Reward-related language processing of maternal speech in infants at different likelihood of ASD

Abstract

This longitudinal study investigated the differential impacts of maternal speech on early socio-communicative development in infants at low likelihood (LL) and elevated likelihood (EL) of autism spectrum disorder (ASD). Using functional near-infrared spectroscopy, we measured cortical responses and connectivity in 6-month-old infants while they listened to their mother’s voice and an unfamiliar female voice. LL infants exhibited extensive cortical activation and robust connectivity in temporal and frontal regions, particularly in areas associated with voice processing, reward, and language functions. In contrast, EL infants showed minimal activation and weaker connectivity in these regions. Specifically, LL infants demonstrated significant connectivity between the superior temporal gyrus and the inferior frontal gyrus on the left side and between the orbitofrontal cortex and language areas, facilitating language processing and reward-related responses to maternal speech. These neural patterns were absent in EL infants, highlighting a neural basis for subsequent language delays. Furthermore, many of these reward-related or language-related networks predicted subsequent language development. Our findings underscore the importance of neural sensitivity to familiar human voices, regarding them as rewards that will eventually facilitate the acquisition of speech.

Introduction

The development of infants’ social and language skills is facilitated through interactions with their caregivers. Speaking to infants is often considered one of the most important signals for promoting successful language development (Kuhl 2004). Even in the prenatal stage, foetuses can distinguish auditory stimuli and demonstrate differential responses to their mother’s voice compared to other voices or sounds (Voegtline et al. 2013, Webb et al. 2015). It has also been suggested that mothers’ voices may facilitate left frontotemporal connectivity as early as birth (Uchida-Ota et al. 2019). Neonates show a preference for their mothers’ voices over those of other females (DeCasper and Fifer 1980). This body of evidence suggests that mothers’ voices undergo distinct processing, separate from environmental sounds or unfamiliar voices, from the gestational phase. Moreover, mothers’ voices contribute to the adjustment of phonemes (Cristià 2011) and prosodic perception (Abboub et al. 2016), as well as facilitating word learning (Barker and Newman 2004, Imafuku et al. 2014) and emotion recognition (Montague and Walker–Andrews 2002). Given the significant role of mothers’ voices, it is postulated that such voices enhance the development of infants’ sociality and language acquisition.

Autism spectrum disorder (ASD) is a type of neurodiversity characterized by repetitive behaviours and focused interests, often accompanied by delays in social and language development. Although these developmental delays have been extensively examined in the literature with a focus on anomalous perception and cognition of social signals in individuals with ASD, one of the major causes seems to derive from a mother–infant communication system based on their attachment (McKenzie and Dallos 2017). Particularly during the first year of life, parental attachment contributes substantially to social and language development via facial and vocal signals (Feldman 2015), fostering essential learning processes. Considering the crucial role of the primary caretaker’s voice, as reviewed above, individuals with ASD may have problems processing familiar voices in the course of social and language development. In other words, indifference to social-emotional voices may impede their development. Although there are some studies on responses to human voices in individuals with ASD, there are no studies during the infancy period, in terms of ASD factors, on responses to mothers’ voices. Consequently, the present study focuses on the processing of the familiar mother’s voice and an unfamiliar voice in early development. Due to the challenges of diagnosing ASD in infancy (Ozonoff et al. 2010, Jones et al. 2014), we targeted sibling infants of children with ASD to examine the developmental aspects of ASD. We refer to such participants as infants at elevated likelihood (EL) of ASD, as opposed to infants at low likelihood (LL) for those without ASD siblings.

With regard to processing human versus non-human voices, a recent study compared behavioural responses to human voices, monkey calls, and artificial sounds between EL and LL groups of 9-month-old infants (Sorcinelli et al. 2019). Both groups showed a preference for human voices over artificial sounds but differed in their preference for human versus monkey voices. Intriguingly, in contrast to no subsequent developmental impact of preferences in LL infants, the EL infants who preferred human voices over other non-human sounds demonstrated higher word comprehension and lower autism-like behaviours at 12 months. Among the very limited studies comparing behavioural responses to their mother’s voice versus a stranger’s voice, children with ASD aged 4–6 years were reported to show less preference for their mothers’ voices compared to typically developing children (Klin 1991). Children with ASD aged 7–9 years performed poorer in recognition and matching tasks of familiar faces and voices compared to their typical controls (Boucher et al. 1998, Boucher et al. 2000). The lack of correlation between face and voice recognition performance suggested distinct processing mechanisms for each modality. In adolescents, a negative correlation between Autism Quotient and voice recognition performance has been shown (Skuk et al. 2019). These findings suggest that EL infants may process auditory social signals differently, particularly showing reduced preferences for familiar mothers’ voices.

