ENVIRONAGE is an ongoing prospective birth cohort established in 2010 (Limburg, Belgium). Mother-child pairs, without planned cesarean section, and where the mother was able to fill out a Dutch language questionnaire, were recruited at birth at the East-Limburg Hospital (Genk, Belgium). When the child reached the age of 4 to 6 years old, they were invited for a follow-up visit at the university study center. Detailed information on the recruitment process at birth is provided elsewhere (20). Between October 2014 and October 2019, 439 mother-child pairs had a follow-up examination. The final ENVIRONAGE population for our analysis consists of 283 children (Supplementary Figure S1). The ENVIRONAGE study protocol is approved by the ethical committees of Hasselt University (Diepenbeek, Belgium, reference no. B371201216090 and B371201524537) and East-Limburg Hospital (Genk, Belgium). It has been carried out according to the Helsinki Declaration, and all mothers provided written informed consent.

Mothers filled out detailed questionnaires at the in order to obtain household and lifestyle information. Data on child sex, maternal and child age, maternal education and average hours the child sleeps were retrieved. Maternal education is coded “low” when mothers did not obtain a diploma, “middle” when obtained a high school diploma, and “high” when obtained a college or university degree. Based on the date of the examination day, a seasonal scale was calculated (Winter is from the 21st of December to the 20th of March, spring from the 21st of March to the 20th of June, summer from the 21st of June to the 20th of September, and autumn from the 21st of September to the 20th of December). Maternal health status regarding the occurrence of preeclampsia, hypertension, infectious diseases or diabetes during pregnancy was retrieved via medical records.

In addition to the detailed questionnaires, mothers also completed the Perceived Stress Scale (PSS) questionnaire to assess their stress levels, and the Strengths and Difficulties Questionnaire (SDQ) developed by Goodman (21) to evaluate the child's behavior using SDQ sub-scales (peer relationship, emotional, conduct, and hyperactivity) and a Total Difficulties Score. Lastly, body anthropometrics of the children, including height (to the nearest centimeter) and weight (to the nearest 0.1 kilograms), were measured by trained staff. Body mass index (BMI) was calculated as the ratio of weight over squared height.

The children's neurocognitive functioning was assessed using the Cambridge Neuropsychological Test Automated Battery (CANTAB) on a touch-screen tablet (CANTAB, Cognitive assessment software, 2019). The CANTAB has been shown to provide reliable measurements of executive functions in children as young as four years old (22). Trained examiners administered four tasks according to standardized instructions. Two of these tasks focused on attention and psychomotor speed: the Motor Screening Task and the Big/Little Circle Task, which assessed response accuracy (in pixels) and/or response latency (in milliseconds). The other two tasks assessed visual recognition and working memory. The Spatial Span task evaluated the maximum number of squares the child could remember in the correct sequence (span length). The Delayed Matching to Sample task measured response latency (in milliseconds), percentage correct (%), and error probability (%) (Figure 1). More detailed information on the CANTAB procedures can be found in Supplementary Text S1.

Child blood was drawn and genomic DNA was isolated from the buffy coat of venous blood stored in EDTA tubes using the QIAamp® DNA mini kit (Qiagen GmbH, Hilden, Germany) following the manufacturer's instructions. DNA quantity and purity were assessed using the Nanodrop spectrophotometer (ND-1000; Isogen Life Science, De Meern, Netherlands) and the Quant-iT™ PicoGreen® dsDNA Assay Kit (Life Technologies, Foster City, CA, USA) using the Omega Fluostar plate reader (BMG LABTECH, Ortenberg, Germany). TL was measured using an adapted qPCR method described by Cawthon (23). Telomeres were measured in triplicate in a single batch. Telomere lengths were expressed as the ratio of telomere copy number to single-copy gene number (T/S) relative to the mean T/S ratio of the entire sample. Detailed sample collection procedures and TL-assay specifications, and reliability parameters are provided in Supplementary Text S2 and have been previously provided for the ENVIRONAGE birth cohort (24).

