Content area
Maturation of the auditory system in early childhood significantly influences the development of language-related perceptual and cognitive abilities. This study aims to provide insights into the neurophysiological changes underlying auditory processing and speech-sound discrimination in the first two years of life. We conducted a study using high-density electroencephalography (EEG) to longitudinally record cortical auditory event-related potentials (CAEP) in response to synthesized syllable sounds with pitch/duration change in a cohort of 79 extremely and very preterm-born infants without developmental disorders. EEG were recorded at 6 timepoints from term to 24 months corrected age, using a pseudorandom oddball paradigm. We found that the infant-P1 component of CAEP showed decreasing latency with age and more focalized cortical source stabilizing in the left primary auditory cortex by 6 months. By 6 months, a negative infant-N1 component emerged, its amplitude increasing with age and source localization showing increasing distribution over the left temporal, parietal and frontal lobes. Mismatch responses demonstrated significant differences in auditory discrimination capabilities starting from 6 months, indicating the infants' ability to detect phonetic differences. There was no correlation between infant-P1 latency, infant-P1 amplitude or mismatch response at term age and gestational age. This study suggests that cortical sound detection occurs very early and is not significantly influenced by the extent of prematurity but rather by corrected age. Early sound detection is followed by cortical sound content processing from about 6 months, with gradual organization along the cortical auditory dorsal stream and mirror neuron system in the first two years of life. Auditory discrimination of speech sounds also significantly changes from around 6 months of age.
The maturation of the auditory system in early childhood is a dynamic process that underlies the development of perceptual and cognitive abilities related to language ( Choudhury and Benasich, 2011). During this period, intricate neurobiological changes shape the auditory pathways, fostering the establishment of efficient and refined neural circuits. Better understanding of auditory maturation is required to address the complexity of sensory processing in the developing brain ( Pelc, 2023). The foundational stages of auditory processing are characterized by neurophysiological changes in the auditory cortex, where synaptic connections are established, refined, and pruned ( Eggemont and Moore, 2012). During the first months of life, the cortical areas encompassing the auditory system undergo rapid development of neuronal connections (Yu and Goodrich, 2014). Auditory cortical areas show increasing differentiation through infancy, as evidenced by gradual decrease in the latency of auditory responses recorded by electroencephalography (EEG), and structural changes in underlying neural pathways shown on diffusion MRI ( Adibpour et al., 2020).
Behavioral and psychophysical studies have suggested that reaction to auditory stimuli can be operational both before and after term birth. The auditory system begins developing in utero ( Graven and Brown, 2008). Behavioral studies suggest that fetuses can respond to auditory stimuli as early as 27 weeks gestational age but more consistently by 35 weeks ( Shahidullah et al., 1994). Exposure to sounds, particularly the mother's voice, plays a critical role in shaping auditory preferences and processing capabilities ( Ćirović et al., 2023; Vogelsang et al., 2023). Prenatal exposure has been hypothesized to allow fetuses to become attuned to speech sounds ( Partanen et al., 2013), including the rhythmic and melodic patterns of language. For example, newborn infants’ cry production reflects the prosodic features of their native language ( Mampe et al., 2009).
Newborn infants, as young as four days old, can show some perception of well-formed syllables ( Ramus et al., 2000; Guttorm et al., 2001) and they can differentiate between sentences in their native language and those in a foreign language ( Dehaene‐Lambertz et al., 2002). Infants’ speech sound discrimination abilities continue to develop, particularly around 6 months ( Lee et al., 2023). As infants grow, they acquire the ability to discriminate between speech sounds from all languages over the first few months of life, and subsequently (from about 6 months) specialize in discriminating sounds that are specific to their own language environment through perceptual narrowing ( Feldman et al., 2021; Kuhl et al., 2005; Werker, 2024).
Perception and cortical connections develop slowly in the second half of the first year of life, and continue to mature throughout childhood, with early learning of lexicon and use of prosody to facilitate learning of the syntactic structure of language during the toddler years ( Gervain, 2022). Long before the emergence of a child's first spoken words, neural activity shapes the base of linguistic competence ( Friederici, 2006). In particular, the processing of phonemes, i.e. small units of sound that form the building blocks of speech, is an essential component of language development ( Friederici, 2005). The neural processes governing the discrimination and processing of non-word syllables in the preverbal stage offer a window into neurophysiological development.
Preterm birth is defined by the World Health Organization, as birth occurring before 37 weeks of pregnancy are completed, with subcategories based on gestational age: extremely preterm (less than 28 weeks), very preterm (28 to less than 32 weeks), and moderate to late preterm (32 to 37 weeks). The early postnatal environment can exert various influences on auditory development. However, preterm infants may show differences in early auditory experience ( Key et al., 2012), which might in turn alter their ability to discriminate sounds later in life (Park et al., 2016). For example, predictable auditory environments have been associated with enhanced attention and cognitive outcomes in infants, while unpredictable noise exposure can lead to stress and dysregulated physiological responses, negatively impacting neurodevelopment ( Wass et al., 2019; Werchan et al., 2022).
The spatiotemporal dynamics of the neural events associated with exposure to speech sounds can be approached by noninvasive EEG. In infants, using high-density EEG with a large number of channels (e.g. 64) enhances spatial accuracy to physiologically relevant levels to document neurodevelopmental trajectories ( Dan et al., 2015; Pelc et al., 2022). This can be used to record cortical auditory event-related potentials (CAEP) in response to acoustic stimuli. CAEP are generated from auditory thalamocortical and corticocortical pathways, including the primary, secondary and tertiary auditory cortex and the association auditory cortex ( Eggermont and Ponton, 2002; Lippé et al., 2009; Costa et al., 2020). They reflect the neural processing of sounds at various levels of the auditory system, including the detection and discrimination of syllables. CAEP show distinctive components in succession. The first peak, which is classically seen around 50–100 ms after stimulus onset in older children (from about 8 years of age) and adults, is not recorded in infants. The first peak seen in infants is recognized as positivity peaking with a latency around 200–300 ms poststimulus ( Lippé et al., 2009). We refer to this CAEP component as infant-P1 (iP1) throughout this paper. It is followed by a negative component that gradually emerges during infancy ( Lippé et al., 2009), which we refer to as infant-N1 (iN1) throughout the paper. It will be important to keep in mind this semantic choice of referring to the first occurring, positive component as iP1, and the subsequent, negative component as iN1, which is phenomenologically consistent at this stage of development, and not confuse them with nomenclature widely used in the adult neurophysiological literature, which calls the equivalents to these components with poststimulus latencies typically around 160–200 ms and 200–300 ms, respectively, P2 and N2, and identifies earlier components seen in adults but not consistently in infants at around 50–80 ms and 80–160 ms poststimulus, respectively, as P1 and N1.
