Infants included in the current investigation participated in two multisite longitudinal studies of brain and behavioral development encompassing three groups: 1) infants with FXS, 2) infants with at least one older full-biological sibling with autism (family history; FH), and 3) infants with a typically developing older sibling and no siblings diagnosed with autism or FXS (control). Infants with FXS were recruited via postings in list serves, family advocacy conferences, research registries, and through a FXS specialty clinic. By design, the intended enrollment age was 6 months, though FXS and FH infants were allowed to enroll at 12 months, and in rare cases 24 months, to increase recruitment in these groups. FXS diagnosis was confirmed by medical records or genetic testing, and full-mutation FXS was confirmed via genetic testing record for all but three participants, two of whom we could not obtain records for, and one participant with mosaicism. We re-ran all analyses excluding the participant with confirmed mosaicism, and effects were nearly identical, so we maintained inclusion of this infant in analyses. Diagnosis of ASD in the older sibling of the infants in the FH group was confirmed through records from a clinical provider and supported by administration of the Autism Diagnostic Interview-Revised (ADI-R) [37] by research staff. Exclusionary criteria included genetic, medical, neurological, and sensory conditions known to affect development in the infant (other than FXS in the FXS group), premature birth or low birth weight (< 2,000 grams), in utero exposure to exogenous compounds likely to adversely affect brain development, contraindication for magnetic resonance imaging (MRI), infant adoption, non-English-speaking family, and any history of intellectual disability, psychosis, schizophrenia, or bipolar disorder in a first degree relative[18, 38]. Research protocols were approved by the Institutional Review Boards at the data collection sites: the University of North Carolina at Chapel Hill, the University of Washington, Children’s Hospital of Philadelphia, and Washington University in St. Louis. Parents provided informed consent for their infants to participate.

Participants in the FH group were given a clinical best estimate judgment for autistic disorder or pervasive developmental disorder not otherwise specified at the 24-month visit using DSM-IV-TR [39] criteria, the Autism Diagnostic Observation Schedule–Generic (ADOS) [40], and the ADI-R, by the clinician who conducted the behavioral assessments and confirmed via video review by a senior psychologist or psychiatrist who was blind to FH status. Participants meeting criteria for either disorder were classified as having an ASD outcome in the study, consistent with DSM-5 [41] criteria which became available after the original study concluded.

ASD classification based on either the DSM, ADOS, or both, at 24 months was only available for 52% of participants in the FXS cohort (n = 16). Missing or unreliable data on the remaining FXS infants (n = 15) was due to a range of factors (e.g., fatigue/fussiness, low mental age, and inconsistent ability to capture reliable information from the prompts in the assessments) or loss to follow up at the 24-month timepoint. Thus, ASD outcome within the group of FXS infants was not analyzed in this study. This is consistent with prior work from our team involving these infant groups [19]. See Table S12 in the Supplement for 24-month ASD data availability for each FXS participant. The final sample included: 77 FH-ASD infants (63 male, 14 female); 280 infants with family history who were not diagnosed with autism (FH-nonASD; 156 male, 124 female); 31 infants with FXS (25 male, 6 female); and 154 control infants (91 male, 63 female).

Participants were administered a comprehensive battery of early developmental behavioral assessments at 6, 12, and 24 months of age, including parent-report and direct assessment measures of adaptive behavior, repetitive behaviors, social communication, and cognitive development. The primary measure of interest in this investigation is the Mullen Scales of Early Learning (MSEL) [42]. The MSEL is a comprehensive developmental assessment that covers the domains of language, motor, perceptual abilities, and cognition in infants and young children. Subtests of the MSEL include receptive language (RL), expressive language (EL), fine motor (FM), gross motor (GM), and visual reception (VR). One control infant displayed floor effects (e.g., T-scores of 20) at one or more timepoints on each domain of the MSEL, and was therefore excluded from the current investigation for suspicion of global developmental delay, not representing typical development.

Developmental trajectories for FXS, FH-ASD, FH-nonASD, and control infants were examined for each domain of the MSEL (RL, EL, FM, GM, and VR) to identify whether differential skill development in specific behavioral domains at specific ages emerged among FXS and FH-ASD groups in infancy.

The primary analyses utilized raw scores from the MSEL. Raw scores reflect the number of items successfully completed on each subscale, with higher scores indicating more advanced skill development. Raw scores represent a constant rate of development within and across subscales which allow for best group comparisons. Raw scores are better suited for studying the group of FXS infants in the current study than standard scores given their greater range and ability to overcome floor effects often observed using standard scores (e.g., T-scores of 20) on the MSEL [20]. Raw scores additionally allow for interpretation of developmental trajectories in the context of growth, where most infants’ scores will improve over time as their skills develop, in contrast to standard scores which must be interpreted in the context of typical development represented by a mostly flat profile over time (i.e., an average person having an average standard score throughout development). Though age equivalent scores have psychometric limitations [43], supplemental analyses utilizing age equivalent scores from the MSEL are included as part of the current study, to provide a direct comparison to other published reports of infants with FXS that utilized age equivalent scores (e.g., Wheeler et al., 2021 [20]).

