Distinct ERP profiles for auditory processing in infants at-risk for autism and language impairment

Early identification of autism spectrum disorder (ASD) is crucial for the formulation of effective intervention programs. Language deficits may be a hallmark feature of ASD and language delay observed in ASD shows striking similarities to that observed in children with language impairment (LI). Auditory processing deficits are seen in both LI and ASD, however, they have not previously been compared directly using Event-Related Potentials (ERPs) in the two at-risk populations. This study aims to characterize infants at-risk for ASD (HR-ASD) at the electrophysiological level and to compare them with infants at-risk for LI (HR-LI) and controls, to find specific markers with predictive value. At 12-month-old, auditory processing in HR-ASD, HR-LI and controls was characterized via ERP oddball paradigm. All infants were then evaluated at 20 months, to investigate the associations between auditory processing and language/ASD-related outcomes. In both HR-ASD and HR-LI, mismatch response latency was delayed compared to controls, whereas only HR-ASD showed overall larger P3 amplitude compared to controls. Interestingly, these ERP measures correlated with later expressive vocabulary and M-CHAT critical items in the whole sample. These results may support the use of objective measurement of auditory processing to delineate pathophysiological mechanisms in ASD, as compared to LI.

Similar -analyses were conducted to compare the P3 mean amplitude for the STD and DEVD stimuli by Hemisphere and by Group (2 × 2 × 3 ANOVA, Model 2), revealing significant main effects of Stimulus type, F (1,58) = 45.80, p < 0.001, ŋ 2 = 0.441 and Group, F (2,58) = 3.65, p = 0.032, ŋ 2 = 0.112. As expected, mean amplitude was overall higher for DEVD (M = 1.82; SD = 1.94) than STD. Interestingly, Bonferroni corrected post-hoc comparisons following the main effect of Group revealed that overall mean amplitude (irrespectively of Stimulus type and Hemisphere) was higher for HR-ASD (M = 1.59, SD = 1.55) than TD infants (M = 0.42, SD = 1.28; Figure 1. Grand average waveforms for the three groups. For graphic presentation purposes only, two virtual channels were created by merging ERPs from all the channels included in two clusters. The two virtual channels corresponding to the left and right fronto-central clusters are shown. The waveform relative to the standard condition (STD, black line) is plotted against the waveforms relative to the frequency deviant condition (DEVF, red line) and the duration deviant condition (DEVD, blue line). In addition, the difference waveforms relative to MMRF (DEVF minus STD; orange dotted line) and MMRD (DEVD minus STD; light-blue dotted line) are plotted. ERPs are time-locked to the onset of the first tone in the pair. Negative voltage is plotted upward. In order to highlight the significant results, the overall time-window of interest where mean amplitude was computed and peak latency was extracted (350-620 ms) is defined through squares. Peak latency. The overall two-way ANOVA model contrasting peak latency of the two difference waveforms (MMRs) by Stimulus type (MMRF vs. MMRD), Hemisphere (left vs. right) and Group (TD vs. HR-ASD vs. HR-LI) revealed significant main effect of Stimulus types, F (1,58) = 269.72, p < 0.001, ŋ 2 = 0.823. As expected, peak latency was longer for MMRD (M = 528.66, SD = 29.89) than MMRF (M = 444.03; SD = 30.64). Interestingly, the main effect of Group was also statistically significant, F (1,58)

Associations between ERPs and ASD/LI outcome at 20 months of age. The associations between
ERPs components at 12-month-old and language/ASD-related outcomes at 20-month-old were assessed. Pearson's correlations were performed between ERP measures at 12 months and LDS percentile score at 20 months, whereas non-parametric Mann-Whitney U test was used to examine the association between ERP measures at 12 months and M-CHAT at 20 months (coded as dichotomous variable, 0 = less than two M-CHAT critical items; 1 = equal or above two M-CHAT critical items 39 ). For ERP measures, overall mean amplitudes of P3 for each stimulus type (i.e., STD, DEVF, DEVD) and overall peak latencies of MMR for each stimulus type (i.e., MMRF and MMRD) were entered in the association analyses. Follow-up data were available for 89% (n = 54), of the larger sample. Since there were no significant differences between groups on ASD/language outcomes (see Table 2), to maximize the statistical power the associations were computed using the combined groups. The analysis revealed that mean amplitude of P3 for DEVD condition was associated to M-CHAT critical scores, Mann-Whitney U = 99.00; p = 0.021 and peak latency of MMRD was associated to expressive vocabulary score, r (54) = −0.291, p = 0.036. Infants with larger P3 amplitude at 12 months were characterized by higher scores in the M-CHAT screening test at 20 months (see Fig. 4A), whereas infants with faster MMRD at 12 months produced more words at 20 months (see Fig. 4B). No significant correlation was found between LDS and M-CHAT [r (54)

