Memory Performance in Adolescents Living With Obesity: Brain Activation During Encoding and Retrieval in a Case‐Control Study

ABSTRACT

Objective

Effects of obesity on brain health have been revealed in adults, including mental health effects and cognitive impairment. Regarding cognition, obesity‐related memory impairment has been specifically described. While this effect could have a major impact on learning abilities during adolescence, few studies have considered this critical period.

Methods

In this present study, a new fMRI memory task based on paired encoding of faces and backgrounds and subsequent face recognition was presented to male adolescents living with obesity (N = 11) and their lean counterparts (N = 15).

Results

Our study shows that adolescents living with obesity exhibited significantly lower face recognition memory performances (F_(1,72) = 9.84, p = 0.002) than their lean counterparts coupled with altered functional cerebral activation patterns during the encoding and retrieval phases of the task. More specifically, during the encoding of the task, a hypoactivation of the right hippocampus and the parahippocampal gyrus was identified in adolescents living with obesity and during the retrieval a hyperactivation of the precuneus (Z > 2.3, cluster‐corrected p = 0.05).

Conclusions

These results suggest that obesity during adolescence is associated with neurocognitive impairment. Future studies should consider adolescence more carefully since this memory impairment could contribute to academic learning difficulties faced by adolescents living with obesity.

Trial Registration

CHUBX 2017/19; CPP number: 2017‐3A02533‐50

1. Introduction

Besides obesity's effects on cardiovascular and metabolic function, obesity has been associated with brain health. In the past decade, a growing number of studies [1, 2, 3, 4, 5] have assessed the impact of obesity on brain and cognitive functions during childhood and adolescence, which are critical periods of brain and cognitive function development. Most of these studies focused on executive functions and the extent to which they could account for this population's food intake disturbances (for review see [6]). A recent meta‐analysis indicates that poorer executive function abilities are associated with higher levels of disinhibited eating in children and adolescents living with obesity [7]. While pediatric obesity is also associated with poorer academic performance [1], more memory complaints [2] and memory impairment [8, 9], less is known about the associated cerebral network dysfunction when obesity is not associated with comorbidities such as diabetes mellitus and when the memory task does not involve food‐related items.

Animal models have demonstrated that the consequences of obesogenic diet intake on memory depend on the age of exposure and are more deleterious when it starts during the juvenile period [10, 11]. Pediatric obesity has consequences on the structural and functional integrity of several cerebral regions, including some involved in memory [3, 12, 13, 14, 15, 16]. Among these alterations, obesity in adolescents seems to present a functional and structural brain fingerprint as their features in the orbitofrontal cortex, the dorsolateral prefrontal cortex, and the posterior cingulate cortex can be used to predict the body mass index (BMI) [17]. Moreover, studies have shown evidence of lower gray matter volumes of memory‐related brain areas such as the hippocampus and the medial prefrontal, frontal and anterior cingulate cortices [18, 19, 20, 21].

There are different memory processes with distinct neural processes, and only a few studies have investigated the neural correlates of relational/episodic memory among adolescents or preadolescents living with obesity. They showed lower performances in episodic memory using tasks involving either complex scenes or paired presentations of creatures and backgrounds [22, 23]. In addition to these memory alterations, decreased engagement of the hippocampus and parahippocampal gyrus during the encoding of an episodic memory task has been reported in adolescents living with obesity [22].

A subcomponent of these relational and episodic memory tasks was reassessed using conditions mimicking encountering strangers in new environments modeled by the superposition of faces presenting different expressions (neutral, sad or happy) on background scenes of everyday life. This task was designed to be performed in the scanner during the encoding and retrieval phases. Since this task evaluates adolescents' cognitive performance, special care was given to the material and conditions applied during the assessment to limit stress. Faces were chosen rather than words to avoid the confounding effects of reading proficiency and socioeconomic status, potential biases impacting cognitive studies on obesity. Moreover, implicit encoding was chosen to limit the stress commonly brought by explicit encoding tasks. Our hypothesis is that adolescents living with obesity would present lower performances than their lean counterparts in this novel memory task associated with brain activation changes during encoding and/or retrieval of the task.

2. Methods

2.1. Participants

Thirty male right‐handed adolescents were initially recruited in this study. Adolescents living with obesity were enrolled from the pediatric endocrinology department of Bordeaux’ Hospital Pellegrin. The lean subjects were recruited through public advertising. This case‐control study procedure was approved by a national human research review board and all participants and their parents provided written informed consent.

Fifteen participants with obesity and fifteen age‐matched control adolescents were recruited. However, only eleven participants with obesity were included in the whole study due to the quality of the MRIs. Right‐handed boys, between 9 and 17 years old, were selected based on their body mass index (BMI) compared to the norms established by the BMI‐International obesity task force. Accordingly, subjects with a BMI superior to the IOTF‐30 curve are considered adolescents living with obesity. Subjects had no other diagnosed neuropsychiatric or chronic disease such as diabetes and no history of psychotropic drug consumption.

2.2. Experimental Design

The protocol started with a clinical examination of the participants at 3:00 p.m. (GMT+1), followed by an MRI session at 3:30 p.m. for participants to have similar hunger states. A snack chosen by the parent present was given after the MRI session and before the final cognitive assessment using the Cambridge Automated Battery [24, 25].

2.2.1. Clinical Examination

During an initial clinical examination, anthropomorphic data were acquired: participants were weighed using a Body Composition Analyzer (Tanita SC‐240 MA) and their height was measured. The body mass index (BMI) was then calculated and the body fat mass (%) and the fat‐free mass (%) were recorded. The puberty stage of the participants was assessed using Tanner stages [26]. Additionally, participants filled the Adolescent Depression Rating Scale (ADRS) [27] and the Revised Children's Manifest Anxiety Scale (R‐CMAS) [28]. For the ADRS, a cut‐off score of 3 was used to determine depressive mood status. They also scored their sleep quality on the two previous nights using item 6 of the Pittsburgh Sleep Questionnaire Index [29]. Based on a parental interview, the family social deprivation score was estimated using the EPICES [30, 31], and a cut‐off score of 30 was used to determine precarity status.

2.2.2. Magnetic Resonance Imaging (MRI) Session

2.2.2.1. MRI Acquisition

MRI sequences were acquired with a 3T SIEMENS PRISMA MRI scanner equipped with a 64‐channel head coil (Siemens Healthcare, Erlangen, Germany) at the bioimaging platform of the University of Bordeaux, France.

The MRI protocol started with a whole‐brain 3D T1‐weighted MPRAGE sequence with the following parameters: TR: 24000 ms; TE = 2.21 ms; TI = 1000 ms; Flip angle: 12°; Field of view: 256 mm × 256 mm, voxel size: 1.0 × 1.0 × 1.0 mm.

Functional scans were acquired using a single‐shot multi‐band gradient‐echo EPI sequence with a whole‐brain resolution of 2.5 mm³ (TR = 700 ms, TE = 30 ms, Flip angle: 53°; multi‐band factor 6; 60 interleaved slices of 2.5 mm; Field of view: 210 × 210 mm). The functional session was acquired according to the following protocol: i/resting state 1, ii/ENCODING phase (9 min), iii/resting state 2, iv/RETRIEVAL phase (8 min). For the present study, only task‐based fMRI data were used.

The total acquisition time of the protocol was about 45 min.

While MRI sessions were systematically conducted between 2 and 4 p.m., participants' satiety scores were controlled before getting in the scanner.

2.2.2.2. Encoding and Retrieval Task Paradigm

An event‐related functional MRI (fMRI) task called MEET‐UP was developed to assess the abilities of adolescents to associate novel faces with different backgrounds, such as meeting strangers in various environments. This task was implemented and ran on the software E‐prime 3.0 (Psychology Software Tools Inc.).

During the encoding phase, subjects were shown a series of 48 different trials comprising a background image of an every‐day life situation (airport's halls, bridges…) for a duration of 5 s, followed by an image of an unknown male or female portrait from the Karolinska Directed Emotional Faces database (http://www.emotionlab.se/resources/kdef) superposed onto the background for the same duration (Figure 1). Three different facial expressions were depicted on these faces (16 images per facial expression): neutral, sad, or happy conditions. A variable interval of time (ISI randomly chosen between 500, 1000 or 1500 ms) was set between two subsequent trials. Questions regarding the identification of the situation depicted on the background image (work, transport, holidays) and of the facial expression (happy, neutral, sad) were asked to bring the subject's attention to the images. Answers were given using a 3‐button response keyboard in the scanner. Mean response rate, facial expression identification and mean reaction time were collected from this encoding step and compared among groups.

FIGURE 1.

Description of the MEET‐UP task. This task consisted of two phases: the encoding phase and 7 minutes later, the retrieval phase, all performed inside the MRI scanner. During the implicit encoding phase, subjects were shown a picture of a scene used as a background on which a portrait was then superposed. Two questions were asked during the trials to bring the subject's attention first to the background (‘Is it a scene of transport, work or holiday?’) and then to the faces. Regarding the latter, three types of facial expressions were presented (neutral, sad or happy) and subjects had to identify the expression during the image visualization. This was performed for 48 trials (16 per type of facial expression) in an event‐related design. During the retrieval phase, an already seen background was displayed, on which three neutral faces were superposed. Subjects had to identify which of the three faces was already presented.

After seven minutes of resting state delay, subjects had 48 test trials during which they were presented with a different background for each trial and then had to retrieve the face previously paired with this background amongst 3 neutral faces (Figure 1). In the background step (maximal duration of 4 s), they were asked if they remembered this background and had to answer with the corresponding yes/no/maybe button. In a second step (maximal duration of 4 s), they were asked to pick the face previously associated with the background. All target faces were presented with a neutral expression during retrieval. A variable interval of time (ISI randomly chosen between 400, 500 or 600 ms) was set between two subsequent trials. The mean parameters (response rate, accuracy, reaction time) were computed for each participant from the retrieval step of the task.

The number of male and female faces was counterbalanced during the 2 phases (24 trials male faces and 24 trials female faces).

2.2.2.3. MRI Preprocessing

Task functional images were preprocessed using FEAT (fMRI Expert Analysis Tool), part of FSL (FMRIB's Software Library, v.6.0). These steps included spatial smoothing using a Gaussian kernel of FWHM = 5.0 mm, multiplicative mean intensity normalization of the volume at each time point; high‐pass temporal filtering (with a 50.0 s cutoff), normalization of the functional images to the standard MNI space via a two‐step process (i) co‐registration of the functional images to the high‐resolution T1‐weighted scan, using FLIRT and the BBR (Boundary‐Based Registration) cost function with motion correction using MCFLIRT (ii) registration of the structural T1 images to the MNI space with ANTs (Advanced Normalization Tools). Images were visually checked by MRI experts for quality control.

2.3. Statistical Analyses

2.3.1. Demographic and Behavioral Analyses

Descriptive statistics for the demographic, clinical and neuropsychological data were computed. Between‐group comparisons of dimensional variables were performed using unpaired Student t‐tests or Mann‐Whitney tests according to the normality of the distribution of the variable of interest. For categorical variables, Pearson's chi‐square tests were performed.

To investigate if facial expressions were impacting behavioral variables of patients with obesity differently in this task in comparison to lean individuals, the following statistical models were used. For behavioral parameters of the retrieval (Response Rate, Accuracy, Reaction time), two‐factor ANOVAs (Group, Facial expression) were performed. To compare the accuracy between groups considering the potential confounding effect linked to the reaction time, an ANCOVA for accuracy with reaction time as a covariate was performed.

A significant threshold was considered for a p‐value < 0.05 (JAMOVI v2.3.24, running on the statistical language R [32]). When appropriate, Bayesian approach was applied to determine whether the just below significance results reflect evidence for the null hypothesis or insufficient power.

2.3.2. Neuroimaging Analyses

2.3.2.1. Whole Brain Analyses

To determine individual activity patterns per facial expression during encoding and retrieval of this new task, an event‐related design was used for the three facial expression conditions divided into successfully encoded images (Hits) or not recognized (Miss), creating 6 separate events at the first level analysis. Each event is modeled with the onset and duration trials convolved with a double‐gamma response function adding temporal derivatives and temporal filtering. At this level, Z‐statistic (Gaussianised T) images were thresholded using clusters determined by Z > 2.3 and a corrected cluster significance threshold of p = 0.05 for each contrast [33].

At the second level analysis, the main effect of the different facial expression conditions and their differences were computed within each group using FLAME1 to identify the activation patterns of the encoding in lean Adolescents and Adolescents living with obesity. Secondly, a Subsequent Memory Effect contrast was computed to assess the differences between the activation patterns of successful encoding versus unsuccessful trials: Hits > Miss according to the facial expression conditions (Supporting Information S1: Analysis 1). Finally, as no differences in all the previous contrasts (Facial expression conditions or Hit/Miss conditions) between groups were observed, all trials were pooled. The retrieval phase was also analyzed following these different steps.

To ensure potential functional modifications did not result from partial volume effects due to lower gray matter volumes in adolescents living with obesity, a voxel‐based morphometry analysis was conducted on gray matter volume maps (t‐test between groups in SPM12, corrected for multiple comparisons) with the total intracranial volume and age as covariates (Supporting Information S1: Analysis 2).

2.3.2.2. Region‐of‐Interest Analyses

Confirmatory regions‐of‐interest (ROI) analyses were also performed on the unthresholded maps acquired during the encoding and retrieval phases. Masks of the hippocampus and the parahippocampal gyrus were created from the probabilistic atlas Brainnetome [34] with the FSL toolbox for each hemisphere. Whole‐brain unthresholded zstat maps of first level FEAT, for each individual, of the analyses of each phase were taken to extract the maximum z‐score value in specific masks with the function fslquery from the FSL toolbox. Comparisons between groups of this level of activation in specific regions were done with the use of Student t‐tests (significant threshold set at p < 0.05).

2.3.2.3. Correlation Between Memory Performances and Neuroimaging Results

To test the association between memory performances and neuroimaging findings, individual z‐score maximum values were extracted with the function fslquery inside the precuneus from the specific cluster peak obtained in whole‐brain group comparisons.

A sphere of 5 mm on the peak of the whole brain results was used. Since assumptions of normality were reached for the 2 variables according to the Shapiro‐Wilk test, Pearson's correlations between behavioral variables and levels of activation were considered significant for a threshold of p < 0.05.

3. Results

3.1. Demographic and Clinical Results

Descriptive and comparative statistics of demographic and clinical variables are presented in Table 1.

TABLE 1.

Participants description: Between‐group comparisons of the demographic and clinical data.

	Lean adolescents (N = 15)	Adolescents with obesity (N = 11)	P
Age (years), mean ± SD	14.07 ± 1.44	14.55 ± 1.57	0.42
Weight (kg), mean ± SD	53.95 ± 10.35	99.75 ± 24.70	< 0.001***
Height (cm), mean ± SD	169.5 ± 10.7	171.1 ± 9.9	0.69
BMI, mean ± SD	18.80 ± 3.37	33.66 ± 5.47	< 0.001***
Fat mass, mean ± SD	10.81 ± 3.57	44.77 ± 8.88	< 0.001***
Lean mass, mean ± SD	84.73 ± 13.78	54.24 ± 8.99	< 0.001***
Tanner's stage, genitals, median (IQR)	4 (1.0)	3 (1.0)	0.72
Tanner's stage, pubic hairs, median (IQR)	4 (1.0)	3 (1.0)	0.49
Sleep quality, median (IQR)	2 (1.0)	2 (1.0)	0.26
EPICES score (min‐max)	0–20.1	0–62.1	—
EPICES score > 30, N (%)	0 (0)	4 (36.4)	0.04*
ADRS score > = 4, N (%)	2 (13.3)	3 (27.3)	0.37
R‐CMAS, median (IQR)	7 (6.5)	15 (10)	0.08

Note: Variables were presented as the mean value ± the standard deviation or the median (InterQuartile Range or IQR) in each group. Student t‐tests, Mann‐Whitney U tests and Pearson's Chi‐square tests were performed as appropriate: ***p < 0.001. *p < 0.05.

Abbreviations: ADRS, Adolescent Depression Rating Scale; R‐CMAS, Revised Children's Manifest Anxiety Scale. EPICES Social deprivation score.

The two groups did not differ for age, height or pubertal stage (genitals or pubic hairs; all p > 0.4), but they differed as expected for weight, BMI, fat mass, and body fat‐free mass (all p < 0.001). The two groups reported similar levels of sleep quality (Mann‐Whitney U test, U = 51.5, p = 0.26, effect size r = 0.23) and no significant difference in anxiety levels (Mann‐Whitney U test on R‐CMAS scores, U = 48.5, p = 0.08, effect size r = 0.34, the Bayes Factor BF₁₀ is of 1.09, indicating that the data are equally likely under the alternative hypothesis and the null hypothesis) and did not differ for depressive mood status based on the ADRS cut‐off score (χ ² Test, χ ² = 0.79, df = 1, p = 0.37; effect size: φ = 0.17). There was a significantly greater proportion of socially deprived families in the group with obesity based on the EPICES cut‐off score (X‐squared = 3.95, df = 1, p = 0.04, effect size: Cramér's V = 0.49).

3.2. Behavioral Results

During the encoding phase, no group effect in facial expression identification was observed when the response rate was taken into account (F_(1,23) = 2.25, p = 0.12, partial η ² = 0.06 for more detail see the Supporting Information S1: Figure 1). No group effect in reaction time was observed during the encoding phase (F_(2,72) = 0.07, p = 0.79, Supporting Information S1: Figure 2).

During the retrieval phase, the mean response rate for all facial expressions was 91.4 ± 1.7% for the control group and 90.1 ± 1.7% for the group with obesity (Supporting Information S1: Figure 3A). The two‐factor ANOVA on the response rate revealed no effect of Group (F_(1,72) = 0.56, p = 0.45, η ² = 0.007), Facial expression conditions (F_(2,72) = 1.49, p = 0.23, η ² = 0.03) or Group × Facial expression interaction (F_(2,72) = 2.87, p = 0.06, η ² = 0.07, Figure 2A). The Bayesian model predicting the response rate of the retrieval phase including group (BF_inclusion of 0.26), Facial Expression (BF_inclusion of 0.30) and Group × Facial Expression interaction (BF_inclusion = 0.31) presents a BF₁₀ of 0.13 in favor of the null hypothesis that confirms the frequentist analysis.

FIGURE 2.

Behavioral results of the retrieval phase, the mean response rate (A) and mean accuracy (B) according to the facial expression and the group. Two‐factor ANOVAs were computed for each variable and significant results were indicated by ** and ##. (A) No effect of group (F_(1,72) = 0.56, p = 0.45) or facial expression condition (F_(2,72) = 1.49, p = 0.23) was observed on the response rate. (B) A group difference was identified in accuracy per facial expression (F_(1,72) = 9.84, **p = 0.002), with an effect of the facial expression conditions (F_(2,72) = 6.23, ##: p = 0.003), which was not different according to the group (interaction group × facial expression: F_(2,72) = 0.08, p = 0.92). Post hoc analyses on the facial expressions indicated a significant difference in accuracy between sad and neutral trials (Tukey Test, p = 0.002).

Accuracy for face recognition during retrieval was 75.1 ± 2.7% for the control group and 64.4 ± 5.7% for the group with obesity (Supporting Information S1: Figure 3B), all performances being well above the chance level (33%). The two‐factor ANOVA on accuracy revealed main effects of Group (F_(1,72) = 9.84, p = 0.002, η _p ² = 0.10) and of the Facial expression conditions (F_(2,72) = 6.23, p = 0.003, η _p ² = 0.13) were observed, with no significant Group × Facial expression interaction (F_(2,72) = 0.08, p = 0.92, η _p ² = 0.002, Figure 2B). No group differences were observed in the response rate (p = 0.31, Cohen's d = 0.19, Supporting Information S1: Figure 3A), but with an a priori hypothesis, the group with obesity had significantly fewer correct answers than the control group (One‐sided t‐test, p = 0.03; Cohen's d = 0.73, Supporting Information S1: Figure 3B).

Regarding reaction times, the group with obesity responded significantly faster than the control group (two‐factor ANOVA, group effect: F_(1,72) = 12.20, p < 0.001, η _p ² = 0.14) without an effect of facial expression (F_(2,72) = 1.03, p = 0.36, η _p ² = 0.02) or interaction between the two factors (F_(2,72) = 0.37, p = 0.69, η _p ² = 0.009, Supporting Information S1: Figure 4). Therefore, to take into account the confounding effect of the reaction time on the accuracy, an ANCOVA comparing the two groups for the facial recognition accuracy per facial emotion has been run using the reaction time as the covariate; in this analysis, the mean accuracy of the group with obesity remained significantly lower than the accuracy of the control group (F_(1,71) = 5.6, p = 0.02, η _p ² = 0.06).

3.3. Neuroimaging Results

3.3.1. During the Encoding Phase

Whole brain analyses in FSL FEAT of the pattern of activated brain regions during the encoding phase of the task revealed a significant bilateral activation of the hippocampus, the temporal and occipital fusiform gyrus, the precuneus, occipital regions, and orbitofrontal cortex in lean subjects (Figure 3A) and in subjects with obesity (Figure 3B) for Z > 2.3 and a corrected cluster significance threshold of p = 0.05.

FIGURE 3.

Pattern of activated regions during the encoding phase in lean subjects (A) and in adolescents living with obesity (B). (A) Activated regions during the encoding phase of the face recognition task in lean adolescents. Results from FEAT analyses in lean adolescents (N = 15) in blue. Z (Gaussianised T) statistic images were thresholded using clusters determined by Z > 2.3 and a corrected cluster significance threshold of p = 0.05. (B) Activated regions during the encoding phase of the face recognition task in adolescents living with obesity. Results from FEAT analyses in adolescents living with obesity (N = 11) in orange. Z (Gaussianised T) statistic images were thresholded using clusters determined by Z > 2.3 and a corrected cluster significance threshold of p = 0.05.

Contrasts on the different levels of activation between the two groups in all trials indicated that the group living with obesity exhibited a general lower activation of the engaged brain areas, and more precisely of the right hippocampus, parahippocampal gyrus, and a higher activation of the cerebellum in the left VIIIa and b and the left VII b (Figure 4A–B and Supporting Information S1: Table 1, Z > 2.3, corrected cluster significance threshold of p = 0.05). Subsequent memory effect analysis revealed no significant group differences during encoding (Supporting Information S1: Analysis 1).

FIGURE 4.

(A) Differences in the levels of activation during the encoding phase between lean subjects and adolescents living with obesity. (A–B) Results from FEAT analyses in the contrast Lean > With Obesity (A) and Lean < With Obesity (B). Z (Gaussianised T) statistic images were thresholded using clusters determined by Z > 2.3 and a corrected cluster significance threshold of p = 0.05. (C) Differences in the levels of activation in ROI analyses during the encoding phase in masks created from the atlas Brainnetome. The light blue masks correspond to the hippocampi and the purple masks correspond to the posterior parahippocampal gyri. Mean peak z‐score values are presented with the standard error of the mean (+ SEM). Student t‐test one‐tailed, *p < 0.05, ***< 0.001.

Further analyses based on regions‐of‐interest for the right and left hippocampi and posterior parahippocampal gyri confirmed the lower activation in the right hemisphere for the group living with obesity (Figure 4C, Student unpaired t‐tests: right hippocampus p = 0.004; right parahippocampal gyrus p = 0.04). No significant correlation was found between recognition performances and the levels of activation in this phase (Pearson's correlation, r = 0.27, p = 0.18).

3.3.2. During the Retrieval Phase

The pattern of activated brain regions during the retrieval phase of the task indicated a significant bilateral activation of the temporal and occipital fusiform gyri, the precuneus and occipital regions in the control group (Figure 5A, Z > 2.3 and a corrected cluster significance threshold of p = 0.05). Similar, but less extended activated regions were found in the group living with obesity (Figure 5B).

FIGURE 5.

Pattern of activated regions during the retrieval in lean subjects (A) and in adolescents living with obesity (B). (A) Activated regions during the retrieval phase of the face recognition task in lean adolescents. Results from FEAT analyses in the lean adolescents (N = 15) in blue: Z (Gaussianised T) statistic images were thresholded using clusters determined by Z > 2.3 and a corrected cluster significance threshold of p = 0.05. (B) Activated regions during the retrieval phase of the face recognition task in adolescents living with obesity. Results from FEAT analyses in adolescents living with obesity (N = 11) in orange: Z (Gaussianised T) statistic images were thresholded using clusters determined by Z > 2.3 and a corrected cluster significance threshold of p = 0.05.

Contrasts on the differences between groups in all trials indicated that the group living with obesity exhibited a higher level of activation in the precuneus (Figure 6A and Supporting Information S1: Table 2) and no regions with a lower activation pattern than the control group (Figure 6B), with Z > 2.3 and a corrected cluster significance threshold of p = 0.05 for both analyses.

FIGURE 6.

Differences in the levels of activation during the retrieval between the lean adolescents and the adolescents living with obesity and extraction of the level of activation and its link with the face recognition performance. (A–B) Results from FEAT analyses in the contrasts Lean > With Obesity (A) and Lean < With Obesity (B): Z (Gaussianised T) statistic images were thresholded using clusters determined by Z > 2.3 and a corrected cluster significance threshold of p = 0.05. (C) Differences in the levels of activation in ROI analyses for the precuneus: Extracted mean peak z‐score values in a mask of a 5 mm sphere around the local peak of the precuneus, in the cluster extracted from the whole‐brain results, between‐group verification of the comparisons with a Student t‐test one‐tailed and Pearson's correlation plot between mean z‐score values of the precuneus mask and the face recognition accuracy. Mean peak z‐score values are presented with the standard error of the mean (+ SEM), *p < 0.05.

In the whole cohort of adolescents (N = 26), this higher activation was negatively correlated with the face recognition performances: in other words, the higher the activation of the precuneus during retrieval, the lower the accuracy (Figure 6C, Pearson's correlation r = −0.39, p = 0.04).

4. Discussion

This neuroimaging study uncovered lower memory performances coupled with an alteration of the neural pattern of activation during both the encoding and the retrieval phases of the task in adolescents presenting obesity with no associated diabetes compared to lean age‐matched controls. During the encoding phase, the right hippocampus and parahippocampal gyrus were significantly less activated in adolescents living with obesity than in lean controls. Another substantial finding of our study was the higher recruitment of the precuneus during face recognition among the adolescents living with obesity.

A new fMRI task, MEET‐UP, was developed based on implicit paired encoding of faces and backgrounds, a memory process commonly used in everyday‐life that was adapted to the targeted adolescent population. Faces were chosen rather than words to avoid the biases associated with reading abilities and socioeconomic levels. The implicit nature of the learning phase limited the impact of stress brought by an explicit task, and a judgment procedure was used to ensure deep encoding [35].

In the present study, adolescents living with obesity had significantly lower performances during the MEET‐UP task than their lean counterparts, even if their performances were above the chance level (33%). Memory impairments have been reported in adults with obesity [36, 37, 38, 39] as well as in very few studies involving children and adolescents living with the same conditions [18, 23], though they were not consistently found in every task. Lynch et al. [23] reported that in a cohort of 588 children and adolescents, aged from 8 to 14 years old, higher BMI was associated with worse visuospatial memory for shapes, but no effect of BMI was found on verbal memory or face recognition memory. In another study more comparable to our task, based on associations between creatures and scenes, relational memory performances were negatively correlated to abdominal adiposity in 126 prepubertal children, aged from 7 to 9 years old, an effect mainly driven by more severe obesity [23]. However, in a later study from Pearce et al. in 2019 [22], using a task with an encoding of background scenes, there were no significant recognition or episodic memory deficits in adolescents living with obesity. This discrepancy with our findings could be related to some procedural differences such as the retrieval of background scenes outside the scanner or the use of colored scene materials more easily recognized due to their richness features. With the new MEET‐UP task, the present study adds evidence to the negative effects of obesity on memory performance in male adolescents without diabetes after a complex associative implicit learning phase with faces and backgrounds.

In both groups of adolescents, the MEET‐UP task was associated during encoding with the activation of the parahippocampal area, classically involved in visualization of scenes, and belonging to a ‘core network’ involved in face perception, for example, the bilateral occipito‐ventral system including the inferior occipital gyri and the lateral fusiform gyrus, in previous neuroimaging studies performed among adults [40, 41]. During encoding, the activated core pattern extended to regions involved in implicit memory such as the hippocampus and the parahippocampal gyrus, the precuneus, as well as the orbitofrontal cortex, already described in face encoding protocols [42, 43]. During retrieval, the core pattern was less extensively solicited, and the hippocampus was not recruited.

Here, lower memory performances observed in adolescents living with obesity are concomitant to a different brain activity occurring during phases of implicit encoding and of retrieval in our task. When the main effect of encoding was compared between groups, lower activations of the right hippocampus and the parahippocampal gyrus were revealed in adolescents living with obesity. This is consistent with a previous neuroimaging study [22]: during the encoding of colored scenes in the scanner, a similar hypoactivation of the right hippocampus and the parahippocampal gyrus was detected in adolescents living with obesity for images successfully recognized outside the scanner. Moreover, it was reported that a lower radial thickness of the hippocampus mediated the relationship between high BMIs and worse memory performances in children and adolescents [18]. These three studies [18, 22, 23] and ours indicate that adolescent populations with obesity present subtle cognitive deficits associated with particular brain patterns of activation that can be uncovered using specifically designed tasks. Although sparse and based on various materials, this set of studies could suggest that memory deficits related to hippocampal impairments can be observed in adolescents living with obesity and can only be detected in conditions presenting a certain level of difficulty.

This study also revealed a hyperactivation of the precuneus in adolescents living with obesity during the retrieval phase of our task. According to the Encoding/Retrieval flip, the posteromedial cortex presents a deactivation during encoding and a hyperactivation during the retrieval phase [44]. Our study shows that compared to lean subjects, the posteromedial cortex in teenagers with obesity presents an increased level of activity during retrieval. Moreover, the level of activation of the precuneus was inversely related to the recognition accuracy in teenagers, including healthy subjects and patients with obesity. This hyperactivation in adolescents living with obesity may reflect a partly inefficient compensation process. The hyperactivation of the precuneus combined with the difference in reaction time observed in adolescents living with obesity might suggest that they employed another strategy which would be more frequently engaged in recollection based on context details for retrieval than the lean subjects. This recollection strategy would require a deeper implication of the precuneus cortex that plays a critical role in transforming contextual representation in a self‐related representation, a process at the start of event recollection [45]. This specific engagement of the precuneus in obesity during adolescence may impact not only episodic memory retrieval but also various cognitive processes such as visuo‐spatial imagery and self‐related processing.

A major limitation of this study remains the small cohort of patients recruited and the replication of this study on a larger cohort of adolescents would be required to confirm our findings. Due to this small number, only part of the differences in activation between the 2 groups has been correlated with behavioral performances, which enable firm causal association. Our results only concern male adolescents, and these results could not be generalized to female adolescents. Moreover, while male and female faces were used in the task, inducing gender effect on response rate in these adolescents, no major effect of gender faces on accuracy of the retrieval phase was found (data not shown). The duration of their metabolic disease and lifestyle parameters such as the composition of their diet and physical activity were not recorded. While the task used here mimick a daily‐life event, encountering strangers in new environments, the use of faces on gray background could limit the ecological plausibility of the task.

This study adds evidence of the deleterious impact of obesity during adolescence on episodic memory and on associated cerebral activity patterns. This memory impairment could contribute to academic learning difficulties faced by adolescents living with obesity, but also more generally to the development of obesity. This was recently suggested by the vicious cycle of obesity and cognitive decline model, proposing that cognitive alterations contribute to reinforce behavior favoring obesogenic diet intake [46].

Author Contributions

Anaïs Emmie Bouvier: formal analysis, writing – original draft preparation, visualization. Bixente Dilharreguy: MRI preprocessing, writing – review and editing. Ernesto Sanz‐Arigita: conceptualization, writing – review and editing. Pascal Barat: patient recruitment, data collection, writing – review and editing. Sylvie Berthoz: conceptualization, data collection, writing – review and editing. Elodie Barse: data collection. Aline Marighetto: formal analysis, writing – review and editing. Guillaume Ferreira: conceptualization, data collection, visualization, supervision, funding acquisition. Gwenaëlle Catheline: conceptualization, data collection, mri preprocessing, formal analysis, writing – original draft preparation, visualization, supervision, funding acquisition.

Funding

This work was supported by the French National Research Agency (ANR‐15‐CE17‐0013 OBETEEN to G.F. and G.C.). E.S.G. was the recipient of a postdoctoral fellowship from ANR OBETEEN (2015–2016) and the clinical research program of the LabEx BRAIN (ANR‐10‐LABX‐43 to G.F; 2016–2017). A.E.B. was the recipient of a PhD fellowship (2022‐2025) from the French government in the framework of the University of Bordeaux's France 2030 program/Grant Programme de Recherche (GPR) BRAIN2030.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Supporting Information S1

Emmie Bouvier, Anaïs , Dilharreguy Bixente, Sanz‐Arigita Ernesto, et al. 2026. “Memory Performance in Adolescents Living With Obesity: Brain Activation During Encoding and Retrieval in a Case‐Control Study,” Obesity Science & Practice: e70147. 10.1002/osp4.70147.