To expand upon our previous report of group differences on the MSEL Early Learning Composite (ELC) measure [19], and for comparison to other work [20], verbal developmental quotient (VDQ) and nonverbal developmental quotient (NVDQ) scores were generated and utilized in a subset of analyses focused on overall cognitive domains. The VDQ is a language-based composite measure, calculated by averaging the age equivalent scores of RL and EL domains, dividing by chronological age, and multiplying by 100. The NVDQ is calculated by averaging the FM and VR age equivalent scores, dividing by chronological age, and multiplying by 100. The NVDQ has been found to be more predictive of school-aged executive functioning than language-reliant developmental quotients [44] and has been found to be associated with specific deficits in white matter development (i.e., lower fractional anisotropy in the left and right uncinate fasciculus), characteristic of FXS at 12 months of age [18].

The analysis of longitudinal group differences using the VDQ and NVDQ allows for a parsing of previous findings of FXS and FH-ASD divergence on the ELC [19], which comprises visual reception, fine motor, and both language scales, into more specific skill composites where verbal and nonverbal abilities are analyzed separately while maintaining the use of composites which include information from multiple domains. Additionally, longitudinal analysis of NVDQ throughout infancy between the groups included in the current study may eliciduate group differences very early in development on this measure of cognitive ability which has been previously shown to be related to early brain differences between FXS and control infants [18].

Separate mixed-effects models estimated the changes in MSEL RL, EL, FM, GM, VR, VDQ, and NVDQ scores from 6 to 24 months of age. These models included random intercepts for each participant. Independent variables included group membership, timepoint, sex, and study site. These latter two variables were included to control for effects related to differences in sex ratio among groups and any potential unaccounted variance related to assessment site. Models also included the interaction of group by age. Trajectories of development for FXS and FH groups were examined in reference to the control group, to understand differences between FXS and FH infants in comparison to typical development. Missing data were assumed missing at random with patterns of missingness not varying by group or other covariates. Approximately 72% of the participants had no missing MSEL data at any of the visits, and 13% of the sample had missing data at the 6-month visit. The rest of the sample had some missingness at either the 12-month visit, 24-month visit, or a combination of two visits among the 6-, 12-, and 24-month visits. Missingness was handled via multiple imputations (m = 5) obtained through multivariate imputation by chained equations. This method fits a sequence of regression models for the missing value and imputes the missing data from its predictive distribution [45]. These chained equations incorporated every covariate in the mixed-effects model to predict the missing value. Predicted values were produced using predictive mean matching with fully conditional specification [46]. Fixed-effects estimates were pooled according to Rubin’s rules [47]. Imputation of missing scores provided more power and mitigated the possibility of bias introduced from listwise deletion of data [48,49,50,51] (see Table 1 for MSEL availability by group and timepoint before imputation). Sensitivity analyses were performed to determine comparability of imputed versus unimputed analyses. The two sets of estimates were deemed comparable (see Table S2 in the Supplement for unimputed results). Least-squares means were estimated to compare groups at different visits with Cohen’s d effect sizes reported. Cohen’s d was calculated using the lme.dscore() function from the EMAtools package [52]. Multiple imputations were estimated using the mice package in R [53], mixed-effects models were estimated using the nlme package [54], and least-squares means were estimated using the emmeans package [55].

Because cognitive level varies greatly within FH-ASD infants and is related to ASD severity [8] and behavioral trajectories [10], we further divided the FH-ASD group to assess the effect of low cognitive level within FH-ASD on MSEL developmental trajectories in comparison to the FXS group, which is characterized by low cognitive level. We parsed FH-ASD infants by whether they had a MSEL ELC score of 70 or lower, which equates to 2 standard deviations below the mean (FH-ASD-Low) at the latest available timepoint in infancy. Group trajectories between FH-ASD-Low (n = 24), FH-ASD-Avg/High (n = 58) and FXS were compared (most of the FXS infants met criteria for low cognitive level, characteristic of this group, see Tables S8, S9, S10 and S11 in the Supplement).

As described above, to provide alternative interpretations to our group comparisons and to investigate whether group trajectories differed dependent on score type used, all analyses were run with age equivalent scores and are presented in Tables S3 and S4 and Figs. S1, S2, S3, S4 and S5 of the Supplement.

Due to the inherent variability in behavioral phenotypes between males and females with FXS, supplementary analyses were conducted to investigate whether patterns of group differences (FXS vs. FH-ASD) in male infants were similar in nature to the combined sample of males and females utilizing both raw and age equivalent scores. Females were not investigated separately due to low sample size for FXS (n = 6); this approach is consistent with other studies of infants with FXS [24].