Discussion
The present study aimed to directly compare two groups of 12-month-olds who were at high-risk for ASD and LI, at the electrophysiological level, and to associate their ERP responses with ASD-related traits and language outcome at 20 months. The results reported here are intriguing and unique, as this is the first study to provide abnormalities in ERP auditory processing profiles in ASD compared to LI high-risk infants and controls. In both ASD and LI at-risk groups, the latency of the MMR is delayed compared to controls. On the other hand, only the group at-risk for ASD shows overall larger P3 mean amplitude compared to controls. More interestingly, these ERP measures (regardless of risk status and clinical outcome) correlate with expressive vocabulary and critical M-CHAT items measured at 20 months of age in the whole sample.  A difference emerged when comparing infants with and without familial risk. Our results suggest that in both ASD and LI high-risk groups the peak latency of the MMR is delayed compared to controls, suggesting slower processing in both HR groups, specifically for fine-grained acoustic discrimination of changes in both frequency and duration. More interestingly, when we correlated the early ERP responses with later expressive language in the total sample, we found a significant association between the latency of the MMR for changes in duration and later expressive vocabulary.
In the domain of language and language-related impairments, our results are in line with previous studies that found that the ability to perform fine non-speech acoustic discrimination in early infancy is critically important to and highly predictive of later language development.
In both typically developing infants and in infants at familial risk for language/learning disorders (LLI) 29,30,32,33,37,40,41 , supporting auditory processing skills as early markers of risk and predictors of language development. Specifically, longer MMR latencies have been found in 2-month-old infants at risk for LLI in response to CV-syllables differing in vowel duration, suggesting delayed processing of the auditory stimulus change 40 . The latency of the mismatch negativity (MMN), which is often reported in older children and adults as an analog of the infants' MMR 36 , has often been explored in older children with the full-blown diagnosis of LI 42 . Despite the methodological variation between studies leading to inconsistent results, at least two studies have reported markedly longer MMN latency in children with LI 43,44 .
On the other hand, the findings are still mixed in the ASD literature 45,46 , with some reports suggesting an intact mismatch response 47,48 and others a delayed mismatch in ASD 45,46,49 , even if there are no infant studies, while convergent literature has suggested more prolonged mismatch latencies in ASD with concurrent LI 49,50 . Roberts and colleagues 49 assessed MMR to tones and speech sounds in children with ASD and in children with ASD + LI. Interestingly, it was found that ASD children showed delayed latencies compared to controls, but a more pronounced delay was found in children with ASD + LI, supporting that the time course of the auditory mismatch may represent a biomarker for ASD, and specifically for language in ASD.
Taken together, our results seem to be in line with previous studies. Since the latency of MMR is similarly impaired in the two HR groups and is correlated with later language abilities, it is possible that delayed MMR indexes a dysfunction of auditory processing skills necessary for language development, thus representing a neural dysfunction common to language impairment in both ASD and LI.
In addition to the described electrophysiological pattern related to the latency of the MMR which seems to be common to the two at-risk groups, different ERP profiles for the two groups emerge concerning the amplitude of the P3 response. Only the group at-risk for ASD showed a larger P3 mean amplitude, primarily for STD and DEVD, compared to infants at-risk for LI and to controls. Furthermore, we found significant correlations between the P3 amplitude for duration changes and ASD-related outcome, suggesting the role of this ERP component as a specific predictor of later socio-communicative development.
These results appear to be consistent with prior evidence that reported a significant advantage in processing low-level auditory stimuli in ASD [51][52][53] . In particular, this result may be aligned with the "Enhanced Perceptual Functioning" theory 54 , positing that ASD is characterized by an enhanced perception of low-level perceptual information and that neural networks underpinning perceptual processing are "over-specialized", which potentially results in a reduced tendency to integrate local information. One question arising from this result, though, is whether the findings are apparently in contrast with the idea that ASD is impaired in processing spectro-temporally complex auditory stimuli when high temporal demand is in place 24 , as is the case with our 'deviant' stimuli characterized by rapid changes 38 . It should be noted, however, that in our study the enhanced processing of the auditory information was not only found in response to 'deviant' stimuli, but additionally in response to the 'standard' stimuli. In the framework of the theoretical accounts of oddball processing 55 , the amplitude of the P3 component in response to 'deviant' stimuli is supposed to reflect the detection of the sensory stimulus feature mismatch and the following update of the stimulus representation. If no stimulus attribute change is detected (as in the case of STD stimuli), no update of the stimulus context is needed, and only sensory evoked potentials are recorded. For this reason, the larger amplitude found in our group at HR for ASD could be interpreted as increased sensory processing mechanisms, probably suggesting an overall enhanced sensitivity to low-level acoustic features, not specific for spectro-temporally complex changes.
Taken together, it could be that enhanced auditory processing might arise from a dysregulated sensory profile, supporting a possible hyper-sensitivity to perceptual level sensory features, especially for simple, low-level auditory stimuli. To further support our hypothesis, correlational analyses showed that different discrimination responses (larger P3 amplitude) are specifically associated with ASD-related phenotypes (i.e. M-CHAT). In particular, only the group at-risk for ASD showed overall larger P3 mean amplitude compared to controls and, regardless of risk status and clinical outcomes, a larger P3 amplitude was significantly associated with more ASD-related symptoms in all infants. Based on this result, we may hypothesize that enhanced neural responses to auditory discrimination processing very early in life carry significant and specific predictive value for ASD symptomatology during the course of development. Consistent with our results, a recent study 51 demonstrated a relationship between enhanced discrimination of low-level perceptual contrasts and specific aspects of the ASD phenotype, measuring auditory discrimination ability in 29 adolescents with ASD, 26 with optimal outcomes (individuals diagnosed with ASD before age five, who had no symptoms at time of testing) and 20 controls. Interestingly, it was found that only the group with ASD showed enhanced pitch discrimination, compared with the other two groups, supporting the hypothesis that enhanced rather than impaired pitch discrimination may lead to ASD symptomatology. Intriguingly, in this study enhanced pitch discrimination also seemed to be associated with early language milestones (i.e. age at which children produced their first words), both in the group with ASD and in the optimal outcome group. Similar associations between atypically larger brain responses to speech and non-speech sounds and poor language scores were found in a recent MEG study conducted on school-aged children with ASD 53 . In addition, in a case-study, the brain responses to auditory stimuli were assessed in a non-verbal child with ASD, using both MEG and EEG 52 . The non-verbal child showed larger ERP responses to non-speech auditory stimuli, as compared to controls and to verbal children with ASD, and this result was interpreted as proof of the association between enhanced discrimination skills and lack of language. This association between enhanced discrimination skills and language impairment was not found in the present study, where distinct patterns of correlations were found between larger P3 amplitude and ASD-related traits, and between delayed MMR and language. Thus, our results are more aligned with the literature on individuals with or at-risk for language impairment 30 , suggesting language deficits to be associated with impaired, rather than enhanced, auditory processing skills. However, it is difficult to compare our results with previous literature, as there are no other ERP studies exploring the correlation between auditory processing and ASD-related traits and/or language skills in ASD at such an early age. Further studies are needed to confirm the present results and disentangle the pattern of associations.
Certain limitations should be taken into consideration. First, language outcomes and ASD related traits were assessed solely by parental report. Therefore, we suggest that future studies should employ direct assessment of language skills and autism-related symptoms. Second, the sample size in each group is relatively small and a replication in independent samples or using larger datasets may further refine the observed effects. In particular, since the incidence rate of ASD among infants identified as HR based on familial risk is about 20% for ASD 4,5 and about 30% for LI 56 , it would be very interesting to examine developmental trajectories of P3/MMR components in a larger at-risk sample for ASD (who themselves will and will not ultimately be affected by ASD), as well as in a larger at-risk sample for LI (who themselves will and will not ultimately be affected by LI). Third, although we made an effort to match the three groups of infants for sex, it should be noted that the sample is not well representative of the male:female ratio commonly found in ASD and LI. Previous literature has shown that sex differences emerge in LI and ASD 4,57 , with boys more commonly diagnosed than girls, and in auditory processing 58 . Animal studies have shown that male mice with cortical malformations are impaired in auditory processing compared to controls, whereas females with the same induced malformations show no deficits 58 . It would be very interesting to test whether P3 responses significantly differentiate males from females. This remains an interesting hypothesis to be tested in future research with better suited sample sizes.
In conclusion, these results support the use of objective measurement of auditory processing to delineate specific pathophysiological mechanisms in ASD. They can help identify infants at-risk for ASD and possibly provide specific markers that help to distinguish ASD from linguistic delays. These results have strong implications for the future implementation of early intervention programs. A better understanding of shared and unique mechanisms underlying different disorders may support the development of more effective interventions by indicating whether interventions developed for one disorder are likely to be helpful for the another, and by identifying specific treatment targets that may be shared across disorders or are unique to each disorder.

Methods
Sample. At 12 months of age, 20 infants at high-risk for ASD (HR-ASD), 19 infants at high risk for LI (HR-LI) and 22 typically developing infants (TD) took part in the study. The TD group consisted of infants recruited by local advertisements in two hospitals in Northern Italy, whereas the HR groups were recruited at the Medea Institute. The recruitment of TD and HR-LI is part of a currently on-going longitudinal project aiming at identifying early risk markers for language and learning impairment 29,37,38 . Recruitment of HR-ASD was made possible through collaboration with the Italian Network for Early Detection of Autism Spectrum Disorders (NIDA Network).
Scientific REPORts | (2018) 8:715 | DOI:10.1038/s41598-017-19009-y All infants were included in the study if: (1) both parents were native-Italian speakers, (2) gestational age was ≥36 weeks, (3) birth weight was ≥2000 grams, (4) Bayley Cognitive Score 59 was ≥7. The criterion for being included in the HR group was to have at least one sibling with a certified diagnosis of LI for the HR-LI group or a diagnosis of ASD for the HR-ASD group. There was no overlap in terms of recruited subjects.
Descriptive statistics of demographic and clinical characteristics are shown in Table 2. Participants were matched at the group level for sex, age, gestational weeks and socio-economic status (SES). Written informed consent was obtained from all parents prior to testing. The experiment was performed in accordance with relevant guidelines and regulations and was approved by the Medea Ethical and Scientific Committee. The M-CHAT is one of the most promising parental questionnaires commonly used in primary care and clinical settings [60][61][62] . It requires the parent to report on the presence of specific child behaviors in a checklist of 23 yes/ no items. The presence of an abnormal behavior was assigned a score of 1 and the total score interpreted. Six items pertaining to social relatedness and communication (critical items) were found to have the best discriminability between children diagnosed with and without ASD. Thus, for the purpose of this study, children were considered as showing symptoms of ASD if 2 or more critical items were "no" (M-CHAT critical scoring 39 ). The critical items are: (1) Does your child take an interest in other children? (2) Does your child ever use his/her index finger to point, to indicate interest in something? (3) Does your child ever bring objects over to you (parent) to show you something? (4) Does your child imitate you? (e.g. if you make a face, will your child imitate it?) (5) Does your child respond to his/her name when called? (6) If you point at a toy across the room, does your child look at it? Language Development Survey (LDS). Infants' expressive language was rated by the Language Development Survey (LDS), which was standardized for the Italian population 63 . LDS includes 310 words arranged into 14 semantic categories (e.g. food, animals, people, vehicles) and contains high frequency words (e.g. daddy), as well as less common words (e.g. supermarket). The number of total words that the child produces corresponds to the total vocabulary score. Norms are available from 18 to 35 months of age 64 . Percentile scores were entered in the analyses. Experiment design. Auditory processing was assessed by means of an electrophysiological task tapping the ability to process rapidly changing and complex auditory stimuli. The paradigm and stimuli were identical to those used in previous research 29,37,38 . A non-speech multi-feature oddball paradigm was used in which pairs of brief complex tones with an inter-stimulus interval (ISI) of 70 ms were presented. Figure 5 schematically represents the experimental paradigm. The first tone in the pair always had a fundamental frequency of 100 Hz and duration of 70 ms. For 'standard' tone-pairs (STD stimuli) the same 100 Hz tone was repeated twice (i.e., 100-100 Hz). Two deviant tone-pairs differing with respect to the second tone were presented: in 'deviant for frequency' stimuli (DEVF), the second tone had a fundamental frequency of 300 Hz; in 'deviant for duration' stimuli (DEVD), the second tone had a duration of 200 ms. The stimuli were presented in a passive oddball paradigm in which 1200 stimuli (80% STD, 10% DEVF, 10% DEVD) were delivered in a pseudo-random order, so that at least three 'standard' tone-pairs were presented before each 'deviant' pair, with an inter-trial interval (offset-to-onset, ITI) varying randomly from 700 ms to 900 ms. EEG data acquisition and preprocessing. During EEG recording, the children were seated on their caregiver's lap in a sound-attenuated and electrically shielded room. All stimuli were presented free field at an intensity of 75 dB via speakers located on either side of and equidistant (95 cm) from the subject. The ERP experiment lasted approximately 25 min., with an additional 10-15 min. before the experiment to prepare the participant for the EEG recording.
Auditory ERPs were recorded from 60 scalp sites using a dense-array EGI recording system (Electric Geodesic, Inc., Eugene, Oregon). Vertex was used as an online reference. EEG was sampled at 250 Hz and bandpass filtered (0.1-100 Hz) online. After recording, the data were processed using EEGLAB 65 and ERPLAB 66 . The EEG data processing procedures were identical to those used in Cantiani and colleagues 29 . Continuous EEG data were bandpass filtered offline at 0.5-30 Hz. Channels with high impedance (>50 KΩ), or visually evident noise, were interpolated with a spherical spline (never more than 12 of the 60 channels). No interpolation was performed on the online reference Cz. The signals were then re-referenced to an average reference. For the STD, only responses to the immediate pre-deviant STD were included in the average. The continuous EEG was segmented according to stimulus type (pre-deviant STD, DEVF and DEVD) with 900 ms epoch lengths. The 100 ms pre-stimulus segment was used for baseline correction. Bad EEG epochs were identified and rejected using two automatic criteria, followed by visual inspection. First, a moving window (200 ms width, 50 ms step) was used to identify segments containing signals with voltage differences >150 μV. Then, trials whose spectrum (in one or more channels) deviated from the baseline by +25/−100 dB in the frequencies >20 Hz were removed 67 .
Analytic strategy. To examine the role of auditory processing, we focused on the large positive response corresponding to the mismatch response (P3 component), reflecting a neural change detection process. Both amplitude and latency of this component were analyzed.
Time windows and electrode sites to be submitted to statistical analyses were selected based on mass univariate analyses applied to a subset of ERP data 29 . This procedure allows the identification of channel clusters and time windows where differences between stimulus types are significant, taking into account the application of appropriate corrections for multiple comparisons 68 (for a full description of this procedure and the permutation test results that drove the selection of the time windows and the electrode sites to be submitted to statistical analyses, refer to Cantiani et al. 29 ). For each participant, ERPs were extracted from a subset of 18 electrodes localized in the left and right fronto-central areas. Data were then averaged in two clusters corresponding to left and right fronto-central areas, each including nine channels (left cluster: E7, E9, E11, E12, E13, E14, E15, E18, E19; right cluster: E2, E3, E53, E54, E54, E57, E58, E59, E60; see Supplementary Information for the graphical representation of the selected electrodes and clusters placement map). Never more than 4 of the 18 channels included in the final fronto-central clusters involved in the analyses were interpolated (M = 0.6, SD = 0.9, range = 0-4). