Introduction

Difficulties in social emotional skills have been described in Anorexia Nervosa (AN) using a variety of methods [1, 2], and emotion dysregulations have been observed [3,4,5,6]. These difficulties in emotional processing are considered a maintenance factor for eating disorders [7] and may be caused by difficulties in analyzing nonverbal emotional signals [8] such as facial cues [9, 10]. However, facial emotion recognition (FER) in AN have reported heterogeneous results: several studies found difficulties in emotion recognition in patients with AN [1, 11,12,13,14,15,16], which could result from elevated levels of anxiety regarding emotional experience, leading to prolonged avoidance of salient facial cues, promoting inaccurate interpretation [17]. Yet, other studies failed to highlight any difference between patients with AN and healthy controls [18,19,20,21]. For example, it has been shown that patients with AN had similar performance in terms of number of correct answers and speed to match both facial identity and emotion compared to their healthy sisters and to unrelated controls [20]. As most trials rely on a static paradigm to assess facial emotion recognition, such discrepant results could be attributed to methodological limitations. Previous literature showed indeed that the use of static paradigm for facial emotion recognition is not associated with daily-life functioning, suggesting that it may also not be sensitive enough to discriminate patients and controls [22, 23]. In daily life, real facial emotions are not static and not always intensely expressed, but instead can often be ambiguous and/or dynamic [24, 25]. The use of a dynamic way to assess the accuracy of emotion recognition could allow a more ecologic and accurate evaluation.

Heterogeneity in the literature could also be explained by other important associated factors that have not been taken into account, such as physical activity (PA), which is recognized as being frequently associated to AN [26, 27]. Physical activity is indeed known to be involved in the regulation of negative emotions and in social integration [28]. Anxiety has also been linked to PA during undernutrition state in both clinical and rodent models [29]. Heart Rate Variability, strongly influenced by PA, has been proposed to be a biological marker of emotion recognition capacity [30]. Several studies hypothesized that physical activity and food restriction could share the similar functionality of regulating negative affect [27]. As pointed out by Kolar et al. [31], patients with AN consider that exercise could be a response to negative affect [32] and higher levels of anxiety are associated with excessive exercise [33, 34]. Engel et al. [35] also found that exercise is preceded by an increase of negative affect and followed by a decrease of negative affect.

Literature, thus, suggests that physical activity may act as a way to regulate emotions related to stressful experiences. Furthermore, acute exercising has been shown to be associated with emotion recognition among healthy participants [36]. However, to our knowledge, no studies were conducted among patients with AN (or other clinical populations); despite the role of PA and emotion deficits in AN having been previously highlighted. Therefore, one could hypothesize that in AN, PA might also influence emotion recognition capacities, and could constitute an important factor to explain differences in FER capacities. Fewer studies are available regarding non-exercise-based physical activity, although it is also increased in patients with AN and may be a key symptom of AN [37,38,39,40], and results remain inconclusive [31]. Non-exercise physical activity refers to low-intensity and daily-life activities, such as walking or fidgeting, which are mainly non-volitional and are not considered as sport activities. As pointed out by previous authors [26, 31], the use of objective assessments and Ecological Momentary Assessment is highly recommended.

The present study aims to compare FER skills between patients with AN and healthy controls using a dynamic paradigm, throughout a morphing emotional task. The second aim of this study is to explore the relationship between FER and PA, including non-exercise-based physical activity using daily-life monitoring through accelerometers. Following the idea that physical activity may constitute an inefficient strategy against negative emotions, we made the hypothesis that PA in patients with AN would be associated with a decreased capacity to recognize negative emotions.

Method

Participants and procedure

Thirty-three female outpatients diagnosed with AN according to DSM-5 criteria were recruited at a psychiatric hospital in Paris. Non-inclusion criteria were: diagnosis of other severe mental illness (e.g., schizophrenia), any chronic or life-threatening medical illness, severe visual impairment and temporary or prolonged mobility impairment, not being fluent in French or inability to provide informed consent. Thirty-three healthy controls were recruited by announcement in school and company facilities. In addition to previously cited criteria, control participants were also excluded if they had a diagnosis of eating disorder and/or if their body mass index was under the normal range for their age (18.5 kg/m2 for an adult).

After comprehensive description of the study, patients and controls (and both parents for participants under 18 years old) provided written informed consent prior to the experiment. The protocol was approved by the Ethics Committee of Cochin Hospital, Paris, France, and the study was conducted according to ethics recommendations in the Helsinki declaration [41].

