Understanding the developmental trajectory of behavioral problems and subcortical structure morphometry in healthy children at 6 years old and long-term impact of early nutrition: the COGNIS study

Abstract

Background

Early behavioral problems may influence adult psychopathology, and early-life nutrition plays a critical role in shaping behavioral outcomes during childhood.

Objective

This study investigated whether subcortical brain volumetry at age six is associated with early behavioral trajectories and the potential influence of early nutrition on this relationship.

Methods

Data from 82 children participants in the COGNIS study were included in the present analysis. During the first 2 months of life, fifty infants were randomized to one of two formula diets until 18 months of age: a standard formula (SF, n = 26) or an experimental formula (EF, n = 24) supplemented with bioactive compounds. A reference group of breastfed infants (BF, n = 32) was also included. Behavioral assessments were conducted using the Child Behavior Checklist (CBCL) at 18 months, 2.5 years, and 4 years. Structural magnetic resonance imaging (MRI) was performed at 6 years to assess volumes of bilateral subcortical nuclei, brainstem, cerebellum, and total intracranial volume. Complete behavioral and imaging data were available for 37 participants (SF = 19, EF = 18). Weights for linear, quadratic, and mixed linear/quadratic growth curves were computed for CBCL total, internalizing, externalizing, and DSM-oriented scales. Non-parametric correlations between CBCL growth curves and subcortical brain volumetry were computed after adjusting for relevant confounding factors. Generalized linear mixed model for repeated measures was performed.

Results

No significant effects of early nutrition on behavioral trajectories were found; in fact, EF and BF groups exhibited similar patterns across internalizing, externalizing, total problems and DSM-oriented scales. CBCL domains followed distinct developmental trajectories, and interestingly, children’s subcortical volumetry of specific brain area at 6 years old, were primarily associated with non-linear behavioral growth curves. Amygdala volume correlated with total problems scores and DSM-oriented scales, while hippocampal volume was linked to internalizing, oppositional defiant, and ADHD-related behaviors. Cerebellar cortex volume correlated with ADHD and externalizing problems, the latter also associating with putamen. Pallidum volume was correlated with internalizing and anxiety symptoms.

Conclusions

These findings suggest that non-linear behavioral growth models more effectively reflect brain–behavior associations. Futhermore, subcortical brain morphometry, particularly of the hippocampus, may be shaped by behavioral patterns during critical developmental windows—most notably around 2.5 years of age.

Supplementary Information

The online version contains supplementary material available at 10.1186/s13034-026-01030-7.

Introduction

Behavioral problems in early childhood are a potential indicator of psychopathological problems at later ages [1]. The global worldwide prevalence rate of mental disorders in children and adolescents is around 13.4%, similar to those reported for other chronic health problems, such as obesity (16.8%) [2]. Consequently, optimal clinical and practical methodologies must be crucial to design policies that address mental health needs of children and adolescents.

It is well known that early nutrition is of vital importance for optimal brain development and subsequent prevention of behavioral problems in children. Adequate intake of essential nutrients, including but not limited to proteins, long-chain polyunsaturated fatty acids (LC-PUFAs), nucleotides, sialic acid, milk fat globule membrane (MFGM) and synbiotics, plays a critical role in shaping both brain structure and function [3–5]. In fact, recent studies have demonstrated that early-life nutritional supplementation can have long-lasting positive effects on cognitive function, brain structure and behavior development in later stages of development [6, 7]. Therefore, promoting and ensuring a well-balanced and nutrient-rich diet during this critical period is necessary for supporting healthy brain development, thus reducing the risk of behavioral problems in children.

While externalizing disorders represent a significant burden, internalizing problems—including anxiety, affective problems, and withdrawal—constitute another major dimension of early behavioral dysregulation. These internalizing spectra show distinct developmental pathways and have been linked to alterations in limbic structures, particularly the amygdala and hippocampus, which are central to fear processing and emotional memory. The developmental trajectories of these problems and their neural correlates remain underexplored in early childhood, despite their profound implications for later anxiety and mood disorders.

Beyond the internalizing-externalizing dichotomy, broader behavioral dimensions—such as total problems, ADHD-specific symptoms, and pervasive developmental issues—provide complementary insights into a child’s mental health. These overarching metrics capture the global burden of psychopathology and may reflect disruptions in large-scale brain networks involving multiple subcortical structures. Understanding how these diverse behavioral domains develop longitudinally and relate to brain maturation is essential for a comprehensive model of developmental psychopathology.

In addition to environmental and genetic factors, both anatomical and functional bases of cortical and subcortical structures, as well as their maturation, also play a key role in the transition from simple signals to symptoms and subsequent mental disorders development [8]. In this line, externalizing spectrum disorders in children, including oppositional defiant disorder (ODD), attention deficit/hyperactivity disorder (ADHD) and behavioral disorders, were traditionally considered as separate entities; however, based on their comorbidity patterns, trait impulsivity has been currently identified as a common factor [9, 10]. Trait impulsivity is controlled by a series of dopaminergic subcortical mechanisms, including the nucleus accumbens, ventral tegmentum, and ventral areas of caudate and putamen [10, 11]. Interestingly, these areas are not only part of the reward circuit of the brain, but also are associated with irritability and anhedonia, which might explain the poor response to incentives observed in externalizing disorders (ADHD, ODD, behavior problems and substance use disorder) [12]. Additionally, above-mentioned structures also receive control inputs during maturation from cortical structures, mainly prefrontal cortex involved in emotional regulation and posterior cingulate cortex, as well as other subcortical structures, including amygdala, hippocampus, periaqueductal gray substance and medial hypothalamus [13, 14]. In fact, current research supports that severity of externalizing symptoms is associated with a reduction in the activity of the amygdala and its connectivity with the prefrontal ventromedial cortex before frightening stimuli [15]. Furthermore, this severity is also negatively correlated with volume of amygdala, thalamus and parahippocampal gyrus [16, 17]. Taken together, these findings suggest that both cortical and subcortical structures play a fundamental role in explaining externalizing and internalizing disorders.

Crucially, the relationship between early behavior and brain structure is dynamic and likely bidirectional across development. While nascent brain circuits constrain and guide behavioral expression, experience-dependent plasticity also suggests that recurring behavioral patterns can shape neural architecture over time. Our study leverages a longitudinal design to ask a specific question within this complex interplay: Do early behavioral trajectories—as dynamic markers of the developmental process—predict later subcortical volumes above and beyond single-timepoint assessments? Testing this predictive association requires modeling brain volumes as the outcome, an analytical approach that statistically tests the informative value of behavioral trajectories for later brain structure, without precluding the existence of underlying bidirectional effects.

The current study was aimed to determine the associations between longitudinal trajectories of early childhood behavioral problems evaluated at 18 months, 2.5 and 4 years old, and the volumes of the subcortical structures most frequently associated with these problems, including accumbens, amygdala, caudate, hippocampus, putamen, pale, thalamus and brainstem assessed at 6y. We evaluated both linear and non-linear trajectories to model the dynamic process of change. This longitudinal approach is fundamentally more informative than a single-timepoint assessment for several key reasons. First, two individuals may exhibit identical scores at a single timepoint, yet have arrived at that state via divergent trajectories (e.g., a stable decline versus a sharp improvement). Modeling non-linear trajectories is critical for uncovering these distinct patterns of change, which are often essential for accurate prognosis and for elucidating underlying mechanisms. Second, the functional form of a trajectory (e.g., accelerating versus decelerating improvement) is a more powerful predictor of subsequent outcomes than a single cross-sectional score. This is because the trajectory’s shape captures the momentum and direction of the change process itself. Thirds, many behavioral, developmental, and clinical processes—such as learning, recovery, and habituation—are inherently non-linear. The use of non-linear models allows for the identification of critical inflection points (e.g., peaks, troughs, transitions between phases) and provides a more ecologically valid representation of the actual phenomenon. A secondary objective was to analyze the potential long-term effects of a bioactive nutrients-enriched infant formula on behavioral developmental trajectories up to 4 years old and its association with brain morphology.