Apart from the behavioural measures, functional brain responses to human voices have also been examined in relation to ASD. However, before focusing on the differences in LL and EL infants, we first review the general function and development of the brain associated with the processing of human and familiar voices. The temporal voice area or temporal voice patch (TVP), primarily located in the superior temporal region, is known to be crucial for voice recognition and identification (Belin et al. 2000). The anterior part of the STS is particularly involved in the recognition of familiar voices (Stevenage 2018). While 4-month-old infants show strong bilateral responses to both human voices and non-human monkey calls (Minagawa-Kawai et al. 2011), by around 6 months of age, LL infants seem to show specific activation to human voices in the temporal regions involving TVP (Grossmann et al. 2010, Blasi et al. 2011, Lloyd-Fox et al. 2012). Studies on infant recognition of mothers’ voices have been studied from a socio-communicative perspective, exclusively in LL infants. The mothers’ voices are specifically learned from the foetal period, and significant frontotemporal connectivity in response to mothers’ voices is already evident at birth (Uchida-Ota et al. 2019). In addition to the temporal area, prefrontal regions also play a role in processing the socio-emotional aspects of the mother’s voice. Neuroimaging of 2-month-old infants with fMRI has exhibited increased activation in the medial prefrontal cortex (mPFC), the orbitofrontal cortex (OFC), amygdala, and left temporal lobe when hearing their mothers’ voices (Dehaene-Lambertz et al. 2010). Activation of the dmPFC, which deals with self-referencing and mentalizing (Gusnard et al. 2001, Bzdok et al. 2013), was observed exclusively for the mother’s voice condition in 6-month-olds (Imafuku et al. 2014). While studies with electrophysiological measures have, in general, shown distinctive event-related potentials to mothers’ voices compared to unfamiliar voices in neonates and infants (deRegnier et al. 2000, Siddappa et al. 2004, Therien et al. 2004, Purhonen et al. 2005, Beauchemin et al. 2011), the specific role of the mother’s voice has been demonstrated. Namely, even in preverbal infants of 9 months old, the N400 component, indicative of semantic understanding, was identified solely via words spoken by their mother (Parise and Csibra 2012). Given these studies, the cognitive significance of mothers’ voices in early childhood extends beyond the mere recognition of voices. It also facilitates language as a form of communication and the understanding of others’ intentions, contributing to the development of language and social cognition.

Among limited studies comparing LL and EL infants, strong activation to voices was observed in the right posterior superior temporal sulcus (STS) only in the LL infants aged 4 to 6 months, in contrast to comparable bilateral STS activation to non-voice stimuli for both LL and EL groups (Lloyd-Fox et al. 2013). Mostly consistent with this, strong activations in the right temporal area, including TVPs and medial prefrontal regions, were specific to voices only in LL infants aged 4–7 months (Blasi et al. 2015). By the age of 2–3 years, EL children begin to show specific activations to voices in different brain regions unlike LL children (Redcay and Courchesne 2008). Differences in functional connectivity have also been found between the two groups. While 3-month-old EL infants showed stronger functional connectivity in the left hemisphere in response to voices, such connectivity gets stronger for the LL group by 12 months, with no differences at 6 and 9 months old (Keehn et al. 2013).

Although almost no infant study has examined brain activation and connectivity during the processing of the mother’s voice by comparing LL and EL infants, studies on children with ASD have provided clearer views. In processing the mother’s voice, children with ASD at 10 years old exhibited a remarkable pattern of underconnectivity in relation to emotion-related association regions, with weaker connectivity between the bilateral STS and reward-related regions (Abrams et al. 2013). Considering the role of the temporal lobe for voice processing (Grossmann et al. 2010; Beauchemin et al. 2011) and OFC as a reward circuit (Haber and Knutson 2010), the strength of connectivity of these areas, especially when hearing mothers’ voices, reflects the preference or reward-worthiness of the mother’s voice. The strength of these connectivities correlated with communication skills scores in 10-year-olds (Abrams et al. 2013), suggesting a crucial role of this reward-related connectivity in development. However, it remained unknown when such a divergence of reward-related connectivity in the context of voice processing emerged.

Consequently, in the current study, we explore differences between LL and EL infants in their early brain substrates for processing familiar mothers’ voices by focusing on connectivity between the temporal area, including TVPs and the OFC, reward-related area. Given the critical role of mothers’ voices in social and language development during infancy as well as weak brain responses to voices in EL infants, it is postulated that mothers’ voice processing in EL infants may have already diverged from that in LL infants after half a year from birth. More specifically, we hypothesize that the mothers’ speeches strengthen functional connectivity in the frontotemporal region in LL infants, whereas, the EL infants may exhibit different cortical activations and weaker connectivity (Abrams et al. 2016; Uchida-Ota et al. 2019). Such an altered neural network reflecting indifference to the mother’s voice is expected to affect subsequent language development. To examine these hypotheses, the present study compared cortical activation and connectivity during listening to mother’s and unfamiliar female voices between LL and EL infants at 6 months, using functional near-infrared spectroscopy (fNIRS). Furthermore, the relationship between brain data and the subsequent development of social skills and language was investigated.