Measures

Measurement of facial emotion recognition skills

Emotion recognition was assessed according to the multimorph technique as previously described [42]. It consists in a 36-trial emotion recognition computer-assisted task, presented in a random order, stimuli being taken from the Ekman and Friesen series [43]. Each trial began with a neutral face gradually morphed into one of the six-prototypical emotions—sadness, anger, happiness, disgust, surprise, and fear—across to forty 2.5% incremental stages. Each picture was presented for 500 ms followed immediately by the next morphed face in the sequence. Each trial, therefore, consisted of a 20-s continuum. In a random sequence, each emotion was presented six times, with an equivalent number of men and women faces.

Participants were asked to report whenever they considered they had recognized each emotion by clicking on a corresponding box. They were asked to wait until they had recognized the expression rather than simply guess. They were also told that they would not get any information on the accuracy of their responses, and that they could change their mind anytime if needed, by clicking on the chosen box while the faces would continue to evolve—regardless of their initial response. Participants had to finally choose one final answer at the end of each trial. A training test, including a single and equivalent emotion for all was performed, using an independent face. To determinate the participant's emotional recognition skills, the number of correct responses at the final stage (stage 40) and the number of stages needed to get the correct identification were recorded for each emotion.

Axis-I disorders, eating disorder symptomatology, anxiety and depression

Axis-I disorders including AN were assessed using the Mini International Neuropsychiatric Interview [44] and the intensity of the eating disorder was assessed using the Eating Disorder Examination Questionnaire (EDEQ) [45]. Depression and anxiety were assessed using the depression scale and anxiety scale of the Hospital Anxiety and Depression scale [46].

Assessment of physical activity

Physical activity was assessed with wrist-worn accelerometers (Vivago Wellness®, Paris, France) [47]. It allowed to continuously record physical activity (total activity counts) and energy expenditure in Metabolic equivalent (METs). The activity signal is recorded on average once per minute, the accelerometer is sensitive to movements in the 0.5–10 Hz range and reacts to omnidirectional changes in acceleration which are stored as activity counts. Data are expressed as the daytime mean in counts-min−1 [48]. Patients and controls were proposed to wear an accelerometer for 7 days. Twenty-seven patients and twenty-two control participants agreed to wear the accelerometer and were included in this sub-analysis.

Statistical analysis

Statistical analyses were performed using SPSS® statistical package for social sciences version 21.0 (IBM, Delaware, Chicago). The internal consistency in our sample of each questionnaire used in the present study has been calculated using Cronbach’s Alpha and is shown in Table 1. Data distributions were checked using the Kolmogorov–Smirnov test prior to analyses. For each emotion, group differences regarding the number of stages needed until a correct answer have been calculated using a repeated-measure mixed ANOVA method, trial (1–6) for each emotion being the within-subjects factor and group (patients versus controls) being the between-subjects factor. For the number of accurate recognitions at final stage when merging all trials and for physical activity, group differences have been calculated using student-t test. Effect sizes were calculated using Cohen’s d. While controlling for depression, regression analyses have been used to investigate the effect of AN (vs control group) on FER. Finally, path analyses were conducted using the PROCESS macro for SPSS, to test for the mediating role of depression in the relation between the group (AN vs controls) and emotion recognition. Correlation analyses have been used to investigate the relationship between FER and PA or depressive symptomatology, in patients versus controls.

Table 1 Sociodemographic and clinical characteristics for patients with Anorexia Nervosa and controls
Full size table

Results

Sociodemographic and clinical characteristics

Sociodemographic and clinical characteristics are presented in Table 1. All participants except one were native European, and all participants were females. Participants’ age range was between 15 and 46 years. Mean age was 25.03 (SD = 7.04) for patients and 26.27 (SD = 6.28) for controls, and groups did not differ according to age (p = 0.452). The majority of participants were singles, for patients (93%) and controls (63%). Body Mass Index (BMI) was 16.34 (SD = 1.30) for patients and 20.84 (SD = 1.98) for controls. Among patients, the mean duration of illness was 7.85 years (SD = 7.04), the mean age of onset was 17.18 (SD = 3.38), and 67% were currently receiving daily psychiatric drugs. Patients had significantly higher scores on the depression (p = 0.002) and anxiety (p = 0.012) sub-scales, and had higher scores for all dimensions of the EDEQ scale (p < 0.001).

Facial emotion recognition

Earliest recognition of facial affect

For each emotion, no significant difference between patients and controls was observed regarding the number of stages until correct answer (see Table 2). Therefore, patients with AN were as fast as controls to recognize facial affect.