Methods

Participants

The initial study sample was based on the 220 (99 girls) healthy full-term infants participating in the COGNIS study, a prospective, double-blind, randomized clinical trial (RCT) designed to evaluate the effects of a novel infant formula on immunological and neurocognitive development in healthy infants (registered at www.ClinicalTrials.gov, Identifier NCT02094547). Detailed information on this project, including study design, subject recruitment and population characteristics, have been previously described [7]. Briefly, a total of 170 healthy Spanish infants aged between 0 and 2 months old were randomized (ratio 1:1) to receive, during their first 18 months of life, either a standard infant formula (SF, n = 85) or an experimental infant formula (EF, n = 85) enriched with MFGM components [10% of total protein content (wt: wt)], synbiotics [Fructooligosaccharides (FOS): Inulin proportion 1:1; Bifidobacterium longum subsp infantis CECT7210 (Bifidobacterium infantis IM1) and Lactobacillus rhamnosus LCS-74], LC-PUFAs, gangliosides, nucleotides and sialic acid. Additionally, 50 exclusively breastfed infants were enrolled as a control group (BF, n = 50).

Maternal educational level was determined (0: primary, 1: secondary, 2: vocational training, 3: university) on the first visit. Maternal IQ was evaluated using Cattell’s intelligence test (G Factor) [18].

Emotional/behavioral evaluation using CBCL (Child Behavior Checklist) test was conducted in COGNIS children at 18 months, 2.5 and 4 years old; however, in current analysis, only participants whose evaluation was completed at the three mentioned time points were considered (CBCL sample, n = 82, 35 girls / SF, n = 26; EF, n = 24; BF, n = 32). Structural Magnetic Resonance Imaging (MRI) was performed on 67 (24 girls) volunteering participants from the COGNIS study at 6 years old (mean age = 6.13; range [6.02–6.57 years old]). However, only the images of 37 participants (10 girls) were of adequate quality for further analysis. In this regard, 46 participants (25 girls) completed CBCL questionnaire at the three time points, but they had no MRI results or images obtained were deficient in quality (“Missing” group). Another set of participants was defined as those who had quality resonances but had not completed all three CBCL measures (“MRI only”, n = 19, 14 girls). Both “Missing” and “MRI only” groups were used in the sensitivity analysis.

The COGNIS study was conducted in accordance with the revised Declaration of Helsinki II Principles [19]. Ethical approval was obtained from the Research Bioethical Committee of the University of Granada, Spain, as well as the Bioethical Committees for Clinical Research of the San Cecilio University Clinical and University Mother-Infant Hospitals of Granada. Prior to their participation in the study, all families were thoroughly informed about the procedures, and written informed consent was obtained from each parent or legal guardian.

Assessment of child behavior problems: CBCL test (child behavior checklist)

The Child Behavior Checklist (CBCL)/1.5-5 inventory for preschoolers is designed to assess the child´s emotional and behavioral problems from 1.5 to 5 years old, and it is used internationally for clinical evaluation and research in early stages of development [20, 21]. The Spanish version of the CBCL/1.5-5 (Unitat d’Epidemiologia i de Disgnòstic en Psicopatologia del Desenvolupament, Barcelona) consists in 101 items divided into three scales (internalizing, externalizing and global). The internalizing scale includes four categories of problems: emotionally reactive, anxious/depressed, somatic complaints and withdrawn. The externalizing scale encompasses two types of problems: attention problems and aggressive behavior. CBCL/1.5-5 questionnaire also includes the DSM-oriented scales used in this analysis: affective problems, anxiety problems, pervasive developmental problems (autistic spectrum), ADHD, and oppositional defiant problems. For closed items, a Likert scale is used with answers ranging from 0 (“not true”), 1 (“something or sometimes true”) or 2 (“very true or often true”), considering only the two previous months to the evaluation. The total direct score on a scale was obtained automatically. Direct scores on the total, externalizing, internalizing and DSM-oriented scales were used to perform the longitudinally study and to compute the developmental trajectories of the participants. Typical scores (T) were used only to facilitate categorization into clinical and borderline clinical problems (for example, in the summary scales T ≥ 64 is usually used for clinical problems, while 60 ≤ T ≤ 63 is used for clinical borderline) [22]. Parents or legal guardians received the CBCL/1.5-5 inventory and appropriate instructions to complete it. CBCL questionnaire was filled out by the mother (81.67%) or the father (18.33%). However, given that there was no difference from answers depending on who completed it (all p > 0.10), this factor was ignored in current analysis.

In current study, the direct scores of internalizing, externalizing or total behavioral problems and DSM-oriented scales (affective problems, anxiety, pervasive developmental problems, ADHD, and oppositional-defiant problems) were used. Externalizing and internalizing scales were considered as substitutes for syndromic scales. Although the total scale is the sum of the externalizing and internalizing, its use provides information relative to the total sum of problems that is not contained in either of two scales separately. Likewise, the direct score is considered instead of the T-score or percentiles to prevent taking normative samples as a reference, which could hide each participant’s trajectory over time [23].

Magnetic resonance imaging (MRI) procedure

Brain imaging was performed when participants were 6.13 years old on average, range = [6.02, 6.57]. Finally, as previously indicated, brain images were evaluated by an expert radiologist to determine their quality and the presence of any brain abnormality, in accordance with the ethical standards of the University of Granada.

Imaging data acquisition

Brain structure scanning was performed using the Siemens Tim Trio 3T MRI system (Siemens, Erlangen, GE) equipped with a 32-channel head antenna, installed in the Mind, Brain and Behavior Research Centre (CIMCYC) (Granada, ES). Head movements were minimized using foam positioners around the participant’s head. A T1w 3D MPRAGE (magnetization-​prepared rapid acquisition with gradient echo) sequence was used with the following parameters: RT = 2300 ms, RT = 3.10 ms, IT = 900, flip angle = 9 °. In each volume, 208 slides of 0.8 mm thickness were acquired, so that the voxel size was 0.8 × 0.8 × 0.8 with a matrix size of 320 × 320.

Subcortical structures morphometry

The structural volumes were processed using Freesurfer’s (v. 6) recon-all script, a brain image analysis suite (available at http://surfer.nmr.mgh.harvard.edu/) running on the UNIX platform of the Alhambra supercomputer from the University of Granada (http://alhambra.ugr.es/). Preprocessing and segmentation of subcortical structures (including cerebellum, thalamus, caudate, putamen, pallidum hippocampus, amygdala, accumbens and brainstem) are fully automated. Processing details have been described elsewhere [24, 25]. Briefly, the automated protocol includes motion correction, non-cerebral tissue elimination, and correction of inhomogeneities in the magnetic field, linear and non-linear registration to the Talairach atlas as well as label dissemination for each voxel according to Freesurfer’s subcortical atlas. In subcortical segmentation, Freesurfer combines both intensity of the voxels with the probability distribution of the different types of brain tissue, as well as the spatial relationships of each voxel in relation to neighboring structures defined in labeled atlases [25]. After segmentation, volumes of each subcortical structure were obtained.

The validity and reliability of this automatic segmentation is well established [26, 27] with intraclass correlation coefficients greater than 0.80 [27]. However, there is a tendency to overestimate the volume of the structures by around 10%, which can lead to an increase in the false-negatives rate [28]. In paediatric population, correspondence between the volumes of structures such as the amygdala and the hippocampus estimated by Freesurfer or manually has been found to be superior to other automatic methods such as FSL-First. Consequently, this method still needs improvement [29] and careful visual inspection is necessary to ensure segmentation quality. For this reason, an expert radiologist assessed the segmentation process, and 12 participants were excluded from this analysis due to low quality images or defective segmentation.

