sizes within subgroups were modest, which may reduce statistical power and generalizability. Additionally, potential confounding factors such as medication use, illness duration, and comorbid conditions were not fully controlled and may influence both immune function and cognitive outcomes. Another limitation of the present study concerns the number of predictors included relative to the sample size of each diagnostic subgroup, which may raise concerns regarding statistical power. However, the analyses were intentionally designed with an exploratory and hypothesis-generating aim. The inclusion of multiple immune-inflammatory and oxidative stress biomarkers, together with relevant clinical covariates, reflects a pathway-based and mechanistic approach intended to capture the multidimensional nature of biological regulation in eating disorders and borderline personality disorder. Accordingly, the present findings should be interpreted as identifying coherent patterns of within-group associations that warrant replication and formal hypothesis testing in larger, independent samples.

The sample included only female participants. Although this reflects the higher prevalence of EDs and BPD among women, it limits the generalizability of the findings to male or non-binary populations. Future studies should examine whether similar psychoneuroimmunological associations are present across genders. Future studies should explicitly examine whether these associations generalize to male and gender-diverse populations.

Besides, immune-inflammatory markers are highly sensitive to external factors, including nutritional status, chronic stress, medication exposure, and lifestyle variables such as smoking.^42–44 Although cotinine levels were controlled, other factors – particularly chronic psychosocial stress and malnutrition – may have contributed to the observed immune profiles, especially in ED populations.^45,46 These influences should be considered when interpreting biomarker–cognition associations.

In this context, medication status represents an important potential confound in the interpretation of the present findings. Psychotropic medications commonly prescribed in ED and BPD populations – including antidepressants, antipsychotics, mood stabilizers, and anxiolytics – have been shown to modulate inflammatory and oxidative stress pathways, including cytokine production, oxidative balance, and intracellular signaling cascades.^35,44 As medication use was not systematically controlled for in the present analyses, it is possible that part of the observed biomarker variability reflects pharmacological effects rather than disorder-specific immune changes.

Similarly, psychiatric and medical comorbidities frequently present in these populations may independently influence immune-inflammatory markers. Chronic psychosocial stress, illness severity, and comorbid affective or anxiety disorders have been consistently associated with immune dysregulation and increased inflammatory activity.^43,46 The combined effects of chronic stress, illness burden, medication exposure, and comorbidity likely interact in complex ways, and future studies with larger samples should explicitly model these factors.

Despite these limitations, the findings have important clinical implications. Identifying specific inflammatory and oxidative biomarkers associated with social cognition deficits could inform novel treatment strategies. For example, targeting neuroinflammation through pharmacological or lifestyle interventions – such as anti-inflammatory agents or dietary modification – may represent a promising adjunct to psychotherapeutic approaches in both ED and BPD populations. Moreover, routine biomarker monitoring could aid in early identification of patients at higher risk for poor social functioning, allowing for timely and personalized intervention.

Ethical disclosures

The appropriate ethical disclosures are included in the submission and I have obtained all consents required by applicable law for the publication of any personal details or images of patients, research subjects, or other individuals included in the materials submitted to this journal. I have retained a written copy of all such consents, and I agree to provide this journal with copies of the consents or evidence that such consents have been obtained if requested.

Conflicts of interest

This work was supported by Instituto de Salud Carlos III (ISCIII) through the Fondo de Investigación Sanitaria (FIS), project FIS16, co-funded by the European Regional Development Fund (ERDF) ‘A way to make Europe’. The authors declared no conflicts of interest whatsoever.