Differentiation between fetal and postnatal iron deficiency in altering brain substrates of cognitive control in pre-adolescence

Participants were recruited by the Brain and Behavior in Early Iron Deficiency study, a collaboration between the University of Michigan and the Children’s Hospital, Zhejiang University School of Medicine. This was a longitudinal cohort project that aimed to evaluate the impact of early ID on neurodevelopment. Healthy full-term infants were enrolled at birth between December 2008 and November 2011 at the Fuyang Maternal and Children’s Health Care Hospital in southeastern China [31]. The infants enrolled in the Brain and Behavior in Early Iron Deficiency study met the following entrance criteria [31]: singleton term, birth weight ≥ 2500 g, 5-min Apgar score ≥ 7, and no delivery complications, maternal or infant health problems, or major congenital deficits. This study was approved by the Institutional Review Boards of the University of Michigan and the Children’s Hospital, Zhejiang University School of Medicine. The demographic information, such as parental age, parental education, parental occupation, and household income, was collected by qualified project personnel when the cohort study was set up (i.e., at 6-week-old and 9-month-old). The family socioeconomic status was assessed by the combination of parental education, parental occupation, and household income (Additional file 1: The computation of family socioeconomic status).

Between October 2019 and October 2021, 115 children aged 8–11 years and their families were invited back to participate in this study. Nineteen children declined to go into the scanner or did not finish the test. Another 25 children were excluded for the following reasons: technical problems (2), low behavioral performance (2), health reasons (1), or too much head motion [20]. After the attrition, 71 participants were included in analyses (mean age = 10.0 years, SD = 0.9, 37 males, 34 females). This Zhejiang University follow-up study was approved by the Institutional Review Board of the Children’s Hospital, Zhejiang University School of Medicine. Before participating in this study, parents and children separately signed consent and assent forms.

Iron status was assessed at birth using cord blood and repeated at 9 months, 18 months, and 8–11 years old using venous blood. Measures of iron status included serum ferritin (SF), zinc protoporphyrin/heme ratio (ZPP/H), mean corpuscular volume (MCV), and/or red cell distribution width (RDW). Hemoglobin (Hb) was used to assess anemia status. Serum C-reactive protein was measured to exclude inflammation that may interfere with the diagnosis of iron status. The iron assay methods were documented in our previous publications [30‐32]. Fetal ID was defined as cord SF < 75 µg/l or ZPP/H > 118 µmol/mol [33‐35]; postnatal ID was defined as normal cord iron status at birth and ≥ 2 abnormal iron measures at 9 months using the following cutoffs: MCV < 74 fl [36], RDW > 14.5% [37], SF < 12.0 µg/l [38], and ZPP/H > 69 µmol/mol [39]; iron sufficiency refers to no ID both at birth and 9 months. Infants with cord SF < 60 µg/L randomly received the iron supplement or placebo according to the design of a small randomized controlled trial [32]. All children with ID anemia at 9 months received iron therapy. In the sample for functional magnetic resonance imaging analyses, there were 20 children with fetal ID, 24 children with postnatal ID, and 27 children with iron sufficiency in early life. Among children with fetal ID, 8 children had ID only at birth, and 12 children had ID both at birth and 9 months. No participant had ID or ID anemia at 18 months and 8–11 years. The 3 groups did not differ in sex, age, head motion, parental age, parental education, parental occupation, household income, and family socioeconomic status (ps > 0.10). More details are provided in Additional file 1.

Participants performed a widely used computer-based behavioral task [22, 40] in the Magnetic Resonance Imaging scanner to measure proactive and reactive cognitive control (Fig. 1, Additional file 1: Protocol of the behavioral task). It included 2 sessions with background cartoon characters of Winnie or Donald. In each session, 2 proactive blocks and 2 reactive blocks were administrated randomly. In proactive blocks, a colored border was presented with one cartoon character as an informative cue, followed by a target picture of food, animal, or objects. In reactive blocks, black borders were presented with cartoon characters as non-informative cues, and colored borders were not presented until the onset of target pictures. Colored borders and cartoon characters jointly determined response rules. According to such rules, participants were asked to respond to target pictures by pressing buttons using the left or right index finger. Thus, proactive blocks allowed participants to figure out the rule of a trial before the onset of targets, but the rule in reactive blocks could not be identified until the onset of targets. There were 20 target pictures in each block, which contained switch and repeat conditions and each condition had 8–11 trials. For switch trials, rules in current trials were changed compared to previous trials, whereas there was no such change in repeat trials. Block duration was 2 min and 13 s with a 17-s break between blocks. Stimuli positions and response rules were counterbalanced across participants.