Statistical analysis

To avoid any bias in the selection of the final sample, comparisons in the confounding variables were performed between final “MRI” sample and “Missing” sample using non-parametric Mann-Withney’s test. Welch’s robust t-test was used on both CBCL scales and volumes of subcortical structures, thus controlling the inequality of groups in size and variance.

Longitudinal differences between COGNIS groups were estimated using a generalized linear mixed model (GLMM) for repeated measures, controlling for sex, gestational age, maternal IQ and maternal educational level. Bonferroni-corrected post-hoc comparisons were used to identify significant pair-wise group differences (adjusted p value < 0.05).

Potential associations between behavioral problems trajectories and subcortical structures volumetry were evaluated in two phases. First, the trajectory of each participant was computed in each of the evaluation’s scales using a set of polynomial contrast approximation. This method is ideal for testing our specific hypotheses about the shape of the trajectories. To this effect, the time lag between evaluations (6 months and 20 months) was considered as well as the fact that two polynomials of first or second degree can be only defined given that there were only 3 CBCL measures. Therefore, ideal trajectories can be classified as: (a) linear (the direct score grows/decreases steadily) with coefficients [-0.71, -0.29, 1]; (b) quadratic (up/down peak on the direct score in the second measure), with coefficients [1 -1.33 0.33]; and (c) mixed, in which the direct score shows an increase/decrease followed by a plateau, Type I, with coefficients [-1.87, 1, 0.87]; or vice versa, Type II with coefficients [0.87, 1, -1.87] (Supplementary Fig. 1). Thus, the non-linear contrast codes, a translation of the children age in months, formally test the hypothesis that the averages of the mean of the middle and final timepoints are significantly different from the other points (Type I trajectory), and that the final timepoint is different from the average of the first two timepoints. Consequently, a linear trajectory with a positive coefficient indicates that direct score has increased steadily over time, while those with a negative coefficient indicates that score has decreased consistently with age. A quadratic trajectory with positive coefficient indicates that score was reduced at 2.5 years, but at 4 years returned at levels similar to the initial ones, while a negative coefficient indicates that score achieved a maximum peak at 2.5 years. Finally, a Type I trajectory describes an increasing/decreasing score (positive/negative coefficient, respectively) in the first time period that then remains constant during the second. Note that the coefficients so obtained are equivalent to standardized regression coefficients.

Spearman’s nonparametric partial correlation coefficient was used to determine the association between the volume of subcortical structures and behavioral trajectory assessed by CBCL, considering sex, gestational age (months), maternal age (years), maternal IQ, maternal educational level (primary, secondary, second grade, University) and total intracranial volume (TIV). The latter factor is included since the volume of subcortical structures is expected to depend on the total brain volume. Associations were assessed using Spearman’s correlation, chosen for its robustness to non-normal distributions - particularly relevant for bounded trajectory weights - and its ability to detect monotonic rather than strictly linear relationships. The criterion of statistical significance was p < 0.025.

All statistical analyses were performed using the SPSS statistical software package (version 28.0; IBM SPSS Inc., Chicago, IL, USA) and Matlab software (version 20.0; Mathworks, US).

Results

General characteristics of the COGNIS study participants

Table 1 shows general characteristics of the children classified into COGNIS groups (type of early nutritional intervention) and with CBCL measurements obtained at three time-points (18 months, 2.5 and 4 years). Significant differences between study groups were found in terms of sex distribution and maternal educational level. In fact, a higher proportion of girls was observed in BF group compared to EF (p = 0.010). Furthermore, mothers of BF infants showed a higher educational level (p = 0.001) compared to mothers of both formula-fed groups (SF and EF). Differences found in maternal age disappeared after Bonferroni post hoc test. To test de impact of maternal education and IQ on the nutritional effect, we conducted additional sensitivity analyses with these two variables as covariates in longitudinal linear mixed models. This allow us to statistically control for the influence of these factors on the behavioral trajectories over time. No effects were observed (all p-values > 0.50). The results of these analysis appear to confirm that the absence of significant differences between nutritional intervention groups persists even after accounting for these key covariates. This finding strengthens the conclusion that the early nutritional intervention, per se, was not a primary determinant of CBCL scores in our cohort.

Table 1.

General characteristics of the COGNIS study participants with the three inventory measurements (CBCL), participant´s with high quality of the brain’s magnetic resonance imaging (MRI) and missing (missing values in the MRI)

		SF (n = 26)	EF (n = 24)	BF (n = 32)	P-value	CBCL (n = 82)	MRI (n = 36)	Missing (n = 46)	P-value
Maternal age (years)		31.6 (6.30)a	31.2 (5.02)a	34.5 (4.87)a	0.042	32.6 (5.55)a, b	30.4 (5.80)a	34.3 (4.80)b	0.005
Maternal IQ		107.5 (14.46)	104.1 (14.15)	109.2 (14.17)	0.420	107.1 (14.24)	104.9 (14.76)	108.9 (13.71)	0.442
Maternal educational level	Primary Secondary VT University	11.5 34.6a 15.4 38.5a, b	20.8 16.7a, b 45.8 16.7b	3.1 6.3b 25.0 65.6a	0.001	11.0 18.3a, b 28.0 42.7a, b	16.7 30.6b 30.6 22.2b	6.5 8.7a 26.1 58.7a	0.034
Gestational Age (weeks)		39.8 (1.27)	39.7 (1.35)	39.4 (1.36)	0.462	39.6 (1.33)	40.0 (1.30)	39.4 (1.32)	0.124
Sex (boys)		65.4a, b	75.0a	37.5b	0.010	57.3	69.4	47.8	0.145
Birth weight (kg)		3.37 (0.47)	3.35 (0.41)	3.37 (0.40)	0.976	3.36 (0.42)	3.39 (0.44)	3.34 (0.41)	0.600
Birth length (cm)		50.92 (2.34)	50.96 (2.47)	50.59 (2.29)	0.806	50.80 (2.29)	51.07 (2.30)	50.60 (2.28)	0.360
HC at birth (cm)		34.93 (1.42)	34.21 (1.17)	34.56 (1.34)	0.237	34.58 (1.33)	34.79 (1.31)	34.40 (1.34)	0.200

Data are expressed as means (standard deviation); Numbers for maternal educational level and Sex are in percentage

Notes. BF: breastfed infants; CBCL: Check Behavior Checklist; EF: experimental infant formula; IQ: intelligent quotient; MRI: Magnetic Resonance Imaging; n = number of cases; SF: standard infant formula; VT: Vocational Training; HC: head circumference

Values not sharing the same suffix (ab) were significantly different in the Bonferroni post hoc test. P-values < 0.05 are highlighted in bold

Additionally, we also evaluated general characteristics of the participant’s complete sample, regardless of type of early nutrition (Table 1). In this case, data are presented into three groups: (i) the participants who completed the three evaluations at mentioned time-points (“CBCL” group, n = 82); (ii) participants who also presented adequate MRI at the age of 6 years (“MRI” group, n = 36); and (iii) the participants with missing data due to absence or low quality MRI (“Missing” group, n = 46). As shown in Table 1, significant differences were only observed in maternal educational level (p < 0.001) between “MRI” and “Missing” groups. These data suggest that maternal educational level could be an important factor in explaining the observed loss of data in current analysis.

3.2. CBCL scales

We first analyzed whether type of early nutrition might have long-term effects on behavioral development from 18 months to 4 years old. For this purpose, a GLMM of repeated measures was used, which included a within-subjects time factor with three levels (18 months, 2.5, and 4 years) as well as a group factor with three levels (SF, EF and BF). Additionally, analysis was adjusted by sex, gestational age, maternal IQ and educational level. In the present analysis no statistical differences were found regarding CBCL scores between COGNIS groups (Supplementary Fig. 2). However, both EF and BF groups seem to show a more similar pattern regarding internalizing, externalizing and total problems (Supplementary Fig. 2A-C). Finally, patterns obtained for DSM-oriented scales (affective problems, anxiety problems, pervasive developmental, ADHD and oppositional defiant) do not differ forexperimental formula-fed infants those who were breastfed (Supplementary Fig. 2D-H).

We next performed descriptive statistics for each of the inventory scales according to the set of participants “CBCL”, “MRI” and “Missing”. As shown in Table 2, there were no significant differences between “MRI” and “Missing” groups in any of the evaluations, although there was a trend in externalizing problems at 2.5 years [t (80) = 1.90, p = 0.061]. Therefore, these results indicate that infant’s behavior evaluation in both “Missing” and “MRI” groups was very similar.

Table 2.

Descriptive statistics of the direct CBCL scores on the DSM-oriented, total, internalizing and externalizing scales

	CBCL (n=82)			MRI (n=37)			Missing (n=46)
CBCL Scales	18 (m)	2.5 (y)	4 (y)	18 (m)	2.5 (y)	4 (y)	18 (m)	2.5 (y)	4 (y)
Internalizing Problems	7.96 (0.94)	9.3 (1.17)	9.89 (1.09)	8.43 (1.03)	10.27 (0.98)	10.65 (1.12)	7.59 (0.86)	8.52 (1.31)	9.28 (1.08)
Externalizing Problems	13.58 (1.1)	13.07 (1.27)	12.3 (1.14)	14.51 (1.2)	14.49 (1.12)	12.35 (1.09)	12.83 (1.01)	11.93 (1.37)	12.26 (1.19)
Total Problems	35.18 (2.93)	36.2 (3.47)	34.23 (2.91)	36.86 (3.32)	38.97 (2.83)	34.68 (2.92)	33.83 (2.6)	33.98 (3.9)	33.87 (2.94)
Affective Problems	2.51 (0.37)	2.86 (0.37)	2.46 (0.32)	2.76 (0.46)	2.78 (0.32)	2.59 (0.31)	2.3 (0.28)	2.91 (0.41)	2.35 (0.33)
Anxiety Problems	3.49 (0.34)	4.06 (0.46)	4.02 (0.45)	3.49 (0.33)	4.19 (0.41)	3.97 (0.45)	3.5 (0.35)	3.96 (0.5)	4.07 (0.46)
Perv. Develop. Problems	3.29 (0.44)	3.52 (0.48)	3.77 (0.47)	3.24 (0.48)	3.84 (0.46)	4.05 (0.54)	3.33 (0.42)	3.26 (0.5)	3.54 (0.4)
ADHD	5.42 (0.41)	4.86 (0.46)	4.58 (0.39)	5.62 (0.42)	5.35 (0.42)	4.49 (0.41)	5.26 (0.41)	4.46 (0.49)	4.65 (0.38)
Oppositional defiant Problems	3.14 (0.32)	3.54 (0.39)	3.65 (0.4)	3.38 (0.33)	4.08 (0.36)	3.62 (0.35)	2.96 (0.3)	3.11 (0.4)	3.67 (0.44)

Data are expressed as means (typical errors)

Note. Perv. Develop. Problems: Pervasive Developmental Problems; ADHD: attention deficit hyperactivity disorder; m: months; y: years

The percentages of participants with clinical problems (T ≥ 64) and clinical borderline (60 ≤ T ≤ 63) were calculated in relation to the preliminary 82 participants for each of the scales (Table 3). Our results show that the most frequent clinical problems were the developmental ones, which reached a maximum at 4 years (20.7%). Considering clinical borderline, the most frequent were the affective and anxious problems at 2.5 years (22% and 18.3%, respectively). Overall, internalizing problems were more common than externalizing, both at clinical and subclinical levels. The internalizing problems showed a growing trend with a maximum at 4 years (12.2% and 17.1% clinical and subclinical, respectively), while the externalizing problems did not show this same tendency. Further analysis was aimed to evaluate the percentage of participants whose T-scores increased or decreased between the three-evaluation time-points on a subject-by-subject basis (Table 3). Overall, a major increase was observed between 18 months and 2.5 years compared to 2.5-4 years period, supporting a non-linear overall trajectory. More specifically, anxiety problems grow by 19.5% between 18 months and 2.5 years, and by 14.6% between 2.5 and 4 years. Similar pattern was observed in affective (18.3 vs. 9.8%), ADHD (9.8 vs. 4.9%), and oppositional defiant (11.0 vs. 7.3%) problems. However, the developmental problems showed a different increase pattern, with a lower percentage in the early period than in the later one (9.8 vs. 18.3%). On the other hand, the decrease pattern is more diffuse since no general trend was observed. In fact, externalizing, ADHD and developmental problems decreased more in the first-time frame than in the second, while affective and, to a greater extent, anxiety problems showed a larger decrease in the later period (8.5 vs. 14.6% and 7.3 vs. 17.1%, respectively).

Table 3.

Prevalence of clinical pathological and clinical borderline problems in the total participants’ sample (n = 82), with the percentage of participants whose T scores increase/decrease from one measure to the next (2.5 y-18 m, 4-2.5 y)

	Clinical pathology			Clinical borderline			Increase		Decrease
CBCL Scales	18 m	2.5y	4y	18 m	2.5y	4y	2.5y-18 m	4-2.5y	2.5y-18 m	4-2.5y
Internalizing Problems	4.9	9.8	12.2	8.5	15.9	17.1	19.5	9.8	7.3	6.1
Externalizing Problems	6.1	8.5	3.7	9.8	7.3	8.5	12.2	6.1	12.2	9.8
Total Problems	6.1	11.0	8.5	12.2	12.2	9.8	12.2	6.1	7.3	11.0
Affective Problems	8.5	11.0	9.8	14.6	22.0	18.3	18.3	9.8	8.5	14.6
Anxiety Problems	4.9	11.0	11.0	12.2	18.3	15.9	19.5	14.6	7.3	17.1
Perv. Develop. Problems	11.0	13.4	20.7	4.9	2.4	7.3	9.8	18.3	9.8	6.1
ADHD	11.0	11.0	3.7	12.2	6.1	7.3	9.8	4.9	15.9	11.0
Oppositional Defiant Problems	4.9	11.0	12.2	0.0	0.0	0.0	11.0	7.3	4.9	6.1

Data are expressed as percentages of the total population

Note. Perv. Develop. Problems: Pervasive Developmental Problems; ADHD: attention deficit hyperactivity disorder; m: months; y: years

We next evaluated the ability to predict participants’ borderline or clinical scores at 4 years old considering previous borderline or clinical scores obtained at earlier ages. For this purpose, the odds of a positive or normal classification, considering a positive at the 18-months-old or 2.5-years-old evaluation, were calculated (Table 4). Overall, our data suggest that the risk of obtaining a borderline or clinical score at 4 years old was significantly higher if it has been obtained previously, except for anxiety problems (OR = 2.16, p = 0.131).

Table 4.