Materials and Methods

Participants

Twenty-five infants aged 6–7 months (14 female; mean age, 206.8 ± 16.0 [SD] days; range, 183–240 days) with an elevated likelihood (EL) for autism spectrum disorder (ASD) (having an older sibling with a clinical diagnosis of ASD) and 26 infants (15 female; mean age, 193.3 ± 12.0 [SD] days; range, 183–242 days) with a low likelihood (LL) (no family history of ASD) were included in the final dataset. An additional 15 infants were excluded due to excessive motion artefacts caused by fussiness or rejection of probe attachment. The experimental protocol was approved by the ethical board of Keio University, Faculty of Letters (no. 150320103). Written informed consent was obtained from all parents.

Procedures

The stimuli consisted of two types of infant-directed speech, varying in familiarity levels. For the familiar condition, referred to as the mother condition, speech recordings from the infants’ own mothers were used. For the unfamiliar condition, speech from mothers of other infants was utilized. These vocal stimuli were pre-recorded at a sampling rate of 44 kHz, 16-bit. The stimuli comprised 18 short Japanese sentences characterized by rich prosody in infant-directed speech. To maintain uniform acoustic properties across conditions, the stimuli from the mother condition were also reused as the unfamiliar condition for another infant. Specifically, the voice of Infant A’s mother served as the unfamiliar voice for Infant B, and the voice of Infant B’s mother served as the unfamiliar voice for Infant C, following procedures detailed in Uchida-Ota et al. (2019). The resultant stimuli used for the final dataset did not differ significantly in most acoustic measures between conditions, as detailed in the Supplementary Information (Supplementary Table S1A, see online supplementary material for a colour version of this table). In addition, subjective ratings of the stimuli (prosodic salience and emotional expressiveness; five-point Likert scales) showed no main effects of Group, Condition, or their interaction (Supplementary Table S1B, see online supplementary material for a colour version of this table). Stimulus sounds were presented at approximately 70 dBSPL, using a block design consisting of 10 seconds of silence followed by 15 seconds of stimulus presentation, during which 4 to 5 sentences were played. Infants listened to the stimuli while seated on their caregiver’s lap inside a sound-attenuated booth. To minimize motion artefacts, an experimenter engaged the infants with silent toys during the recording. Trials of the two stimulus conditions were presented pseudo-randomly until a minimum of five artefact-free trials were acquired for each condition.

An fNIRS system (ETG-7000, Hitachi Medical Co., Japan) was employed to measure infants’ frontotemporal brain regions (see Supplementary Fig. S1, see online supplementary material for a colour version of this figure). We used the modified virtual registration method (Tsuzuki et al. 2007) to estimate brain regions underlying each channel. Further details of the fNIRS measurement, including virtual registration, are provided in the Supplementary Information (fNIRS measurements).

To explore the relationship between brain responses and subsequent development, developmental assessments using the Kyoto Scale of Psychological Development were conducted at 6 and 9 months of age. Additionally, parents completed the MacArthur-Bates Communicative Development Inventories (j-CDI) (Ogura et al. 1993) when infants were 9 months old. Details of each test and questionnaire are provided in the Supplementary Information (Developmental Tests).

Analysis

fNIRS data underwent preprocessing, including artefact removal, filtering, and conversion to changes in oxygenated (oxy-) and deoxygenated (deoxy-) haemoglobin concentrations (mM·mm) using the modified Beer-Lambert law. This study focused on oxy-Hb signals to facilitate comparison with fMRI studies (Strangman et al. 2003). Cortical activation was analysed using filtered, baseline-corrected oxy-Hb signals. Significant activation periods were identified, and an analysis window was set to 5 seconds post-stimulus onset. Activations were evaluated with maximum t-statistic permutation (10,000 iterations) that controls the family-wise error rate across channels and remains robust to the correlated, non-Gaussian nature of infant fNIRS data obtained from modest samples. Functional connectivity was assessed by calculating phase locking values (PLVs) between seed regions of interest (ROIs) [STGs: CH10, 35 and OFCs: CH9, 27, see Supplementary Fig. S1 (see online supplementary material for a colour version of this figure) for channel positions and numbering] and other channels. Synchronization indices were statistically evaluated. PLVs were evaluated with Wilcoxon rank-sum tests, and unadjusted p values are reported as p_Wilc. PLV captures sub-second phase synchronization and has been validated for task-based fNIRS in both infants and adults (Uchida-Ota et al. 2019, Cheng et al. 2024). The effects of cortical activation and PLVs on developmental indices were analysed using robust regression models. Detailed methods are available in the Supplementary Information (Analysis).