Longitudinal study of CAEP has the potential to provide insights into infants’ ability to detect, process and discriminate between phonetic differences along developmental trajectories of syllable processing. It can also support some prediction of developmental cognitive and language outcomes ( Depoorter et al., 2018; Kailaheimo-Lönnqvist et al., 2020). Longitudinal studies of speech sound discrimination in infants, using mismatch deviation event-related potentials (mismatch responses, MMR), have documented development trends pointing to critical changes around the age of 6 months ( Virtala et al., 2022; Werwach et al., 2022).
To our knowledge, few electrophysiology studies have examined sex differences in the maturation of auditory cortex function in infants with sufficient rigor. Using magnetoencephalography, a cross-sectional study of 48 typically developing girls and 66 typically developing boys aged 2–24 months found no sex differences in the latency of the equivalent of the EEG iP1 response ( Y Chen et al., 2023).
The existing literature remains unclear with regard to the possible influence of gestational age on the development of CAEP. In infants born extremely preterm, iP1 amplitude recorded around 35 weeks gestation tended to be reduced when neonatal complications occurred, such as bronchopulmonary dysplasia or retinopathy of prematurity ( Suppiej et al., 2015). Melo et al. (2016) found longer iP1 latency in infants born between 26 and 36 weeks than in fullterm infants, which they ascribed to differences in maturation. Some other studies have documented shorter iP1 latency in infants born preterm compared to fullterm infants, suggesting accelerated development, which authors hypothesized to be related to differences in myelination or earlier exposure to environmental auditory stimulation ( Pasman et al., 1992; Gottschalck Cavalcanti et al., 2020). However, the levels of environmental sound and language exposure of extremely and very preterm infants in the neonatal intensive care unit (NICU) could be detrimental to development ( O'Callaghan et al., 2019).
In the present study, we prospectively recorded CAEP using high-density EEG in response to two different non-word syllables presented with an oddball paradigm in a cohort of preterm-born infants without developmental disorders from term age to 24 months corrected age in order to document the maturation of auditory processing of syllables. We studied the evolution of the topography of responses, cortical sources, and automatic discrimination using MMR ( Herman et al., 2023). With regard to accruing knowledge on maturation of cortical circuits involved in auditory processing, as observed in prior studies using electrophysiological and neuroimaging techniques, which have shown a greater differentiation of auditory regions with increasing age, we hypothesize that (1) the processing of auditory information becomes increasingly localized within the auditory cortex as children age, particularly from the age of 6 months, with enhanced lateralization in the left hemisphere. Considering previous evidence showing that the brain's ability to discriminate speech sounds becomes more specialized and efficient during this critical period, coinciding with the development of language-related neural pathways, we hypothesize that (2) auditory discrimination of speech sounds undergoes a significant refinement starting at 6 months of age, characterized by improved sensitivity to speech contrasts. In addition, we explored the influence of sex on the latency and amplitude of CAEP. Finally, we explored if gestational age influences cortical processing of auditory information at term age, as the exiting literature is unclear about the issue
2 Materials and methods2.1 Participants
We performed a prospective exploratory study between 2016 and 2018 in two tertiary university hospitals in Brussels, Belgium. We aimed to evaluate neurophysiologic brain maturation from term age to 24 months corrected age using high-density EEG in children born preterm before 32 weeks of gestational age. Infants born before 32 weeks face significantly higher risks of mortality, morbidity, and long-term neurodevelopmental challenges, making this an essential threshold for both clinical management and research. We anticipated that it would make our proposed testing more relevant to families and treating clinicians, and this was deemed ethical by the Institutional Review Board. Age is given as corrected age (i.e. from term age) throughout the paper. Exclusion criteria were severe congenital malformations, genetic neurodevelopmental syndromes, congenital infections, neonatal hypothyroidism, neuromuscular diseases, and neurodegenerative diseases. Infants with abnormal brain MRI (ventriculomegaly, white matter changes, or other parenchymal lesions), epilepsy, or significant developmental delay (−2SD) in at least one domain (cognitive, motor or language) of the Bayley Scales of Infant and Toddler Development (3rd edition) were secondarily excluded from analysis.
2.2 Stimuli and experimental paradigmWe used a pseudorandom oddball paradigm in which a series of repeated synthesized [da] syllable sounds (220 Hz fundamental frequency, 100 ms duration) were used as standard stimuli (85%) and randomly occurring [dɑ] syllable sounds (110 Hz fundamental frequency, 265 ms) were used as deviant stimuli (15%). The first and second formant of the standard stimulus [da], were stable from 25 ms (which corresponds to the beginning of the phoneme [a]) and their mean were respectively 831 and 1427 Hz. The first and second formant of the deviant stimulus [dɑ] were stable from 45 ms (which corresponds to the beginning of the phoneme [ɑ]) and their mean were respectively 722 and 1060 Hz.
The choice of speech syllable stimuli differing both in frequency and duration of [da] and [dɑ] was based on predictable linguistic patterns in Belgian French, the language to which the participants were exposed. The pitch differences between [da] and [dɑ] were verified by statistical analysis (Welch's test) of a sample of recordings of 10 words containing either of these syllables in various positions from a female and a male native Belgian French speakers (KP, BD).
Stimuli were generated by a Dell PC using eevoke software (ANTneuro, Hengelo, Netherlands). A total of 595 standard stimuli and 104 deviant stimuli were delivered binaurally to each child in a quiet room at 77 dB intensity (sound pressure level) through two speakers located laterally at a distance of 50 cm of the subject's head, with offset-to-onset interstimulus interval of 1200 ms. This interstimulus interval was selected to balance the duration of auditory memory trace and the ability to detect changes in the stimulus in infants, regardless of their arousal level ( Bartha-Doering et al., 2015).