Conditional probabilities and odds ratio (OR) for the CBCL scales obtained in total sample (n = 82)

CBCL Scales	p (O = 1/(A = 1| B = 1)	p (O = 0/(A = 1 | B = 1)	OR	SE	P-value
Internalizing Problems	0.75	0.16	16.33	1.813	0.000
Externalizing Problems	0.70	0.22	8.17	2.109	0.005
Total Problems	0.80	0.19	16.62	2.045	0.000
Affective Problems	0.78	0.27	9.68	1.794	0.000
Anxiety Problems	0.50	0.32	2.16	1.663	0.131
Perv. Dev. Problems	0.52	0.15	6.06	1.738	0.001
ADHD	0.78	0.27	9.28	2.325	0.008
Oppositional Defiant Problems	0.50	0.11	2.086	2.828	0.005

Note. O: evaluation at 4 years; A: evaluation at 18 months; B: evaluation at 2.5 years; OR: Odds-Ratio; SE: standard error of the OR; 0: normal; 1: borderline or clinical pathology. Perv. Develop. Problems: Pervasive Developmental Problems; P-value: Level of significance

Finally, non-parametric Spearman’s correlations between the three CBCL evaluations time-points in the total sample, controlling for confounding variables, were obtained (Table 5). All correlations were positive and statistically significant (p < 0.01). Overall, our results suggest that the scores are serial dependent and, therefore, the percentiles at a specific time point (t) might be partly predictable from the percentiles at previous time points (t-1 and t-2). Using the algorithm described by Lee and Preacher (2013), which contrasts the differences between correlated correlation coefficients, no significant differences were observed in any of the scales (data not shown). According to these results, the scores are consistent over time, thus supporting the use of trajectory computation for determining potential effects of direct scores on subcortical structures volume, compared to classical approximation ways based on general linear model.

Table 5.

Spearman’s partial correlations between the three evaluations of CBCL (n = 82), controlling by sex, gestational age, maternal IQ and educational level

CBCL Scales	18 m-2.5y	18 m-4y	2.5-4y
Internalizing Problems	0.352*	0.334*	0.493*
Externalizing Problems	0.562*	0.514*	0.547*
Total Problems	0.482*	0.491*	0.607*
Affective Problems	0.300*	0.356*	0.504*
Anxiety Problems	0.417*	0.203	0.439*
Pervasive Developmental Problems	0.290*	0.315*	0.377*
ADHD	0.423*	0.397*	0.427*
Oppositional defiant Problems	0.340*	0.334*	0.499*

Note. ADHD: attention deficit hyperactivity disorder; m: months; y: years. * p < 0.05

Volumetry of subcortical structures

The volume of subcortical brain structures for MRI participants and for those with quality MRI but had not completed all three CBCL measurements (“MRI only”) are described in Table 6. As can be seen, no significant differences in brain structure volumetry were found, indicating there is no selection bias in the final sample (“MRI” group).

Table 6.

Volume (mm³) of subcortical brain structures in “MRI” and “MRI Only” participants

Structure	MRI (n = 37)	MRI Only (n = 18)	P-value
L. Cerebellum white matter	13,298 (277.2)	12,824 (400.9)	0.196
L. Cerebellum (cortex)	58,108 (973.6)	56,400 (1409)	0.186
L. Thalamus	7366 (99.2)	7139 (155.8)	0.102
L. Caudate	3893 (66.2)	3828 (99.5)	0.472
L. Putamen	5284 (96.3)	5138 (137.5)	0.249
L Pallidum	1959 (36.2)	1901 (52.5)	0.228
Brain stem	17,812 (335.9)	17,101 (498.7)	0.117
L. Hippocampus	3716 (64.6)	3661 (83.4)	0.493
L. Amygdala	1462 (48.3)	1443 (57.7)	0.747
L. Accumbens	576 (19.2)	564 (24.2)	0.622
R. Cerebellum white matter	12,418 (238.9)	11,956 (344.3)	0.144
R. Cerebellum (cortex)	57,455 (962.1)	55,628 (1413.9)	0.156
R. Thalamus	7350 (99.1)	7130 (152.6)	0.108
R. Caudate	4000 (65.2)	3911 (99.7)	0.318
R. Putamen	5387 (94)	5244 (134.2)	0.246
R. Pallidum	1808 (32.3)	1760 (46.7)	0.256
R. Hippocampus	3873 (60.6)	3800 (78.6)	0.326
R. Amygdala	1652 (26.8)	1619 (38.9)	0.355
R. Accumbens	644 (12.5)	629 (18.2)	0.368

Data are expressed as means (standard error of the mean); P-value: level of significance; L: left; R: right

Developmental patterns of the CBCL scales

The proportion of participants whose trajectory can be described mainly by one of the four defined patterns {linear (L), quadratic (Q), Type I and Type II} are shown in Supplementary Table 1. Each proportion has been calculated from the correlation coefficient between the three CBCL measurements and the ideal trajectory coefficients. A value of 0 was defined as cut-off point in the magnitude of the correlation, so that correlations less than or equal to 0 are considered negative trajectories, while those greater than 0 are considered positive. For example, to properly interpret the results obtained, it should be taken into account that a value of 0.61 in linear internalization means that 61% of the participants tended to increase their percentile over time, while the remaining 39% tended to decrease it. Likewise, a value of 0.43 in the quadratic internalization means that 43% of the participants showed a minimum in the measurement obtained at 30 months, while 57% showed a maximum at that same age. A value of 0.58 in Type I internalization means that 58% of the participants increased their percentile between 18 months and 2.5 years and remained at values ​​similar to those of 2.5 years when they were evaluated at 4 years; consequently, 42% of the participants decreased their percentile at 2.5 years and remained approximately at that value at 4 years. Since the classification is not exclusive, each participant has a weight in each of the trajectories. Interestingly, using Mann-Whitney’s test, there is only a significant difference between the “MRI” and “Missing” samples in total quadratic trajectory score (p = 0.03), which once again confirms that the final sample (“MRI”) is not biased.

Relationship between the volumes of subcortical brain structures and CBCL trajectories

The partial correlation analysis of the evolution trajectories of direct CBCL scores and the volumes of subcortical brain structures is summarized in Table 7. The same correlations for the T-scores can be found in Supplementary Table 2. As shown in Supplementary Fig. 3, the volume-CBCL-trajectory relationship was both linear and non-linear, and 9 brain structures whose volume is related to CBCL trajectories were identified: bilateral hippocampus, accumbens, amygdala (right), pallidum (left), bilateral cerebellar cortex and bilateral putamen. The interpretation of these models is mathematically direct: for instance, a positive sign of the linear component indicates a net increase in the subcortical structure’s volume over time, whereas a negative sign denotes a net decrease. This same principled interpretation extends to higher-order components; a significant positive quadratic term, for example, would indicate an accelerating growth curve, while a negative term would signify a decelerating or inverted U-shaped trajectory.

Table 7.

Non-parametric correlations between developmental trajectories determined by CBCL scales and the volume of brain subcortical structures

Path	Scale	Structure	Rho	P-Value
Linear	Internalizing Problems	L. Hippocampus	-0.413	0.021
Linear	Internalizing Problems	R. Hippocampus	-0.418	0.019
Quadratic	Internalizing Problems	L. Pallidum	0.429	0.016
Quadratic	Internalizing Problems	R. Hippocampus	0.434	0.015
Type I	Internalizing Problems	L. Pallidum	-0.488	0.005
Type I	Internalizing Problems	L. Hippocampus	-0.460	0.009
Type I	Internalizing Problems	R. Hippocampus	-0.500	0.004
Linear	Externalizing Problems	L. Putamen	-0.451	0.011
Linear	Externalizing Problems	R. Putamen	-0.407	0.023
Type I	Externalizing Problems	R. Cerebellum (cortex)	-0.407	0.023
Type II	Externalizing Problems	L. Putamen	0.484	0.006
Type II	Externalizing Problems	R. Putamen	0.455	0.010
Linear	Total Problems	R. Amygdala	-0.488	0.005
Linear	Anxiety Problems	L. Pallidum	-0.477	0.007
Type I	Perv. Develop. Problems	R. Amygdala	-0.474	0.007
Type II	Perv. Develop. Problems	L. Accumbens	0.408	0.023
Linear	ADHD	L. Cerebellum (cortex)	-0.585	0.001
Linear	ADHD	R. Cerebellum (cortex)	-0.639	0.000
Linear	ADHD	R. Amygdala	-0.483	0.006
Quadratic	ADHD	R. Hippocampus	0.459	0.009
Quadratic	Oppositional Defiant Problems	R. Hippocampus	0.448	0.011
Type I	Oppositional Defiant Problems	R. Amygdala	-0.504	0.004

L: left; R: right; Perv. Develop. Problems: Pervasive Developmental Problems; ADHD: Attention Deficit Hyperactivity Disorder. Spearman’s correlations. P-value: level of significanceOnly significant correlations are presented.