Table 2 Average number of stages until correct answer and number of accurate emotional recognition at final stage: differences between patients with AN and controls
Full size table

Accuracy of emotion recognition

Regarding the accuracy of recognition for each of the targeted emotions, patients correctly recognized disgust more frequently than controls (p = 0.018). Patients recognized more frequently sadness and overall emotions, although this was only a trend (p < 0.10). The importance of effect size can be considered medium for disgust (d = 0.60), and small for sadness (d = 0.47) and overall emotions (d = 0.44). No significant difference was found for anger, joy, surprise, and fear (see Table 2).

Accuracy in recognizing disgust (R2 = 0.85, β = − 0.29, t = − 2.43, p = 0.018) and sadness (R2 = 0.55, β = − 0.24, t = − 1.94, p = 0.057) was predicted by being in the AN group (vs. control group). But after controlling for depression, this was only a trend (β = − 0.23, t = − 1.78, p = 0.080 for disgust and β = − 0.25, t = − 1.90, p = 0.062 for sadness). No multicollinearities were detected when conducting a collinearity diagnosis (VIF < 3).

Therefore, regression analyses suggested that depression might account for at least some of the relationship between AN diagnosis and emotion recognition accuracy. However, the bootstrapping indirect effect of group (AN vs controls) on disgust (CI 95% [(− 0.39)–(0.06)]) and sadness (CI 95% [(− 0.19)–(0.33)]) recognition through depressive symptomatology was not significant (see Fig. 1).

Fig. 1
figure 1

Depressive symptoms (HAD-S) as a mediator of Anorexia Nervosa (vs. control group) and the number of recognition of disgust and sadness morphing faces

Full size image

Anxiety, depressive symptomatology and facial emotion recognition

The speed at which disgust was recognized by patients was significantly correlated with depression (r = − 0.39, p = 0.025,), as was the level of accuracy (r = 0.34, p = 0.05), while no such link was detected among controls. Anxiety was not associated with any of the emotions, in the clinical group and in the control group (p > 0.05) (see Table 3).

Table 3 Correlations (pearson r) between facial emotion recognition, physical activity, anxiety and depressive symptoms among patients with Anorexia Nervosa and controls participants
Full size table

Physical activity and facial emotion recognition

Physical Activity (PA) monitored during 7 days did not statistically differ in the two sub-groups in which participants wore accelerometers (see Table 1). Among patients, the number of accurate recognitions for sadness was negatively associated with the level of physical activity (r = − 0.40, p = 0.037), and not the other emotions (p > 0.05), including disgust (p = 0.956), while no association was detected among controls. Considering that physical activity did not differ between patients and controls and was not associate with depressive symptomatology, it did not appeared relevant to conduct mediation analyses including PA.

Discussion

Patients did not appear to experience difficulties in recognizing basic facial emotions and may have an increased sensitivity to specific facial emotions, especially regarding disgust and, to a lesser extent, sadness. This result is interesting considering that disgust is an emotion related to smell and taste, involved in food, which could be associated with eating disorder [49, 50]. Moreover, among patients, we detected that higher sensitivity to disgust is correlated with depressive score, an association which is not observed in healthy participants. Our study, thus, supports the hypothesis that disgust sensitivity could be an aspect of depressive symptomatology in AN, and can be detected through dynamic and ecological facial recognition. Further studies may help to determine if this result reflects a higher emotional sensitivity or rather an attentional bias toward disgust. Negative attentional bias have indeed been previously highlighted among patients with AN [51, 52], including attentional bias towards faces [53,54,55]. Therefore, the present studies may support the idea of a negative attentional bias toward disgust, increasing depressive symptoms in AN.

Disgust sensitivity in AN has been seldom addressed in the literature, and previous studies showed heterogeneous results [49]. In fact, several studies on disgust recognition found no difference between patients with AN and controls, and, contrary to our results, other studies observed that patients were less accurate in disgust recognition [25, 56, 57]. This points out the necessity to keep investigating disgust sensitivity and recognition in anorexia nervosa, for example, by conducting subgroup analyses for AN subtypes to further understand such discrepancies.

While being surprising regarding the literature, results remain coherent with the difficulties in regulating negative affect that have been highlighted in AN [5, 6]. We could hypothesize that the relational difficulties described in AN [58] may not be due to altered FER capacities but rather to later stages of emotional processing, such as misinterpretation of recognized emotions under social and complex circumstances [21]. These results appear contradictory with previous studies, which concluded to emotional cognition impairment in AN [11,12,13, 15, 16, 57]. However, as previously pointed out, prior researches examining FER in AN mostly used static paradigms and the results were heterogeneous [18,19,20]. In paradigms using static images, patients with AN appear to have more difficulties in FER than in more dynamic paradigms, where their performances were similar to controls [21]. This highlights that heterogeneity of results can depend on the type of evaluation used, and confirms the necessity to expand FER exploration using more dynamic and ecological evaluations.