Data was analyzed using JMP Pro 16.2.0 software. Participants' characteristics were presented as means (SD) or frequencies (%). All latency data, i.e., response latency of the Motor Screening Task, Big/Little Circle task, and Delayed Matching to Sample task were log-transformed (log10), to better comply with assumptions of model linearity. The association between the different cognitive outcomes and TL was assessed using multivariable-adjusted general linear models. The threshold for statistical significance was set at a 95% confidence level (p < 0.05). We adjusted our models for a priori-selected covariates that may be associated with both the dependent (cognitive outcomes) and independent variables (TL at 4–6 years old). In model 1, this includes the child's sex and age. In model 2, we further adjusted for the child's BMI, average sleep hours of sleep the child has on a typical day and night, maternal education, and the examination season. The normality of the residuals was visually evaluated by Q-Q plots. Estimates are presented as the difference in cognitive outcome for a 1-IQR increase in TL. For data that were log-transformed, we back-transformed the estimates and expressed them as a percentage difference.

To assess the robustness of our findings, we performed several sensitivity analyses on the computerized measures of cognitive function. First, we excluded children who showed any signs of disinterest or were distracted while taking the cognitive tests to account for possible response errors. Second, we adjusted our models for the child's behavior over the past six months by using four sub-scales of the SDQ questionnaire: the peer relationship and emotional subscales, and the conduct and hyperactivity subscales, each scored as normal vs. not normal. Further we additionally adjusted our models for PSS scores and maternal health conditions during pregnancy like diabetes, preeclampsia, hypertension or infections. Lastly, we stratified by sex.

Characteristics of the 283 participants are provided in Table 1. Children were on average (SD) 4.57 (0.41) years old, and approximately half of them were boys (51.9%). Most participants were of European ethnicity (94.0%), with an average BMI of 16.05 (1.26) kg/m². The season in which the examinations took place was for 64 (22.6%) participants during autumn, 88 (31.1%) participants during spring, 71 (25.1%) participants during summer, and 60 participants (21.2%) during winter. The relative telomere length ranged from 0.64 to 1.47. Accompanying mothers were on average 30.1(4.4) years old, and the majority had a college education or higher (62.9%).

SDQ data was derived from 256 participants and categorized results into “normal” and “not normal” for each subscore. For the majority of participants (82%), the emotional symptom score falls within the “normal” range. The same is true for the hyperactivity score, where 77% exhibited behaviors classified as “normal”, signifying a typical level of hyperactivity. Further, 78% of participants demonstrated “normal” conduct, indicating a standard range of behavioral conduct. Within the domain of peer relationships, 84% of participants showcased “normal” patterns, underscoring healthy social interactions. Moreover, 95% of participants exhibited “normal” prosocial behaviors, meaning they have a high degree of positive engagement and cooperation with others. The combination of these individual sub-scales resulted in a comprehensive assessment of participants’ behavioral characteristics, quantified by the Total Difficulties Score. 86% of participants received a “normal” classification according to this measure.

The outcomes for the four CANTAB tasks are summarized in Table 2. The median (p25–p75) response latency of the Motor Screening task referring to the average time the child needed to touch the target successfully, was 964.4 (803–1148.9) milliseconds. The median error, measured as pixel units from the target center made during the same task, was 14.3 (12.2–16.0). In the Big/Little circle, the response latency was similarly calculated as the average time to select the correct circle and was observed to be 1092.4 (965.7–1228.6) milliseconds. During the Spatial Span task, participants demonstrated an ability to accurately reproduce a median of 3 (2–3) squares in the correct sequence. Finally, in the Delayed Matching to Sample task, participants needed a median time of 4133.1 (3282.8–5134.0) milliseconds to accurately select the correct pattern on the first try for all presented sample patterns. The percentage of trials answered correctly on the first try was 45.0 (35.0–55.0)% and the probability of error given a previous correct answer was 0.6% (0.4–0.7).

In unadjusted analysis, we only observed a negative association between TL and the error made during the Motor screening task (Figure 2). A 1-IQR increase in TL was associated with a 0.58-pixel unit lower pixel error (95% CI: −1.04 to −0.12; p = 0.014). After adjustment for child sex and age the estimates were comparable and a 1-IQR increment in TL was associated with a 0.57-pixel unit decrease. In fully adjusted models this association remained significant (95% CI: −1.04 to −0.10; p = 0.017). No other cognitive outcomes related to attention and memory were found to be associated with childhood TL (Table 3).

Results of the sensitivity analyses are shown in Supplementary Table S1. The association between TL and the error made during the Motor screening task remained robust after the exclusion of children who were disinterested or unfocused during the tests, and for additional adjustment for SDQ results, PSS results and maternal health conditions during pregnancy. After stratification for child sex, a 1-IQR increment in TL was associated with a 0.86 decrease in this error in boys and a and 0.49 decrease in this error in girls, and no significant interaction was observed (p-interaction = 0.93).