A negative correlation between linear trajectory in internalizing problems and both right and left hippocampus volume were found; however, when the trajectory is quadratic, with a minimum at 2.5 years, an increase in volume of the left pallidum and right hippocampus was observed. Likewise, when the trajectory is Type I with a maximum at 2.5 years that continues at 4 years, both left pallidum and both hippocampus decrease in volume (Fig. 1, upper row). In addition, a Type II trajectory (with a plateau at the minimum reached at 2.5 years) in externalizing problems is associated with greater volumes of left and right putamen (Fig. 1, lower row).

Fig. 1.

Internalizing (upper row) and externalizing (lower row) paths and volume subcortical structures

Regarding total problems, a negative correlation was observed between the linear trajectory and the volume of the right amygdala (Fig. 2), indicating that an incremental trajectory of these problems associated with a reduction in right amygdala volume. Interestingly, no other trajectory correlated significantly with the volumes of this subcortical structure.

Fig. 2.

Lineal relationship between the right amygdala volumetry and the total CBCL score path

On the other hand, the linear trajectory of anxiety problems correlated negatively with the volume of the left pallidum (r = -0.477, p = 0.007, Fig. 3), then left pallidum volumetry seems to be smaller when there is a steady increase in the trajectory of anxiety problems.

Fig. 3.

Relationship between volume of left pallidum and anxiety paths

Likewise, in terms of developmental problems, Type I trajectory negatively correlated with the volume of the right amygdala (r = -0.474, p = 0.007, Fig. 4 left); however, a positive correlation was found between Type II trajectory and right nucleus accumbens volumetry (r = 0.408, p = 0.023, Fig. 4 right). The first result implies an increase in the size of the amygdala when development problems decrease and stabilize at 2.5 years. The second result indicates that the volume of right nucleus accumbens tends to be higher the greater the reduction of developmental problems at 4 years.

Fig. 4.

Relationships between right amygdala and right accumbens with development paths

Regarding ADHD problems, significant correlations were observed with both left and right cerebellar cortex, right amygdala and right hippocampus (Fig. 5). Thus, the volumes of both left and right cerebellum, as well as right amygdala, negatively correlated with a linear trajectory (r = -0.585, p = 0.001; r = -0.639, p < 0.001; r = -0.483, p = 0.006, respectively). Therefore, both cerebellar and right amygdala volumes are smaller the higher the level of problems with increasing age. Conversely, the right hippocampus volume positively correlated with a quadratic trajectory of ADHD problems (r = 0.459, p = 0.009), which suggests that the volume of the hippocampus could be positively affected by a reduction in the level of problems at a minimum peak around the age of 2.5 years.

Fig. 5.

Relationships between ADHD paths with volumes of cerebellum (left and right), right amygdala and right hippocampus

Finally, the Type I trajectories of oppositional-defiant problems were negatively associated with the right hippocampus and right amygdala (r = -0.476, p = 0.007; r = -0.504, p = 0.004, respectively) (Fig. 6). Therefore, the volume of both subcortical brain structures is higher the greater the reduction of these problems at 2.5 and 4 years with respect to the level observed at 18 months.

Fig. 6.

Relationships between right hippocampus and right amygdala with oppositional-defiant type I paths

Discussion

This research was primarily aimed to understand the developmental trajectories of behavioral problems during the first 4 years of life and determine their potential associations with children’s subcortical brain structures volumes. Moreover, potential long-term effects of early nutrition on such trajectories and their potential associations with brain morphology at 6 years old were also evaluated. Within COGNIS framework, we have previously reported that bioactive compounds-enriched infant formula, containing MFGM components, synbiotics, LC-PUFAs, gangliosides, sialic acid and nucleotides, provide a protective role against behavioural problems development at 2.5 years old (affective problems) compared to standard formula-fed infants, and similar to those effects found in breastfed infants. In fact, these results showed significant higher increase of internalizing, externalizing and total problems, as well as ADHD and Oppositional Defiant problems from 18 months to 2.5 years in the SF group, compared to EF and BF infants [7]. Nevertheless, in the current longitudinal study, a similar effect on behavioural development could not be identified, which can be mostly explained by the smaller sample size included in the present analysis. Interestingly, as previously reported, our results also suggest that maternal factors, including maternal IQ and educational level, lead to better mental health and behavioural development in childhood [7]. While maternal education and IQ factors may influence the baseline level of behavioral problems, our longitudinal analysis, controlled for these covariates, suggests that the trajectory of change did not differ significantly between nutrition groups. This provides a more nuanced interpretation, indicating that behavioral problem outcomes are not linked to the specific early nutritional intervention in this cohort. In other previous publication, we demonstrated the importance of early nutrition up to 18 months of life, on later brain structure at 6 years old and neurocognitive outcomes, similar in EF children to BF ones, and better than those fed SF [7]. Therefore, our results highlight the inclusion of both early nutrition and maternal factors into designed policies that address mental health needs of children and adolescents. However, it is noteworthy that important sets of results in the present analysis of COGNIS study data has yielded two key findings: the first relates to behavioral developmental trajectories during the first four years of life, and the second to the association between these trajectories and the volume of children’s subcortical brain structures at 6 years of age. Overall, results obtained here suggest that behavioral trajectories lead to different relationship between behavioral problems up to 4 years and subcortical brain structures morphometry. An important consideration in interpreting our findings is the directionality of the observed associations. Our analytical model tested how early behavioral trajectories predict later subcortical volumes, an approach justified by our temporal sequence of measurements. However, we interpret these results not as evidence of a unidirectional effect, but as a snapshot of the likely bidirectional brain-behavior dynamics that unfold across development. The associations we observe may reflect a process whereby early-appearing behavioral tendencies, which are themselves rooted in neural function, become embedded and reinforced through experience-dependent plasticity, subsequently influencing the maturation of subcortical structures. Future studies with cross-lagged models or even earlier neuroimaging timepoints are needed to disentangle the precise temporal and causal dynamics of this complex relationship.

The behavioral trajectories we have observed here seem to have a relationship with volumes of hippocampus, amygdala, pallidum, cerebellum, putamen and accumbens throughout the first four years of life, in which higher levels of problems over time were associated with reduced volume of these structures. Interestingly, results obtained from non-linear trajectories also suggest that subcortical nuclei morphometry, particularly the hippocampus, could be linked to behavioral problems in a critical period around the age of 2.5 years old. Taken together, our results emphasize the need for early and effective prevention and detection of behavioral problems during the first years of life, thus reducing the risk of psychopathological disorders later in life. These findings align with the concept of sensitive periods in early brain development. The age of 2.5 years coincides with a phase of rapid hippocampal synaptogenesis and the onset of experience-dependent pruning, processes that may render this structure particularly plastic and vulnerable to environmental and behavioral influences. The non-linear trajectories observed around this developmental window may therefore reflect periods of heightened neural plasticity, where behavioral patterns could exert a more pronounced effect on hippocampal maturation and, consequently, on the volume observed at 6 years of age. This potential mechanism underscores the importance of this specific age as a critical window where behavioral interventions might be most efficacious in steering neurodevelopmental trajectories toward more adaptive outcomes.