2.3 EEG recording and preprocessingEEG recordings were acquired at the following corrected ages: at term (40 weeks gestational age ± 2 weeks), 3 months (± 2 weeks), 6 months (± 2 weeks), 12 months (± 2 weeks), 18 months(± 2 weeks), and 24 months (± 2 weeks). EEG was recorded in a quiet environment using the Active Two system (Biosemi, Amsterdam, Netherlands), at a sampling rate of 2048 Hz ( Fig. 1 a,b). Recording sessions lasted 16 minutes. We used EEG headcaps with 64 electrodes based on the international 10/10 system or 128 electrodes based on the BioSemi equiradial system according to the infant's head circumference. Summed ear lobe potentials were used as reference.
In order to compare EEG data between age groups, the 128-channel EEGs in the equiradial system were reduced to 64 channels equivalent to the 10/10 system by retaining the signal from the common electrodes to both systems and interpolating the others using the spherical spline method ( Perrin et al., 1989). EEG signal analysis was performed on MATLAB R2021a using EEGLAB ( Delorme and Makeig, 2004). The sampling rate was first reduced from 2048 Hz to 512 Hz using linear interpolation. The EEG signal was then filtered using Butterworth zero-phase filters with high-pass cutoff frequency of 0.5 Hz and low-pass cutoff frequency of 30 Hz.
Visual inspection of the epochs (-500–1000 ms) was performed and epochs with artifacts were manually rejected. These artifacts include repetitive eyeblinks, electrode disconnection, and severe movement and muscle artifacts. Epochs were excluded when 5 or more channels were contaminated. Individual contaminated channels were interpolated using the spherical spline method ( Perrin et al., 1989). We avoided interpolation in the frontotemporal region, where auditory responses are expected to be observed. We used the -500 to 0 ms window to apply baseline correction.
We included for analysis each channel individually and only channels for which CAEP components were reliably observed, based on visual inspection. CAEP amplitudes were quantified as mean amplitude, calculated by averaging the voltage values across all sample points within a 40 ms time window centered around the component peak. CAEP latencies were quantified using the 50% area latency measure, which calculates the time point that divides the area under the curve of the iP1 and iN1 components into two equal parts within the same time window ( Luck, 2014). These analyses were performed on grand average across all participants for each of the 64 channels individually for each time point.
Mismatch response (MMR) was obtained by subtracting CAEP evoked by standard stimuli from those evoked by deviant stimuli.
2.4 Source localizationThe Brainstorm toolbox ( Tadel et al., 2011) was used for CAEP source estimation, based on protocols published previously ( Stropahl et al., 2018). Grand average across all participants of CAEP related to the standard ([da]) stimulus were used for source analysis for each age group, considering each of the 64 channels individually. Age-adapted infant MRI templates from O'Reilly et al. (2021) and Biosemi 64 10–10 electrode location file were used to compute the forward head model using boundary element method as developed in OpenMEEG ( Gramfort et al., 2010; Kybic et al., 2005). MRI segmentation to generate boundary element method surfaces using Brainstorm was performed using default number of vertices per layer (scalp 1922, outerskull 1922, innerskull 1922). We used the tissue conductivity values used in infants by O'Reilly et al. (2021) (gray matter: 0.3300 S/m; skull: 0.0041 S/m; scalp: 0.3300 S/m).
Source estimation was performed using the minimum-norm imaging method after noise covariance matrix was computed from recording over the -500 to 0 ms baseline window block by block in order to avoid effects of slow shifts in the data ( Tadel et al., 2011). Minimum-norm imaging estimates the amplitude of brain electrical currents at each grid point derived from the forward head model. Those electrical currents (matrices) were normalized using the dynamical statistical parametric mapping approach ( Dale et al., 2000) resulting in unitless z-scores maps of spatiotemporal brain activity. We selected the option of constrained dipole orientation which models one dipole perpendicular to the cortical surface for each vertex ( Tadel et al., 2011).
2.5 Statistical analysisDescriptive and bivariate statistics were performed with R-studio version 4.3.2 (R Foundation for Statistical Computing, Vienna, Austria). Continuous quantitative variables were analyzed in terms of mean and standard deviation (SD) if normality of the distributions was confirmed by graphical analysis, or in median and interquartile range (IQR) if not. Binary qualitative variables were described in terms of percentage. To test associations between variables and group assignment we used t-test and Chi-squared test after verifying the conditions of application.
In addition, we performed t-tests with false discovery rate correction for multiple comparisons ( Benjamini and Hochberg, 1995) on grand average CAEP across all participants for each of the 64 channels at the 6 timepoints to determine whether iP1, iN1, and MMR differed significantly from the baseline ( p<0.05).
We performed repeated measures ANOVA to study the evolution of the CAEP latency and amplitude over time. The Greenhouse-Geisser or the Huynh-Feldt corrections were used (in function of the epsilon value) in order to take the sphericity violation into account. We used the Tukey correction to correct for the risk of type I error inflation due to multiple comparisons. In order to consider the evolution over time as well as the sex dependency, we performed within-between ANOVA. The same corrections for sphericity and multiple comparisons were applied as above.
Cluster-based permutation, a nonparametric tests developed by FieldTrip ( Maris and Oostenveld, 2007), was used in Brainstorm (version 2024 January) as statistical method to compare MMR (deviant – standard stimuli) in the latency range from 0 to 800 ms poststimulus in the time and space domains in different age groups solving the multiple comparisons problem. MMR was compared between the different age groups and the term age for each channel at each timepoint, using two-sided t-tests. The cluster-alpha representing the threshold for considering a sample as a candidate member for one of the clusters was set at 0.05. The clusters are formed by those candidates, taking into account their spatial and time adjacency, and the fact that the statistic of the t-test has a positive or negative sign. A minimum number of 2 of neighborhood channels was required for a selected sample to be clustered. The statistics of each cluster was calculated by taking the sum of the t-values within each cluster. A permutation test was then performed. The reference distribution for this test is obtained by means of the Monte Carlo method. The number of randomizations was set at 1000 in order to get the distribution of the maximal cluster-level statistic that may occur by chance. To reject or accept the null hypothesis, we compared the observed cluster with the largest statistic with the obtained reference distribution and calculated the proportion of random partitions with a larger statistic. This proportion is the p-value.