Results

In response to voice stimuli, both LL and EL groups exhibited distinct time courses of hemodynamic changes (Supplementary Fig. S2, see online supplementary material for a colour version of this figure). The LL group showed a gradual increase in oxy-Hb, whereas the EL group displayed either a gradual decrease or rapid fluctuations across several channels. Bootstrap permutation tests indicated that speech elicited significant cortical activation in multiple regions for LL infants (Fig. 1, see Fig. 6D for channel number). Specifically, in the mother condition, LL infants showed significantly increased activation in broad cortical regions, including bilateral superior temporal gyri (STGs) (CH10: P _FWER = .003, Cohen’s d = .80 [hereafter d]; CH35: P  _FWER = .008, d = .74; P  _FWER from a maximum t-statistic permutation test; see Supplementary Tables S2 and S3 for detailed statistics, see online supplementary material for a colour version of these tables), and supramarginal gyri (SMGs) (CH19: P  _FWER = .035, d = .54; CH44: P  _FWER = .036, d = .54) (Fig. 1a). In contrast, during the unfamiliar condition, significant activation was observed in more localized areas, with sustained activation in the right STG (CH29: P  _FWER = .045, d = .51; CH30: P  _FWER = .040, d = .55) (Fig. 1b). Conversely, the EL group showed minimal changes in cortical activation under either condition, with none resembling the activation patterns observed in the LL group (Fig. 1c and d). Comparing groups under each condition, LL infants exhibited significantly stronger cortical activation in the mother condition, particularly in widespread regions encompassing temporal areas (e.g. bilateral STGs CH10: P  _FWER = .007, d = .73; CH35: P  _FWER = .040, d = .54; SMG CH19: P  _FWER = .048, d = .51) (Fig. 1e and Supplementary Table S3, see online supplementary material for a colour version of this table). Under the unfamiliar condition, group differences were less pronounced, although LL infants showed stronger activation in the left OFC (CH4: P  _FWER = .028, d = .60) and weaker activation in the bilateral dorsolateral prefrontal cortices (dlPFCs CH21: p_FWER = .023, d = .60; CH37: P  _FWER = .014, d = .66) (Fig. 1f and Supplementary Table S3, see online supplementary material for a colour version of this table). For supplementary purposes, we compared the conditional differences (mother-unfamiliar) between groups; several channels, including bilateral STGs (CH10, 35), showed significant LL > EL differences (Supplementary Note S1 for details).

Figure 1.

Cortical activation in response to speech stimuli. (a) Channels showing significant oxy-Hb changes compared to the baseline for LL infants in the mother condition, viewed from the left and right sides (FWER-corrected). Red indicates positive changes; blue indicates negative changes. Green circles indicate channels with statistically significant changes (P  _FWER < .05; p_FWER from a maximum t-statistic permutation test). (b) Channels for LL infants in the unfamiliar condition. (c) Channels for EL infants in the mother condition. (d) Channels for EL infants in the unfamiliar condition. (e, f) Channels showing significant differences between groups under each condition. Red indicates greater activation in the LL group compared to the EL group, while blue indicates the opposite. See Supplementary Figs S2 and S3 (see online supplementary material for a colour version of these figures) for detailed statistics.

Figure 6.

Significant effects of cortical activation of ROIs in response to mothers’ speech on language and social developmental indices (a, b, c). See Table 1 for all statistically significant effects. (a) Association between activation of the left OFC (CH9) and language and social scores (KSPD) at 9 months. Green indicates LL group, and red indicates EL group. Scatter plot and fitted lines are shown. Asterisks indicate statistically significant effects (P < .05), n.s. indicates non-significant effects. (b) Association between activation of the right STG (CH35) and number of phrases comprehended (j-CDI) at 9 months. (c) Association between activation of the left STG (CH10) and language and social scores (KSPD) at 6 months. (d) Channel positions and numbering from left and right side views. Green circles indicate ROIs with significant effects on developmental indices (P < .05).