First, in relation to the developmental trajectories, our results indicate that the CBCL evaluations are consistent over time, showing a high degree of autocorrelation. Interestingly, except for the anxiety scale, these trajectories might allow a significant prediction of possible diagnostic categorization (borderline or clinical) from measurements taken up to 2.5 years earlier. Moreover, from an analytical point of view, these data also suggest that, when repeated measurements are performed, an analysis approach that does not take into account the autocorrelation of the measurements (for example: analysis of variance with repeated measures) will produce unreliable results [23], with incorrect statistical significances [30]. In contrast to other development curve analysis strategies, such as linear mixed models or mixed growth models [31], the one used here offers the advantage of modelling the set of possible curves in a concrete way, as well as assigning a weight for each participant in each of these curves. Therefore, it can be used regardless of the total number of participants in the sample. Based on this developed methodology, our results suggest that CBCL scales in children with typical development do not seem to follow the same development curves. In fact, in the case of both anxiety and developmental internalizing problems, the greatest weight corresponds to a linear trajectory, while total, externalizing and ADHD-type problems follow a quadratic scheme, with a peak at the age of 2.5 years. Likewise, affective and oppositional-defiant problems follow a Type I pattern, with a plateau starting at 2.5 years. Finally, the total scale most often follows a Type II pattern with a plateau up to 2.5 years old.

Once analyzed the developmental trajectories in our study population, we next evaluated their potential relationships with the volume of subcortical brain structures. A total of 9 structures, 3 bilateral (cerebellum cortex, hippocampus and putamen), 2 in the left hemisphere (accumbens and pallidum), and one in the right hemisphere (amygdala), showed significant correlations with developmental trajectories. Specifically, regardless of the category of problems, a linear trajectory is significantly associated with all mentioned brain structures. In all cases, the volume of these structures decreases as the negative slope of the curve becomes steeper, thereby exacerbating the level of problems over time. On the other hand, quadratic trajectory is positively associated with right hippocampus and pallidum, so that the volume of these structures seems to increase when there is a minimal peak at 2.5 years old. Thirdly, Type I curve is negatively associated with different brain structures, including pallidum, bilateral hippocampus, right cerebellar cortex and right amygdala. These data suggest that volume of mentioned brain structures is smaller the greater the trajectory weight, supporting that low score at 18 months, although followed by a subsequent increase, are associated with higher structural volume. Finally, Type II trajectory is positively associated with bilateral putamen and left accumbens, which implies that their volume is greater when the scores have their minimum after the 2.5 years old. Having in mind these results, it is feasible to think that there could be a critical age around 2.5 years in relation to the development of brain structures, not determined by the score at that age, but its relation to the obtained at the other ages.

Regarding internalizing problems, our results indicate a bilateral hippocampal volume reduction when the problem trajectory is linear, but an increase in the left pallidum and right hippocampus volume when this trajectory is quadratic. Our results with the linear trajectory are consistent with previous research on volume of subcortical structures and internalizing problems in the pediatric and adolescent population, including depression, anxiety or phobias, where hippocampal volume reduction and volume increase of the left pallidum have been observed [32–36].Taken together, these patterns suggest that the developing brain responds to internalizing symptoms in multifaceted ways. A linear, progressive hippocampal decrease may signal a trajectory of cumulative allostatic load, potentially identifying individuals at higher risk for later psychopathology. Conversely, a non-linear trajectory with volumetric increases might indicate a more plastic, adaptive, or susceptible neural phenotype. Future longitudinal studies incorporating direct measures of HPA axis function and environmental factors are crucial to disentangle these mechanistic possibilities. Although the role of hippocampus in psychopathological problems is well known (Macpherson and Hikida, 2019), the potential effects of the increase in the volume of the pallidum are still poorly understood. In this sense, this nucleus is part of three cortico-subcortical circuits (sensory-motor, cognitive and limbic) through which facilitates or suppress behavior via regulation of the subthalamic nucleus (STN) and the substantia nigra pars reticulata (SNr) activities. Interestingly, these two structures are involved in the regulation of the thalamus, one of whose functions is salience attribution to stimuli. Consequently, reduced control by the pallidum on the STN/SNr pair results in the reduction of thalamic activity and the appearance of symptoms characteristic of depression or anxiety [11]. Although this seems to explain its potential effects on internalizing problems, further studies are still needed.

Concerning externalizing problems, our results highlighted bilateral putamen (linear and Type II trajectory) and left cerebellar cortex (Type I trajectory). Due to both linear and Type II trajectories include the difference between the two-time limits (4 years and 18 months) but of opposite sign, we will discuss both together. Furthermore, considering the externalizing DSM scales, our results showed a reduction in the volume of right amygdala and bilateral cerebellar cortex in ADHD with linear trajectory, reduction of right amygdala (Type I trajectory) and increase of left accumbens (Type II trajectory) in developmental problems. Also show a reduction of right hippocampus and amygdala in oppositional defiant problems (Type I trajectory). Thus, except for the right cerebellum, there seems to be no common nuclei between the overall score of the externalizing scale and the specific scales of this type of problem. However, using functional MRI, it has been observed that both putamen and bilateral accumbens are activated in adolescents with externalizing symptoms in response to reward, with respect to non-reward [38]. Moreover, in cases with clinical levels, both a dysfunction of the amygdala and structures that process the reward led to emotional empathy disorders as well as problems in learning by strengthening and taking of decisions [39]. On the other hand, both structural integrity and connectivity of the cerebellum seem to be crucial factors explaining the called “p factor” [(general factor of psychopathology [40], due to poor efficiency in the operation of information processing systems could be based on externalizing and internalizing disorders. Consequently, both factors have been proposed as risk markers for developing any psychopathology [41]. Therefore, the cerebellum not only play a key role in movement control and motor coordination but also seems to control visual attention and operational memory [42, 43]. Currently, its consideration as a node of the dorsal attentional network (DAN) also allows it to influence on the maintenance of attention over time (sustained attention) [44]. Taking into account these considerations, and although not as differentiated as the cerebral cortex, the cerebellum seems to have a relevant role in complex cognitive functions such as attention, working memory, and social and emotional processing [45].

In adolescent patients with ADHD, lower volumes of the hippocampus and amygdala have been observed with respect to the control group with usual development [46]. In a recent study, conducted under the umbrella of the ENIGMA collaboration [47], in which morphometric alterations were addressed in 1713 participants diagnosed with ADHD and 1529 controls, at ages between 4 and 63 years, it was observed that, with volume effect sizes is higher in children than in adults, showed a volumetric reduction in accumbens, hippocampus and putamen, in addition to total intracranial volume and caudate. Our results are consistent with those of this study, although they also suggest that this is especially true in children with linear developmental trajectories, but not so much in quadratic trajectories, in which the right hippocampus could benefit if the level of the problem is reduced at 2.5 years old. Unfortunately, there is no published literature on the effect of non-linear trajectories on the development of cortical or subcortical structures, so we can only speculate on the potential benefits of an intervention aimed to reduce the level of problems at an early age.

The role of the amygdala and its developmental trajectory at an early age is not yet well known, especially with regard to its potential consequences on behavioral problems [48]. Although it is a very small structure (approximately 0.3% of the total brain volume), it is attributed a key role in the processing of outgoing environmental information, emotional learning, sensory information and prior knowledge linking as well as in the ability to produce an adaptive response. In this regard, the amygdala is important in social cognition, since it is probably the one that organizes the ocular exploration of space, especially the faces of others, the evaluation of the trust that can be placed on others or the identification facial emotional expression [49]. Consequently, several psychiatric disorders have been linked to a malfunction of the amygdala, including obsessive-compulsive disorder, anxiety or problems in emotion regulation [50]. Moreover, this malfunction could also contribute to the development of problems related to autistic spectrum disorders [48]. Structurally, a greater volume of the bilateral amygdala has been observed in 5-year-old children with ASD [51]. In line with these findings, our data also suggest that the volume of the amygdala could be a good indicator for the trajectory of total problems. In this sense, the number of neurons in the amygdala grows postnatally, leading to about a 40% increase in volume at a mature stage. However, in patients with ASD, an abnormal developmental trajectory has been found, resulting in a substantially lower final number of mature neurons [52]. It is also important to note that abnormal development of the amygdala is also characterized by an excessive number of dendritic spines in ASD patients [53], which can lead to an inappropriate emotional development sequence [48]. In addition, in healthy patients, a change in the connectivity of the amygdala with the medial prefrontal cortex is observed over time. While in childhood a positive relationship is observed, where the increases in one structure are associated with increases in the other, in maturity a negative relationship is described so that, increases in the one decrease the activity of the other [50, 54]. This pattern is altered in ASD patients, thus supporting the potential role of malfunction of the amygdala in this disorder [55]. The nucleus accumbens also seems to play a crucial role in ASD. In fact, study carried out in animal model showed that that the decrease in the serotonin levels injected into the accumbens causes social behavior deficiencies. These difficulties are reversed when the dorsal raphe is activated [56]. These findings suggest that the underlying cause could be brainstem-born abnormalities in the serotonergic circuit that spread to the subcortical structures and the cerebral cortex [57].

Oppositional defiant disorder (ODD) is perhaps the most frequent childhood disorder, with prevalence rates around 5% in the 6–16 age group [58]. Its characteristics are well known and related to respect for/negativity before authority and norms, animosity and disobedience. A recent meta-analysis suggests that insecure and disorganized attachment is associated with a higher probability of developing ODD but cannot be considered a necessary cause [59]. In the long term, it is associated with difficulties in accomplishing academic and professional achievements and developing antisocial behaviors [60], which also associates ODD with “insensitivity traits’’ (“callous traits”) [61]. These characteristics, in 8-to-11-year-old children correlate negatively with the total volume of gray and white matter, and modestly with the volume of the right amygdala [61]. A decrease in hippocampal activation has been observed in tasks that require effort to regulate emotions [62].

In the current study, no significant association were observed between trajectories of affective problems and subcortical volume. This null finding is surprising because there is abundant literature that suggests that affective problems such as depression are associated with volume reductions or increases in subcortical structures such as the hippocampus [36] or the amygdala, respectively [63]. A possible explanation of this difference is that it is possible that this relationship emerges later in time. For example, Jaworska et al. (2016) participants were 12–25 years old, and Albaugh et al., (2017) used an even larger range of age. Similarly, a more recent study has a smaller range of age but start when children were 6 years [64].

Strengths and limitations

The main contribution of the present research is that a longitudinal strategy is used, that is, the trajectories of behavioral problems are evaluated throughout the first years of life (18 months, 2.5 and 4 years of age), instead of a transversal approach, in which each measure of the problem level is considered in isolation. We believe that this longitudinal approach allows us to determine more precisely if the brain structures morphometry anatomy is really related to the evolution of the problem. Furthermore, the analysis of our “Missing cases” indicate that the final sample is not biased neither in terms of the CBCL scales nor regarding the confounding factors. Nor does our final sample differ from the “Missing cases” at same-quality MRI.

The results of our research, although consistent with the literature, have some limitations that restrict their generality and interpretation. The first limitation is the smaller size of the final sample compared to the standards in literature. However, very few published studies use a longitudinal strategy like the one used here, which adds extra value to our results. The second limitation is related to the small number of observations available to obtain the developmental trajectories and the temporal distance between them. Unfortunately, it was not possible to obtain cerebral structural images at the ages of 18 months and 2.5 years old, which made it impossible to compare and associate the trajectories of the subcortical structures and the behavioral development. Third, our analytical approach prioritized statistical power over stringent multiple comparisons correction, given the exploratory nature of this initial study on a novel typology of developmental trajectories. While this methodological choice increases sensitivity to detect potentially novel effects, it also elevates the risk of false positives. Consequently, our more suggestive findings should be interpreted with caution. The definitive validation of our conclusions and the precise determination of their effect sizes will require replication in larger, independent samples using confirmatory significance thresholds and rigorous corrections. Such future research will be essential to distinguish robust effects from preliminary findings. Finally, our results are relative to volume of structures, not function. The relationship between structure and function is not yet entirely clear, although it is common to assume that greater volume implies greater functionality. However, functional studies are necessary to strengthen the interpretations regarding the structure-function relationship.

Conclusions

The relationship between the morphometry of subcortical brain structures and behavioral problems seem to be influenced by the trajectories of these problems. The hippocampus, amygdala, pallidum, cerebellum, putamen, and accumbens appear to be the most influenced structures by the history of behavioral problems during the first four years of life. Generally, trajectories with higher levels of problems over time are associated with reduced structure volumes. However, the observed non-linear trajectories suggest that the morphometry of subcortical nuclei, particularly the hippocampus, may be influenced by behavioral problems during a critical period around 2.5 years of age. Building on our previously published findings at this age, the new data presented here further emphasize the importance of supplementing infant formula with bioactive components naturally present in human milk as strategy for reducing later behavioral problems and changes in brain morphology. Based on our previously published data and the present analysis, it is suggested that such supplementation of infant formulas could help protect children from behavioral problems during their early years, promote optimal brain morphology development in childhood, and reduce the risk of mental illnesses in adulthood. The present study also highlights the need for further adequately powered studies with long-term follow-up.

Supplementary Information

Below is the link to the electronic supplementary material.

Abbreviations

ADHD

Attention Deficit/Hyperactivity Disorder

BF

Breastfed infants

CBCL

Child Behavior Checklist

EF

Bioactive compounds-enriched infant formula

FOS

Fructooligossacharides

GLMM

Generalized linear mixed models

MFGM

Milk fat globule membrane

MRI

Magnetic resonance Imaging

LC-PUFAs

Long-chain polyunsaturated fatty acids

ODD

Oppositional Defiant Disorder

RCT

Randomized Clinical Trial

SF

Standard infant formula

T

Typical Scores

Author contributions

Conceptualization, CC, AC; Study design and methodology, CC; AC ; Formal analysis, EC-V, AN-R, AC; Investigation, EV-C, AN-R, FH ; Data curation, EV-C, AN-R, AC ; Writing-original draft preparation, EC-V, AN-R, JAGS ; Writing-review and editing, JAGS, AC, CC ; Supervision, CC; Project administration, CC; Funding acquisition, Rd-C, MTP-H, CC ; All authors have read and approved the final manuscript. The results of this study are likely to be included in the Doctoral Thesis of EC-V, in the context of the Doctoral Program in Psychology at the University of Granada.

Funding

This project has been originally funded by Laboratorios Ordesa, S.L. Contract University of Granada General Foundation, No. 3349 and SMARTFOODS (CIEN) Contract University of Granada General Foundation, No. 4003, Spanish Ministry of Economy, Industry and Competitiveness. Furthermore, the project follow-up has been funded by HORIZON 2020 EU DynaHEALTH Project (GA No. 633595).

Data availability

The data presented in this study are available on request form the corresponding author. The data are not publicy available due to ethical reasons.

Declarations

Ethics approval and consent to participate

The COGNIS study was conducted in accordance with the Declaration of Helsinki and approved by the Institutional Review Board (or Ethics Committee) of Research Ethics Committee of the University of Granada, the Bioethical Committees for Clinical Research of the Clinical University Hospital San Cecilio and the Mother-Infant University Hospital of Granada, Spain. The project was registered at www.ClinicalTrial.govno.: NCT01634464. Informed consent was obtained from all subjects involved in the study.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.