The significance level was adjusted for multiple comparisons performed for the different age comparisons using false discovery rate correction.
In order to explore the influence of gestational age on early processing of auditory information, we performed correlation tests between gestational age and iP1 latency and amplitude as well as MMR on the recording performed at term age. This test was done during active (REM) sleep. Pearson's and Spearman's correlation coefficient were used to quantify the relationship between continuous variables of normal and non-normal distribution.
2.6 Ethical aspectsThe Ethics Committees of Hôpital Universitaire des Enfants Reine Fabiola and Centre Hospitalier Universitaire Saint-Pierre, Brussels, Belgium, approved the study. The research was conducted in accordance with the principles embodied in the Declaration of Helsinki and in accordance with Belgian and EU statutory requirements. Signed informed consent to participate in the study was obtained from all participants’ parents or legal guardians.
3 Results3.1 Population data
High-density EEG were recorded prospectively from 102 infants. By the age of 24 months, 84 of them (40 girls, 44 boys) met all inclusion criteria, including normal development, audition and brain MRI. Mean gestational age at birth was 29 weeks (SD 2, range 24–31). Mean birth weight was 1195 g (SD 290, range 500–2100). Among them, 78 had received antenatal corticosteroids for lung maturation in anticipation of preterm birth; this occurred without any clinical complications. All 84 infants were included for data analysis and their parents/caregivers agreed to attend all six scheduled appointments. Seventeen participants completed only one of the recording sessions, 21 completed two sessions, 12 three sessions, 19 four sessions, 6 five sessions, and 6 participants completed all six recording sessions (attendance tended to decrease with age), whereas 3 children did not come to any of the scheduled sessions ( Fig. 1c). The number of EEG recorded at each timepoint was 49 at term age, 48 at 3 months, 47 at 6 months, 35 at 12 months, 29 at 18 months, and 29 at 24 months. Of these, the following numbers of EEG recordings were withdrawn due to artifacts at each timepoint: 7 at term age, 5 at 3 months, 7 at 6 months, 2 at 12 months, 2 at 18 months, and 4 at 24 months. There were no statistical significant differences in terms of gestational age, birth weight and sex between infants whose EEG were analyzed and those whose EEG were not analyzed because of lacking data or presence of excessive EEG artifacts. The numbers of CAEP epochs that were analyzed (i.e. those that remained after preprocessing) were similar at the various age points ( Table 1 ).
3.2 Longitudinal evolution of cortical responsesFig. 2 shows superimposed CAEP traces over all epochs in all children over 64 channels at each timepoint for both the standard and deviant stimuli. Channels exhibiting iP1 and iN1 components following the standard stimulus that were significantly different from the baseline are shown for the 6 timepoints on Supplementary Figures 1 and 2, respectively. At term age, CAEP following standard stimulus consisted of a positive component (iP1). The average across channels of the iP1 latency decreased by 43 ms (SEM=0.16) between term age and 3 months, 39 ms between 3 and 6 months (SEM=0.14), 65 ms between 6 and 12 months (SEM=0.22), 18 ms (SEM=0.13) between 12 and 18 months, and 12 ms (SEM= 0.10) between 18 and 24 months (p<0.001 between all successive timepoints). ( Fig. 3 ). The decrease in iP1 latency with respect to age was seen in girls between term age and 24 months and in boys between term age and 18 months (p<0.001 between all successive timepoints). It occurred in a different way through time for girls in comparison with boys (p<0.001) ( Fig. 4 ).
The average across channels of the iP1 amplitude increased by 5.1 µV (SEM=0.27) between term age and 3 months (p<0.001). The further increase in amplitude between 3 and 6 months was not significant. It brought the mean amplitude of iP1 to 7.3 µV (SEM=0.36). Thereafter, iP1 amplitude decreased in average by 3.8 µV (SEM=0.21) between 6 and 12 months (p<0.001) ( Fig. 3). The evolution of iP1 amplitude with age was similar in girls and boys, though a significant difference was observed at the 18 months timepoint (p<0.001) ( Fig. 4).
Fig. 5 illustrates the scalp topography of the peak of iP1. High amplitude iP1 peak (>75% maximum amplitude) was seen widely over the frontocentrotemporal regions at term age, with maximal amplitude in frontal areas. From 3 months of age, high amplitude iP1 peak was less widely distributed, with maximal amplitude varying between the temporal regions (T7 and T8 at 3 months) and slightly more anteromedial regions (subsequently).
From 6 months of age, iP1 was followed by a negative component (iN1) ( Fig. 2). The average across channels of the iN1 latency decreased by 82 ms (SEM=0.26) between 6 and 12 months (p<0.001) and 54 ms (SEM=0.19) between 12 and 18 months (p<0.001) ( Fig 3). The decrease in iN1 latency between 6 and 12 months was more marked for the girls than for the boys with respectively, in average, a difference of 81 ms (SEM=0.40) and 60 ms (SEM=0.30) (p<0.001) ( Fig. 4). The average across channels of the iN1 amplitude increased with age, by 0.99 µV (SEM=0.15) between 6 and 12 months (p<0.001), 0.40 µV (SEM=0.10) between 12 and 18 months (p<0.001), and 1.05 µV (SE=0.10) between 18 and 24 months (p<0.001) ( Fig. 3). The increase in iN1 amplitude was seen in girls and boys between 6 and 24 months (p<0.001) with respectively, in average, an increase of 5.9 µV (SEM=0.29) and 1.58 µV (SEM=0.22) ( Fig. 4).
3.3 Cortical sourcesFig. 5 additionally shows cortical sources of iP1 component following standard stimulus displayed according to z-scores of signal strength. Sources were mostly located within the left temporal lobe including the primary auditory cortex (Heschl gyrus, Brodmann area (BA) 41), secondary auditory cortex (BA42), the middle temporal gyrus (BA21), superior temporal gyrus (including Wernicke's area, BA22) and anterior prefrontal cortex (BA10 and BA11). Signal strength z-scores were much higher at 3 and 6 months than at term age with gradual focalization centered on left BA41 and BA42. Thereafter z-scores decreased, in relation to higher variance associated with lower numbers of participants and epochs, but localization of the cortical source remained in the temporal lobe at 12 and 24 months (data not shown).
Fig. 6 shows sources of iN1. Cortical sources were mostly located within the left hemisphere, with signal strength increasing from 6 months to 24 months. At 6 months, sources were localized in the primary auditory cortex. At 12 months, they were in the posterior inferior frontal gyrus (including the Broca's area, BA44 and BA45). At 18 months they extended from Broca's area to the temporal pole. At 24 months, sources were distributed in Broca's area, Wernicke's area, the parietal lobe, as well as the prefrontal cortex bilaterally at 24 months.
3.4 Auditory discriminationChannels exhibiting MMR that were significantly different from the baseline are shown for the 6 timepoints on Supplementary Figure 3. At each timepoint, MMR showed an initial negative component (early mismatch negativity, E-MMN) followed by a positive component (P-MMR) ( Fig. 2). From 12 months, these components were consistently followed by a late negative component (late mismatch negativity, L-MMN). The amplitude of E-MMN increased with age. P-MMR amplitude increased initially, reaching maximal values at 6 months, and subsequently decreased and stabilized from 12 months. L-MMN remained stable between 12 and 24 months. Fig. 7 illustrates the topographical distribution across the scalp and amplitude of MMR at term age, which is used as the reference age, and all the other ages at three moments within the significant latency identified with cluster-based statistics, with the exception of the comparison between 3 months and term age as cluster-based statistics did not show a difference between these ages (adjusted p=0.40). For each age comparison (except 3 months-term), the third line in the figurines shows the topography and activity strength of the clusters differentiating the two age groups. From 6 months on, cluster-based statistics demonstrated a statistical difference compared to term age. Between 6 months and term age, MMR was significantly different over the frontotemporal regions in the 270 to 428 ms latency range, i.e. P-MMR range at 6 months (adjusted p=0.02; maximum sum=+8093; cluster size=2273). Between 12 months and term age, the statistical difference was more pronounced in the 309 to 744 ms latency range, i.e. the L-MMN range, in particular over the right parietal regions (adjusted p=0.007; maximum sum=-20213; cluster size=7030). Between 18 months and term age, the difference more pronounced over the centroparietal regions (latency 342–800 ms, L-MMN range; adjusted p=0.007; maximum sum=-33796, cluster size=9897). Between 24 months and term, the difference was more pronounced over the central regions (latency 313–800 ms, L-MMN range; adjusted p=0.007; maximum sum=-30638; cluster size=8611).
3.5 Relation between iP1 and mismatch response at term age, and gestational ageAnalyses were performed on the three channels showing the highest significant t-values, indicating the largest signal deviation from the baseline at those locations compared to the other channels (Suppl. Fig. 1 and 3). For iP1, the analyzed channels were AFz, FC1, and FC6. For MMR, the channels were AFZ, AF3, and AF4. The correlation between iP1 latencies at term age and gestational age showed no significant linear relationship. Similarly, there was no significant correlation between iP1 amplitude at term age and gestational age. In addition we did not find any statistical significant linear relationship between the latency and amplitude of the MMR (E-MMN) and gestational age.
4 DiscussionWe investigated CAEP in response to hearing speech syllables, recorded longitudinally for the first two years of life using high-density EEG in a large cohort of children who were born extremely or very preterm without developmental disorders in order to study early functional brain maturation of auditory processing. Our study is neither strictly longitudinal nor purely cross-sectional, since only 6 participants completed all six sessions and we averaged data at each time point across all available participants rather than tracking individual trajectories. The study can therefore be regarded as semi-longitudinal, with group-level time course analysis. CAEP are generated from auditory thalamocortical and corticocortical pathways, including the primary and association cortex. It must be noted that the components that we refer to as iP1 and iN1 throughout this paper, based on their polarity and order of occurrence, sometimes go by other names in the literature, either in reference to findings in adults, who show additional earlier components, or to latency. For example, Lippé et al. (2009), following much of the adult literature, call the earlier components seen in adults P1 and N1, and therefore they call the first components seen in infants P2 and N2. Earlier components thought to be equivalent to those seen in adults have occasionally been described in infants, but they are seen inconsistently and show very low amplitude, which statistical analysis has not demonstrated to be different from 0 µV ( Lippé et al., 2009). As for latency, it is known to decrease with maturation, making latency-related nomenclature (e.g. P150) difficult to use in developmental studies.
4.1 Maturation of CAEPIn our study, only iP1, the first positive component of CAEP following stimulus, was identified consistently before 6 months of age. From the age of 6 months, iP1 was consistently followed by a negative component (iN1). iP1 amplitude increased markedly between term age and 3 months, and later stabilized around lower values after 6 months, while iN1 amplitude tended to increase. These features are consistent with other reports, although many of these are from cross-sectional studies performed on a single channel. iP1 and iN1 were the most observed and described components in a systematic review of the maturation of CAEP in children ( Silva et al., 2017).
The occurrence of the iP1 component was similar to our findings in a cross-sectional study of 40 children aged 1 month to 5 years (analyzed on the FCz channel only, although 128-channel EEG recordings were performed) ( Lippé et al., 2009). In the latter study, iN1 had very low amplitude with wide dispersion from 1 to 6 months, i.e. in a period when we failed to identify a negative component with consistency. The same study additionally recorded CAEP in 11 adults aged 20–30 years, highlighting a positive (P1) and a negative (N1) component prior to iP1 (called P2 in that context), and the authors concluded that the mature CAEP with all four components only emerge between 5 years and adulthood.
Shafer et al. (2015) examined 19 children two to six times from 3 months to 4 years on 4 channels (left and right of Fz, T7 and T8) from 63-channel recordings, and pooled the data with 30 other single recordings. Consistent with our findings, they found iP1 at all ages, and observed clearly present iN1 from 6 months of age. In older children and 14 adults aged 24–40, they additionally described a later positive component (which they call P2) occurring after iN1.
We found decreasing latencies of iP1 during the first two years, which is also consistent with many other reports. Shafer et al. (2015) showed a decrease in iP1 latency up to 4 years of age. This evolution can likely be ascribed to maturational processes that improve efficiency of signal transmission, such as myelination and synaptic pruning. From the third trimester of gestation until 6 months of age auditory maturation processes involve mainly the brainstem and cortical layer I ( Eggemont and Moore, 2012). The latter contains axons from Cajal-Retzius cells and dendritic branching of pyramidal cells from deeper layers that make synapses with afferent axons from the brainstem. This period is characterized by the beginning of acoustic radiation myelination. Maturational changes occurring from 6 months on mostly involve the development of thalamocortical connections to deep layers of the auditory cortex (IV, V and VI). Cortical layer II and III mature later, until late childhood.