However, the use of a dynamic paradigm does not appear to fully explain the heterogeneous results in previous literature, since other studies using multimorphing in eating disorders report no difference between patients and controls participants [59]. Heterogeneity of these results could also be explained by other factors that have not necessarily been taken into account in previous studies. For example, in our sample, PA appears to be involved with less ability to recognize sadness, this being observed only in patients. In coherence with previous studies [27, 31], we could hypothesize that PA may be a way to regulate negative affect; thereby decreasing the ability of emotion recognition. The fact that the relation between PA and decreased sensitivity to sadness is not highlighted among controls appears particularly interesting in supporting the idea that PA can be considered as an emotional regulation strategy specific to patients with AN. These results contribute to recent findings [31] which showed that PA was associated with less negative affect in adolescents with AN only, but not in the control group. Further studies are needed to explore if AN patients with higher PA use PA to avoid intense emotional stimuli in their daily relational functioning.

It is also important to point out that other factors which were not investigated in this study may also be involved in the heterogeneous results regarding AN and FER, such as alexithymia or attention to faces. For example, alexithymia has been suggested to explain emotion processing deficits in eating disorders, and difficulties in recognizing emotions among patients with AN has been suggested to be due to reduced attention to faces and particularly the eye region [60]. Therefore, it appears necessary to reproduce our results while controlling for more factors.

This study allowed us to draw hypotheses regarding on one hand the sensitivity of patients with AN to disgust and sadness (which could be related to social cognition impairment and difficulties to regulate negative affects), and on the other hand the potential role of depressive symptomatology in the bias toward disgust. Results are also promising regarding the potential link of PA with FER, and highlight the necessity to better comprehend the variability of FER in AN. These data, although based on a limited sample, strongly encourages to keep investigating facial emotion recognition in AN using dynamic paradigms.

Finally, results also lead to the hypothesis that physical activity could influence the relationship between AN and FER by decreasing FER sensitivity to negative affect such as sadness, and suggest that FER impairment sometimes found in patients with AN could be due to high levels of physical activity rather than AN disorder per se. To support this hypothesis, further studies should investigate FER differences between AN patients with or without excessive physical activity.

Strengths and limits

The number of participants enrolled was relatively small, and exclusively composed by women. The absence of a consensual definition of excessive physical exercise [61] prevented us to precisely explore the relationship between PA and FER. Moreover, some patients refused to wear accelerometers, these patients being potentially more at risk of excessive physical exercise. Besides, the influence of pharmacologic treatment on FER could not be analyzed because of the limited size of the sample, although previous studies on FER [42, 62] described no effect of medication on this type of dynamic paradigms. Finally, although this study aims to increase the accuracy of FER assessment using a more stronger and ecological dynamic paradigm through multimorphing, it is important to note that this cannot capture the complexity of real-life emotions which are influenced by the context of recognition and interpretation [63]. Conversely to daily life, multimorphing lacks the context in which emotions are interpreted, especially regarding social contextualisation which could play an even important role in AN due to the social difficulties that have been highlighted in this population [1, 2].

This study also has several strengths. First of all, it is, to our knowledge, the first study exploring the relationship between FER skills and PA. Second, the use of multimorph technique and Ecological Momentary Assessment (through continuously recording PA during 7 days) provides an objective, ecological and dynamic assessment. It is especially important to mention that monitoring non-exercise-based physical activity in clinical samples remains a challenge, and valid clinical data were missing in previous literature; even though this may be a key symptom in AN [38] associated with poor outcomes and a longer duration of treatment [28, 39, 40]. The present methodology, therefore, contributes to this literature by providing ecological findings on non-physical activity in AN. Finally, our protocol includes both clinical and control participants, allowing us to highlight that the relationship between FER and PA differs in patients with AN (compared to healthy participants).

What is already known on this subject?

Difficulties in social emotional skills have been previously highlighted in AN, but results were heterogeneous on FER: some studies suggest that patients with AN had impaired FER abilities [1], others failed to highlight any difference between patients with AN and controls [18].

Moreover, previous studies suggest that PA may act as an emotional regulation strategy in patients with AN [27], but non-exercise-based physical activity was far less studied and there were no data available on the relationship between PA and FER.

What this study adds?

This study highlights that patients with AN are capable of recognizing facial emotions as much as healthy participants, but could have higher sensitivity toward disgust and sadness. This study is also the first to show that higher levels of PA were associated with a decreased ability to recognize sadness, but only in patients with AN.