Participants were scanned in a Siemens 3.0-T scanner (MAGNETOM Prisma, Siemens Healthcare Erlangen, Germany) using a 60-channel coil. Before formal scanning, children learned how to stay still in a mock scanner. Then, participants completed a series of structural and functional scans in the real scanner. The structural images were acquired with a T1-weighted magnetization prepared rapid gradient echo sequence: TR (repetition time) = 2300 ms, TE (echo time) = 2.32 ms, slice thickness = 0.9 mm, voxel size = 0.90 × 0.90 × 0.90 mm³, voxel matrix = 256 × 256, flip angle = 8°, field of view = 240 mm. During task scanning, a total of 546 whole-brain volumes were collected using a T2-weighted gradient echo planar imaging sequence with multi-bands acceleration (TR = 1000 ms, TE = 34 ms, slice thickness = 2.50 mm, voxel size = 2.50 × 2.50 × 2.50 mm³, voxel matrix = 92 × 92, flip angle = 50°, field of view = 230 mm², slice number = 52, MB-factor = 4).

Preprocessing neuroimaging data consisted of several steps. First, slice timing and head motion were corrected using AFNI (Analysis of Functional NeuroImages, Cox, 1996). Then, tissue segmentation was conducted to extract brains using SPM12 (Statistical Parametric Mapping 12, https://​www.​fil.​ion.​ucl.​ac.​uk/​spm/​). Structural and functional images were normalized to the MNI (Montreal Neurological Institute) space using ANTs (Advanced Normalization Tools, http://​stnava.​github.​io/​ANTs/​). Finally, spatial smoothing was conducted with a 5-mm full-width-at-half-maximum Gaussian kernel. For first-level analyses, multiple regression analyses were conducted using the 3dDeconvolve command within AFNI. Events during encoding were convolved with the hemodynamic function to create a total of 8 regressors of interest (2 sessions: session 1 vs. 2; 2 control conditions: proactive vs. reactive; 2 switch conditions: repeat vs. switch) for cue- and target-elicited brain activity. Six head motion parameters were included as covariates. To alleviate the impact of head motion, we censored the volumes with framewise displacement > 0.5 mm. All subjects included for statistical analyses had mean framewise displacement (FD) within 0.13 to 0.44 mm.

Linear mixed modeling (IBM SPSS Statistics 22, IBM Corp., Chicago, IL, USA) was used to analyze behavioral performance as indexed by reaction times and accuracy. In fixed-effect models, Group, Control, Switch, and their interactions were included as predictors. The subject factor was included in the random-effect model. We compared the models that contained only main effects to the ones that also included interactions. We selected the best-fitting models with the lowest Akaike’s Information Criterion (AIC) values. The selected models and their AIC values were presented in Additional file 1. For neuroimaging data, the 3dlme command in AFNI was used to test the main effects of Group, Control, Switch, and their interactions in predicting cue- and target-elicited brain activity. The 3dClustSim mixed model autocorrelation function indicated that clusters with uncorrected p < 0.001 and a minimum of 27 voxel size remained significant after multiple comparison corrections (p_corrected < 0.05). Such correction was carried out for all reported neuroimaging results. The results below focus on differences between groups in behavioral performance and brain activity.

Children in all groups showed faster responses in proactive vs. reactive conditions (β =  − 428.0; 95% CI, − 483.5 to − 372.4; p < 0.001), but there was no significant difference in reaction times between groups. For accuracy, no group showed statistically significant differences between proactive and reactive conditions. The fetal ID group was lower in overall accuracy than both the postnatal ID (β =  − 0.06; 95% CI, − 0.10 to − 0.01; p = 0.02) and iron-sufficient groups (β =  − 0.08; 95% CI, − 0.12 to − 0.04; p = 0.001). More details are provided in Additional file 1: Table S4.

Consistent with behavioral results, all groups showed significant differences in cue- and target-elicited brain activity between proactive and reactive conditions (Fig. 2; Additional file 1: Table S5-S6). Additionally, such condition differences in brain activity varied between groups, as evidenced by significant Group × Control interactions at 6 clusters for cue-elicited brain activity (Table 1, Fig. 3) and at 21 clusters for target-elicited brain activity (Table 2, Fig. 4).