The precise functional significance of CAEP components is yet to be clarified. Čeponienė et al. (2005) found differences in amplitude of CAEP components evoked by speech or nonphonetic sounds in 14 children aged 7–10. iP1 amplitude was higher following the nonphonetic stimulation whereas iN1 amplitude was higher following the syllable sound. The authors suggested that these components might relate to specific aspects of sound processing, with the earlier iP1 component reflecting sound detection and automatic attention mechanisms, and the later occurring iN1 indexing sound content processing. If confirmed, this hypothesis would provide an interesting interpretation of the early occurrence of iP1 in development, followed by the emergence of iN1 from 6 months. By 6 months of age, infants have been shown to display perceptual narrowing for speech sounds in the language to which they had been exposed, indicating sensitivity and specialization for language-specific phonetic contrasts from that age ( Kuhl et al., 1992). There is also evidence that infants between 6 and 12 months can discriminate speech sounds not present in their native language, with sensitivity to these contrasts as early as 6 months of age, suggesting an initial broad sensitivity to phonetic contrasts that becomes more specialized over time with exposure to their native language (Werker and Tees, 1984; Polka and Werker, 1994).
4.2 Topographical distribution and source localizationIn our study, the iP1 component was recorded with high amplitude over wide (frontocentrotemporal) regions at term age. At 3 months of age, iP1 localization became more restricted to the centrotemporal regions, with maximal amplitude around T7/T8, which is the region in which other authors preferentially performed analyses based on less recording channels ( Lavoie et al., 1997; Shafer et al., 2015), although we found higher iP1 peaks more anteromedially from 6 to 24 months.
Considering the EEG volume conduction effect, where brain-generated electrical signals spread through the brain and surrounding tissues, complicating the precise identification of the origin of brain activities, we complemented these analyses with source analysis ( Michel and Brunet 2019). We found gradual focalization of the generators of the iP1 component from the frontotemporal region at term age to the left primary auditory cortex (Heschl's gyrus) by 6 months. These results, though potentially limited by spatial resolution, assumptions in forward modeling, and depth bias, which contribute to uncertainty in the localization of the sources, add to a recent study of sound processing maturation that showed that source generators were more circumscribed within the temporal and prefrontal gyrus, bilaterally, in a group of 9 month-old infants compared to 4 month-old infants ( Polver et al., 2023).
The marked difference between the cortical localization of iP1 between term age and 3 months would require intermediate timepoints to track whether this component is representing the same cortical sources at these two timepoints. Yet, 24 participants in the study were recoded both at term age and three months, making our findings of narrowing down of the topography of iP1 peak amplitude and its cortical generator in the first few months strongly supportive of our first hypothesis that auditory information becomes more localized within the auditory cortex with age in young children. The primary auditory cortex later plays a central role in the detection and processing of auditory stimuli, including sound localization, frequency analysis, and discrimination ( King et al., 2018). Additionally, it contributes to higher-level auditory functions such as speech comprehension and auditory memory ( Hickok, 2009). Our results suggest that the brain response to auditory stimuli undergoes marked developmental changes even within the first few months of life, indicating a refinement in the organization of auditory processing pathways to the auditory cortex as the infant matures ( Friederici, 2005; Friederici, 2006). This might reflect cytoarchitecture and other maturational changes of the acoustic radiation, which extends from the posterior thalamus to the auditory cortex in the temporal lobe ( Eggermont and Moore, 2012; Adibpour et al., 2020). In particular, neurofilaments proliferation within the axons of the acoustic radiation have been documented to develop a few weeks after birth, followed by myelination of those axons from 3 to 4 months of age ( Moore and Guan, 2001).
The second year of life is also a critical period for speech production development, characterized by significant milestones in vocabulary growth, sound imitation, sentence formation, and interactive communication ( Gervain, 2022). In contrast to iP1, we found that iN1 cortical generator studied longitudinally evolves from a focalized situation to a more complex situation which may be linked to several generators at 24 months, localized in the left primary auditory cortex, posterior parietal cortex, premotor cortex and prefrontal cortex (including Broca's area). The localization of iN1 sources thus appears to be consistent with structures that are involved in integrating auditory information with motor functions and understanding of speech. The auditory dorsal stream is a strongly left-dominant pathway extending from the Sylvian parietal-temporal region to areas in the frontal lobe involved in motor planning and execution, including the premotor cortex and Broca's area ( Hickok and Poeppel, 2007). It is involved in the spatial localization of sounds (“where” pathway), speech perception, auditory-motor integration, and the coordination of auditory information with motor actions. The auditory mirror neuron system is another structure that is distributed within the posterior parietal cortex, premotor cortex and Broca's area and entertains important functional links with the auditory dorsal stream. Auditory mirror neurons are involved in the representation and imitation of speech sounds and gestures, facilitating the learning of new words ( Iacoboni, 2009). They are in turn also involved in understanding emotions and contextual factors that are important for developing effective communication skills.
4.3 Auditory discriminationMMR is an event-related potential that can be used to study processing of changes in acoustic stimulus features. MMR reflects the nervous system's ability to detect deviations from an expected auditory stimulus, relying on sensory memory, prediction, automatic auditory processing, and hierarchical neural integration ( Näätänen and al., 2007). The response typically observed in adults is negative, i.e. mismatch negativity. It is commonly interpreted as being indicative of mature sensory memory and inhibitory control. In contrast, positive MMR is more common in infants and young children, likely reflecting ongoing neural maturation, with a gradual transition to negative MMR as the auditory system and inhibitory mechanisms mature.