Functional connectivity analysis using phase locking values (PLVs) was conducted with regions of interest (ROIs) as seeds. Specifically, we focused on STGs (CH10 and 35) involved in voice processing and OFCs (CH9 and 27) implicated in reward processing. Initially, Shannon’s Entropy index confirmed significant phase locking compared to phase-randomized PLVs. Analysis of connectivity seeded by the left STG (CH10) (Fig. 2 and Supplementary Tables S4 and S5, see online supplementary material for a colour version of these tables) showed subtle group differences in the number and regions of significant connectivity under both conditions. Notably, significant connectivity with the left inferior frontal gyrus (IFG) (CH7: P  _Wilc = .047, effect size r = .29) was observed exclusively in LL infants during the mother condition. Conversely, connectivity seeded from the right STG (CH35) (Fig. 3 and Supplementary Tables S4 and S5, see online supplementary material for a colour version of these tables) revealed more pronounced group differences. LL infants exhibited greater connectivity compared to EL infants under both conditions, particularly demonstrating robust connectivity with various frontal areas including bilateral dlPFCs and left OFC (CH9: P  _Wilc = .039, r = .31) during the mother condition (Fig. 3a). Notably, LL infants did not exhibit connectivity between right STG and left OFC during the unfamiliar condition.

Figure 2.

Functional connectivity seeded by the left STG (CH10). (a) Connectivities in response to hearing mothers’ speeches for LL infants, viewed from the left, top, and right sides. Dots on the brain surface indicate fNIRS channels, and red lines between dots represent significantly strong connectivity between these channels (P  _Wilc < .05; p_Wilc from an unadjusted Wilcoxon rank-sum test), while pink lines indicate connectivity approaching significance (P  _Wilc <.1). Green dot indicates the seed channel. (b) Connectivities for LL infants in the unfamiliar condition. (c) Connectivities for EL infants in the mother condition. (d) Connectivities for EL infants in the unfamiliar condition. (e, f) Connectivities with significant differences between groups under the conditions. Red lines indicate significantly greater connectivity in the LL group compared to the EL group (P  _Wilc < .05), blue lines indicate the opposite. Pink and light blue lines indicate connectivities approaching significance (P  _Wilc < .1).

Figure 3.

Functional connectivity seeded by the right STG (CH35). See the caption of Figure 2 for further details of (a-f).

Connectivity analysis with OFCs as seeds highlighted consistent between-group trends for the right OFC (Fig. 4) and significant group and condition differences for the left OFC (Fig. 5). Specifically, LL infants exhibited strong connectivity between the left OFC and right temporoparietal channels unique to the mother condition. Similar specificity was observed with both OFCs connecting to the left IFG and STG. Significant connectivities included left OFC (CH9) to left IFG (CH7: P  _Wilc = .040, r = .30) and left dlPFC (CH17: P  _Wilc = .004, r = .30), which were absent in other conditions and groups (Figs 4a and 5a and Supplementary Table S4, see online supplementary material for a colour version of this table). In inter-group comparisons, connectivity between left OFC (CH9) and left IFG (CH7) was significantly stronger in LL compared to EL infants (P  _Wilc = .042, r = .30) (Fig. 4e and Supplementary Table S5, see online supplementary material for a colour version of this table).

Figure 4.

Functional connectivity seeded by the left OFC (CH9). See the caption of Figure 2 for further details of (a-f).

Figure 5.

Functional connectivity seeded by the right OFC (CH27). See the caption of Figure 2 for further details of (a-f).

Furthermore, the impact of activation and connectivity strength in response to mother’s speech on developmental indices at 6 and 9 months was analysed using generalized linear models. Activation in ROIs more strongly predicted developmental outcomes in the LL group (Table 1 and Fig. 6). For example, increased activation in right STG (CH35) was significantly associated with higher scores for phrase comprehension (Fig. 6b, β = 13.91, Z = 2.31, P = .021) at 9 months in LL infants but was associated with lower vocabulary comprehension scores (β = −6.08, Z = 3.22, P = .001) in EL infants (Table 1). Additionally, greater activation in left OFC (CH9) was significantly associated with higher language and social scores at 9 months in LL infants (Fig. 6a, β = 13.75, Z = 2.66, P = .014). Regarding connectivity (Table 2 and Fig. 7), stronger connectivity between left STG (CH10) and left IFG (CH7) was associated with higher vocabulary production (β = 8.05, Z = 3.02, P = .003) and phrase comprehension (Fig. 7c, β = 2.59, Z = 2.91, P = .004) scores at 9 months in LL infants. Stronger connectivity between right STG (CH35) and left OFC (CH9) was associated with higher vocabulary production scores at 9 months in EL infants (Fig. 7d, β = 7.26, Z = 3.29, P = .001). Additionally, stronger connectivity between right OFC (CH27) and left IFG (CH7) was associated with higher phrase comprehension scores at 9 months in LL infants (β = 2.60, Z = 2.12, P = .034).

Table 1.

Significant effects of cortical activation of ROIs on language and social development when hearing mothers’ speech.

Group	Region (ROI)	Channel number	Developmental index	Age at developmental index (months)	β	Z	P
LL	Left OFC	9	Language and social (KSPD)	9	13.75	2.66	.014
LL	Left STG	10	Language and social (KSPD)	6	15.06	2.66	.001
LL	Right STG	35	Phrase comprehension (j-CDI)	9	13.91	2.31	.021
LL	Right STG	35	Actions and gestures (j-CDI)	9	7.06	1.74	.082
EL	Right STG	35	Vocabulary comprehension (j-CDI)	9	−6.08	−3.22	.001

Table 2.

Significant effects of functional connectivity seeded by ROIs on language and social development when hearing mothers’ speech.

Group	Region (ROI)	Channel number	Associated region	Associated region channel number	Developmental index	Age at developmental index (months)	β	Z	P
LL	Left OFC	9	Left dlPFC	17	Vocabulary production	9	13.37	2.12	.034
EL	Left OFC	9	Right STG	35	Vocabulary production	9	7.26	3.29	.001
LL	Left OFC	9	Right PMC	43	Phrase comprehension	9	2.96	2.08	.037
LL	Left OFC	9	Right PMC	43	Vocabulary production	9	9.48	2.09	.036
LL	Left STG	10	Left IFG	7	Phrase comprehension	9	2.59	2.91	.004
LL	Left STG	10	Left IFG	7	Vocabulary production	9	8.05	3.02	.003
LL	Right OFC	27	Left IFG	7	Phrase comprehension	9	2.60	2.12	.034
EL	Right OFC	27	Left dlPFC	17	Vocabulary production	9	−13.08	−2.27	.023
LL	Right OFC	27	Right dlPFC	38	Phrase comprehension	9	1.99	2.36	.018
EL	Right OFC	27	Right dlPFC	38	Vocabulary comprehension	9	−2.56	−2.58	.010
LL	Right OFC	27	Right dlPFC	38	Vocabulary production	9	9.35	6.47	.000
LL	Right OFC	27	Right PMC	43	Actions and gestures	9	-2.04	−2.21	.027
EL	Right OFC	27	Right PMC	43	Phrase comprehension	9	−2.02	−2.17	.030
EL	Right OFC	27	Right PMC	43	Vocabulary comprehension	9	−4.67	−3.24	.001
EL	Right STG	35	Right IFG	33	Vocabulary production	9	6.20	1.90	.058

Figure 7.

Significant effects of functional connectivity seeded by ROIs when hearing mothers’ speech on language and social developmental indices. All significant effects were observed only on developmental indices measured at 9 months and not at 6 months. For each seed ROI, the developmental index with the smallest p value is shown as representative (a–d). See Table 2 for all statistically significant effects. (a) Association between PLV (left OFC (CH9) and left dlPFC (CH17)) and number of vocabularies produced (j-CDI) at 9 months. Green indicates the LL group, and red indicates the EL group. The scatter plot and fitted lines are shown. Asterisks indicate statistically significant effects (P < .05), n.s. indicates non-significant effects. (b) Association between PLV (right OFC (CH27) and right premotor cortex (CH43)) and number of vocabularies comprehended (j-CDI) at 9 months. (c) Association between PLV (left STG (CH10) and left IFG (CH7)) and number of phrases comprehended (j-CDI) at 9 months. (d) Association between PLV (left OFC (CH9) and right STG (CH35)) and number of vocabularies produced (j-CDI) at 9 months. (e) Channel positions and functional connectivities which had significant effects on developmental indices viewed from the top. Green circles indicate ROIs and purple circles indicate associated channels. The purple dotted lines indicate functional connectivity, which had statistically significant effects on developmental indices (P < .05).

Discussion

In this study, we aim to uncover differences in the impact of maternal speech on early socio-communicative development between two infant groups with differing likelihoods of ASD. We measured cortical functional responses and connectivity using fNIRS in LL and EL infants aged 6 months while they listened to speech from their mother and an unfamiliar female. Our findings revealed distinct cerebral responses between the LL and EL groups. Activation in bilateral temporal areas, frontal areas, and functional connectivity among the bilateral STGs, OFCs, and the left IFG—regions involved in processing voice, reward, and language (Belin et al. 2000, Minagawa-Kawai et al. 2009, Price 2012)—differed between the groups. Upon hearing maternal speech, the LL group exhibited widespread cortical activation in temporal and parietal areas, whereas the EL group showed minimal activation. For unfamiliar voices, the LL group showed activation in the right STG, but over a more restricted region than for maternal speech, whereas the EL group showed minimal activation. Additionally, functional connectivity involving the language and reward networks was observed only in the LL group when exposed to maternal speech. These connectivities included the STG–IFG on the left, right OFC–left IFG, and left OFC connected either to the left IFG or right STG. These connectivities were not observed in the LL group when exposed to unfamiliar voices, nor in the EL group for either type of speech. Taken together, this study is the first to reveal that 6-month-old EL infants process maternal speech differently compared to LL infants, particularly in voice-processing regions in relation to socio-emotion, reward, and language functions. Below, we discuss such functions in different sections by referring to activation, connectivity, and brain association with behaviour together.

Socio-emotional voice processing and the right frontotemporal network

The right temporal area, particularly the STG, is involved in processing vocal prosody, including voice identification (Belin et al. 2000, Stevenage 2018) and emotional prosody (Zatorre et al. 1992, Sander et al. 2005). This function is evident even in early development stages (Homae et al., 2006, Arimitsu et al. 2011, Mahmoudzadeh et al. 2013). One notable difference between groups in our study was observed in the right STG and the adjacent temporoparietal area. For LL infants, maternal speech elicited broad cortical activation in the right temporoparietal region, including the STG, SMG, dlPFC, and inferior frontal gyrus (IFG). This was not significant for EL infants.

The right temporoparietal area includes regions involved in socio-emotional processing. The right SMG, part of the temporoparietal junction, is crucial for mentalizing and theory-of-mind (Adolphs et al. 2002, Saxe and Powell 2006), while the right dlPFC is involved in maintaining attention and affective memory (White et al. 2023). The OFC is associated with reward processing in the context of mother–infant attachment (Minagawa-Kawai et al. 2009, Grossmann 2013, Minagawa et al. 2023, Shi et al. 2023). Therefore, it can be assumed that LL infants process maternal speech via the right STG for speaker and emotion identification, extending to higher-order processing, such as affection, mentalizing, and reward. Our PLV analysis supports this, showing significant connections from the right STG to bilateral dlPFCs, right IFG, and left OFC in LL infants during the mother condition.

Considering the roles of the dlPFC in attention and affective memory (White et al. 2023) and the right IFG in processing positive emotional voices (Johnstone et al. 2006, Grossmann et al. 2010, Hampshire et al. 2010, Zhao et al. 2021), we propose a similar process for LL infants. After the right STG analyses the identity and emotional properties of the voice, various networks are recruited to capture the positive emotional content (dlPFC and IFG) and the reward value of familiar stimuli (OFC). This connectivity between the right STG and the prefrontal cortex aligns with findings in LL newborns, interpreted as part of the reward network (Uchida-Ota et al. 2019).

In contrast, EL infants showed no significant activation or strong connectivity from the right STG in response to maternal or unfamiliar speech. Children with ASD often exhibit a lower preference for familiar and maternal voices (Klin 1991, Baucher et al. 1998), suggesting that EL infants may not effectively use maternal voices as salient, affective cues from 6 months of age. For LL infants, the response amplitude to maternal voice was associated positively with language development at 9 months. EL infants showed a positive association between the connectivity strength of the right STG–left OFC and vocabulary production at 9 months. This right STG connectivity is crucial later in development, as its strength is associated with social communication skills in 10-year-olds (Abrams et al. 2016, Abrams et al. 2019) and is weaker in children with ASD (Abrams et al. 2013). Despite these findings by Abrams and colleagues, the developmental origins of these connectivity differences remained unclear. Our study is the first to uncover that even preverbal EL infants at 6 months have a different brain network in response to a familiar voice, lacking the TVP–OFC connection, which could serve as a potential biomarker for ASD diagnosis.

Language-related voice processing and the left frontotemporal area

Broad activation was observed in the left temporal and part of the parietal areas, particularly under the mother condition for LL infants. Notably, this included strong responses in the posterior language areas, such as the left STG and SMG. These areas are implicated in phonological processing, sound segmentation, rule learning, and auditory working memory (Price 2012, Minagawa et al. 2017), suggesting higher-order phonological processing beyond voice familiarity. PLV analysis further revealed significant connectivity between the left IFG and left STG, classic language areas that were specific to maternal speech in LL infants. These findings reflect a facilitated language-related network in response to maternal speech, specifically in LL infants, with predictive implications for later language development. For instance, the left STG response was positively associated with language and social abilities at 6 months, and the strength of the left STG–IFG connectivity predicted language comprehension and production at 9 months in LL infants. This maternal-specific language facilitation aligns with previous neuroimaging studies involving neonates (Uchida-Ota et al. 2019) and 2-month-olds (Dehaene-Lambertz et al. 2010), as well as behavioural studies (Montague and Walker–Andrews 2002, Cristià 2011). Such facilitation by a familiar caregiver appears absent in EL infants, potentially contributing to their language difficulties, including phonological processing issues (Redcay and Courchesne 2008, Eyler et al. 2012). However, despite weaker activations and the lack of IFG–STG connections, EL infants showed a comparable number and type of connectivity projections from the left STG for both conditions, suggesting less pronounced differences compared to the right hemisphere.

Reward-related brain areas connecting to the language area

While the socio-emotional and language aspects of voice processing have been discussed separately thus far, PLV analysis with the OFC as a seed in both hemispheres sheds new light on their interconnectedness. Specifically, significant connectivity was observed between bilateral OFCs and the left IFG, corresponding to Broca’s area, and the left dlPFC exclusively when LL infants listened to maternal speech. Considering the role of the left dlPFC in executive function control, particularly in language comprehension (MacDonald 2000, Klaus and Schutter 2018), maternal speech, which likely induces feelings of reward or motivation in the OFC, may engage functional coupling with the left dlPFC and IFG, thereby facilitating its efficient processing. Additionally, significant connectivity between the left IFG and left STG observed in this study may further enhance this reward-related language network comprising OFC–IFG–STG. Based on this network, it is plausible that the language network (i.e. IFG–STG) observed during maternal speech may be facilitated by a motivational drive rooted in the OFC. Notably, these connectivities were evident exclusively for LL infants. This suggests that EL infants, lacking such reward-related language networks, may experience developmental delays. Regression analysis with subsequent development supports this hypothesis, as many OFC connections exhibited a positive association with language development at 9 months (Table 2). Children with ASD are often indifferent to social stimuli, characterized by reduced eye contact and facial attention. Human voice, a critical social signal, may reflect this indifference, particularly towards familiar voices, evident in the absence of OFC connectivity in EL infants. This divergence likely results in different developmental trajectories for language acquisition. Thus, reward-related language networks have the potential to serve as early biomarkers for predicting language delays.

Limitations

The present study has a limitation in that our follow-up data only covered language and social abilities up to 9 months of age. A definitive diagnosis of ASD is not possible until around 3 years of age, and language development follows different trajectories across individual infants by that age. Therefore, we cannot draw firm conclusions about the effects of reward-related neural responses on later language delay. However, it is important to note that a robust body of research has shown that infant siblings of children with ASD often experience difficulties in language and social development, even among those who do not ultimately receive an ASD diagnosis (Ozonoff et al. 2011, Hudry et al. 2014). Thus, our results are at least significant in demonstrating early neural differences between the LL and EL groups. Nevertheless, longitudinal follow-up into toddlerhood is indeed essential to determine whether the early reward-related neural differences observed here predict later language trajectories (Morrel et al. 2023). In fact, we are currently conducting such a follow-up study, which will help clarify how and when reward-related brain responses in EL infants impact their developmental outcomes.

Conclusion

Our study revealed significant differences in how maternal speech impacts early socio-communicative development in infants with varying ASD likelihoods. We found that LL infants showed widespread cortical activation and robust connectivity in response to maternal speech, particularly in areas related to the processing of voice, reward, and language. In contrast, EL infants exhibited minimal activation and connectivity in these regions. The lack of connectivity between voice-processing regions and reward-related areas in EL infants suggests a potential early biomarker for ASD. Additionally, the absence of significant reward-related language connectivity in EL infants highlights possible pathways leading to language delays. Our findings highlight the significance of neural sensitivity to familiar human voices, recognizing them as rewards crucial for speech acquisition.

Author contributions

Eiichi Hoshino (Data curation [equal], Formal analysis [lead], Methodology [equal], Validation [equal], Visualization [lead], Writing—original draft [lead], Writing—review & editing [equal]), Masahiro Hata (Data curation [supporting], Formal analysis [supporting], Investigation [lead], Methodology [supporting], Resources [supporting]), Mingdi Xu (Data curation [supporting], Formal analysis [supporting], Investigation [equal]), and Yasuyo Minagawa (Conceptualization [lead], Data curation [equal], Formal analysis [equal], Funding acquisition [lead], Investigation [equal], Methodology [lead], Project administration [lead], Resources [lead], Supervision [lead], Validation [equal], Writing—original draft [equal], Writing—review & editing [lead])

Supplementary data

Supplementary data are available at SCAN online.

Conflict of interest: The authors declare that they have no competing interests that could have appeared to influence the work reported in this paper.

Funding

MEXT Kakenhi (Grant-in-Aid for Scientific Research (S):19H05594, Grant-in-Aid for Specially Promoted Research: 23H05410, Grant-in-Aid for Scientific Research (A): 24H00176), World Premier International Research Center Initiative (WPI), MEXT, Japan, and CREST research project on social signalling (JPMJCR19A2), JST, Japan.

Data Availability

Due to the ethical restriction, the data cannot be made open access to the public. However, the data could be available from the corresponding author upon reasonable request.