MMR to phonetic stimulation can be recorded at term age ( Kostilainen et al., 2020; Virtala et al., 2022; Panzani et al., 2023; Navarrete-Arroyo et al., 2024). Using an oddball paradigm, we found that auditory discrimination of speech sounds evolved during the two first years of life in terms of both polarity and topography. MMR became significantly different from term age from the age of 6 months. Three components of the MMR showed variation with age, namely E-MMN, P-MMR, and L-MMN, as previous documented in infants by Virtala et al. (2022). In adults, E-MMN has been hypothesized to be sensory-related, reflecting automatic detection, and L-MMN memory-related, corresponding to higher level cognitive processing ( Tata et al., 2005). P-MMR amplitude was significantly higher at 6 months than at term age, consistent with our second hypothesis. L-MMN was present from 12 months and persisted through subsequent timepoints. This longitudinal developmental trajectory of MMR aligns with a cross-sectional study performed in children of various ages during the second year of their life, showing a shift from positive MMR to adult-like mismatch negativity with the evolution of speech perception ( Cheng and Lee, 2018). Other factors than age influence the presence of the MMR components, as suggested by Virtala et al. (2022), who additionally suggested variation related to features of the deviant stimulus. In our study, the MMR topographical distribution showed gradually stronger signal in the frontoparietal regions bilaterally from 6 to 24 months, similar to late MMR documented in 15 university students ( Tata et al., 2005).
4.4 Influence of sex on cortical processing of auditory informationThe sex differences we observed, with seemingly earlier maturational changes in girls, are notable but difficult to interpret at this stage. While some sex differences have been documented in brain structure and function development in infants, no clear picture emerges of general earlier maturation in girls or boys. In particular, there is limited evidence for sex differences in the maturation of auditory cortex neural responses during infancy ( Y. Chen et al., 2023). When present, such differences often interact with a variety of other factors ( Etchell et al., 2018).
4.5 Influence of gestational age on cortical processing of auditory information at term ageWe explored the influence of gestational age on early auditory processing. This could reflect maturation dynamics. Previous reports of CAEP comparing preterm and term infants failed to show clear results. Melo et al. (2016) found significantly longer iP1 latency in a group of infants born at 26–36 weeks than in a group of infants born at 37–41 weeks, but no significant difference in amplitude in the neonatal period (the precise age at the time of recording was not provided). Similarly, Suppiej et al. (2015) found no significant difference in iP1 amplitude between extremely and very preterm infant recorded around 5 weeks before term age, but there was also no significant difference in iP1 latency at 1 month corrected age between infants born at 31–36 weeks and infants at 37–42 weeks studied by Gottschalck Cavalcanti et al. (2020). However, the same study documented significantly shorter iP1 latency in the preterm group compared to the fullterm group recorded at 3 months corrected age, suggesting accelerated development of the cortical auditory pathways which they ascribed to early exposure to socially relevant environmental stimulation.
Like all physiological systems, the development of the central auditory nervous system is reliant on experience during the early years of life. Behavioral studies have documented reactions to noise in both fetuses and preterm infants from 25–26 weeks gestation ( Graven and Browne, 2008). Studying the effect on CAEP of a 6-week active engagement training with auditory stimulation between 4 and 7 months compared to passive stimulation and a naive age-matched control group, Benasich et al. (2014) found significantly shorter iP1 latencies in the active engagement group compared to the other two groups, but differences in iP1 latency were not specifically studied between the passive stimulation and control groups.
Looking at a much earlier passive stimulation period, we did not find any correlation between the latency and the amplitude of iP1 and MMR (E-MMN), and gestational age. Although this is consistent with previous findings of undetectable influence of early extrauterine life experiences prior to term age on EEG complexity as measured by multiscale entropy ( Pelc et al., 2022), absence of evidence is obviously no evidence of absence. It appears to challenge the notion of a lower boundary for the critical period of auditory development in preterm infants, thought to be around 25 weeks gestation, i.e. when the auditory system is sufficiently developed to respond to sound ( Graven and Browne, 2008). But it must be stressed that preterm infant auditory exposure to language exposure has been documented to be very limited (about 5 times less) compared to typically-developing fetuses ( Monson et al, 2023), suggesting that the population we studied may not be optimal to address this issue.
4.6 Limitations and future perspectivesOur study involved a relatively large sample size and was conducted longitudinally, which are strengths compared with existing literature. However, as with any longitudinal studies, particularly in infants, some appointments could not be kept, which resulted in missing data, and participant attrition occurred over time in spite of the strategies we implemented to minimize dropout. Attrition is a recognized issue in physiological studies in 0–24 month-old infants ( Baek et al., 2023). Yet, we verified that there were no statistical differences in gestational age, birth weight, and sex between infants whose EEGs were analyzed and those whose EEGs were not analyzed due to insufficient data or excessive EEG artifacts, and the number of analyzed CAEP epochs was similar across the various age points.
It must be stressed that the preterm infants we studied represent a selective sample of otherwise healthy babies, calling for caution in generalizing findings to other preterm infants. Because brain maturation is characterized by particularly rapid and intense increase in brain volume, complexity, and connectivity during the third semester of gestation, preterm birth can significantly impact brain development and neurodevelopmental outcomes, as it is associated with brain vulnerability. Preterm infants, particularly those born before 32 weeks gestation, are more susceptible than fullterm infants to brain injury from intraventricular hemorrhage and white matter damage (periventricular leukomalacia), which can disrupt brain development and impair crucial neural pathways. Various impacts on brain maturation, such as altered cortical folding, structural covariance networks, and brain volumes have been associated with poor neurodevelopmental outcomes in spite of possible compensatory mechanisms, such as accelerated cortical folding and increased gray matter volume ( Kim et al., 2020; Young et al., 2020). Nevertheless, many infants born preterm demonstrate resilience despite the risk or adversity (Masten, 2001), and achieve typical developmental milestones. For example, 68% of 6–10 year-old children born under 32 weeks gestation have been found to have a typical development profile, and 82% of the total cohort had no major developmental disorders, i.e. cerebral palsy, blindness, deafness, autism or epilepsy ( Brévaut-Malaty V et al., 2010), similar to the group we studied.
In this study, pitch and duration of the standard and deviant stimulation can be considered as naturalistic as they were based on predictable linguistic patterns of the language to which the participants were exposed. However, differences in both parameters makes fine comparison with studies changing only one stimulus feature difficult. Another limitation concerns the fact that we did not stratify most of our analyses according to arousal state. Several studies have demonstrated the robustness of auditory processing in infants and children with relative stability across different states of arousal ( Hirasawa et al., 2002, Taga et al., 2018). High-density EEG can be considered as another strength of this study. While it offers excellent temporal resolution, it has limitations in spatial resolution compared to other imaging modalities such as fMRI. This, and inherent ambiguity associated with the inverse problem, chiefly its mathematically ill-posed nature lacking a unique solution, can make it challenging to precisely localize neural sources of CAEP components. In future studies, integrating EEG with other neuroimaging techniques could provide a more comprehensive view of auditory processing development.
The proxy we used for preterm-age experience is very summary and preliminary. In the future, differences in early auditory experiences, parental interaction, and socio-economic status as well as other environmental factors that could influence auditory development should be recorded and controlled ( Pelc and Gajewska, 2018). Preterm birth may be associated with medical complications that affect developmental trajectories; however, we excluded infants with any medical complications or atypical neurodevelopment, reducing heterogeneity and enhancing the potential informative value of this cohort. Although similar neural processing of speech sounds at term age has been demonstrated to be similar between preterm and fullterm infants ( Kostilainen et al., 2020), an additional comparison group of infants born at term would have strengthened (or nuanced) our conclusions. We did not record the specifics of the early postnatal environment of the participants but they were all admitted in the NICU, an environment characterized by medical infrastructure and controlled conditions that make it significantly different from a home environment (as documented e.g. by O'Callaghan et al., 2019). Future studies should more finely examine hemispheric specialization. They should also explore the associations between the observed cortical responses and behavioral measures of early language development. Finally, the phonetic stimuli we used do not capture the full complexity of auditory processing capabilities. Future studies could employ a wider range of auditory stimuli to explore different aspects of auditory discrimination and processing. Yet, we believe that this study provides a useful framework for future research on language development in early childhood, across typical and atypical development. For example, investigating CAEP in infants at risk for developmental disorders, such as autism spectrum disorder, dyslexia or other developmental impairments, could help identify early biomarkers for these conditions ( Guttorm et al., 2001; Bosl et al., 2011; Virtala et al., 2022; Piazza et al., 2023), which might in turn lead to early detection and intervention strategies. The iP1 component has been identified as a useful biomarker for assessing central auditory maturation in children with hearing impairments, including those with multiple disabilities ( Sharma et al., 2013), with clinical utility in evaluating the effectiveness of early interventions, such as hearing aids and cochlear implants, in children with sensorineural hearing loss and auditory neuropathy spectrum disorder ( Campbell et al., 2011). This may require routine auditory screening for preterm infants to identify those at risk of delayed auditory development ( Kamenov and Chadha, 2021). Investigating the neuroplasticity of the auditory system in response to early interventions, such as auditory training or speech therapy ( Donadon et al., 2020), could provide valuable insights into optimizing rehabilitation strategies for infants with atypical auditory development. Such plasticity has long been observed across various species and has both theoretical and clinical significance ( Ruben and Rapin, 1980). These findings underscore the importance of early interventions in optimizing rehabilitation strategies for infants with atypical auditory development.
5 ConclusionThis longitudinal study provides new insights into the spatiotemporal dynamics of early auditory processing through a detailed examination of the maturation of high-density CAEP in extremely and very preterm infants without developmental disorders over the first two years of corrected age. Our findings demonstrate significant developmental changes in the auditory processing system, particularly in the evolution of the iP1 and iN1 components, their cortical sources, and the infants' ability to discriminate phonetic differences. The very early occurrence of iP1 suggests very early cortical sound detection, while its latency and amplitude do not appear to be significantly influenced by the extent of prematurity in infants born extremely or very preterm, but by corrected age. iP1 is followed by the emergence of iN1, which possibly indexes cortical sound content processing, from about 6 months with subsequent organization of iN1 source generators over the first two years along the auditory cortical dorsal stream and mirror neuron system, providing a potential functional link to speech processing. Auditory discrimination of speech sounds also significantly changes starting at around 6 months of age, i.e. long before children master receptive or expressive speech.
CRediT authorship contribution statementKarine Pelc: Writing – review & editing, Writing – original draft, Project administration, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Aleksandra Gajewska: Writing – review & editing, Writing – original draft, Project administration, Formal analysis, Data curation. Natan Napiórkowski: Methodology, Investigation, Formal analysis, Data curation. Jonathan Dan: Writing – review & editing, Writing – original draft, Methodology. Caroline Verhoeven: Writing – review & editing, Writing – original draft, Validation, Methodology, Formal analysis. Bernard Dan: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Conceptualization.
Declaration of competing interestThe authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
The author is an Editorial Board Member/Editor-in-Chief/Associate Editor/Guest Editor for [Journal name] and was not involved in the editorial review or the decision to publish this article.
AcknowledgmentsWe are grateful to the children who participated in the study and their family, as well as the neonatal unit teams. We thank Karin Duponcelle and Irit Daniel for technical assistance, Stewart G. Boyd for providing the synthesized syllable sounds used for stimulation, Laura Stubbe, Guy Cheron, Ana Maria Cebolla and Mathieu Petieau for discussion, and Talia Dan for help with figures. This work was supported by the Fondation Roger de Spoelberch, the Fondation JED and the Fonds Iris Recherche. The Fondation Roger de Spoelberch supports the Primebrain project in which infants are longitudinally assessed; the assessments include high-density EEG and ERP. The support is financial and the authors retain full scientific independence. The Fondation JED and the Fonds Iris Recherche each contributed towards the cost of the recording and analysis equipment for high-density EEG. The support is financial and the authors retain full scientific independence. The authors have stated that they had no interests that might be perceived as posing a conflict or bias.
Supplementary materialsSupplementary material associated with this article can be found, in the online version, at doi:10.1016/j.neuroimage.2025.121115.
Appendix Supplementary materialsImage, application 1
| Standard stimulation | Deviant stimulation | |
| Term age n=42 | 402 (105)/187–521 | 73 (15)/43–92 |
| 3 months n=43 | 393 (134)/174–578 | 72 (27)/36–103 |
| 6 months n=40 | 308 (115)/106–553 | 59 (23)/16–94 |
| 12 months n=33 | 325 (138)/126–547 | 60 (19)/26–95 |
| 18 months n=27 | 351 (130)/213–517 | 68 (35)/36–100 |
| 24 months n=25 | 373 (97) /102–527 | 73 (16) /18–98 |
Copyright Elsevier Limited Apr 15, 2025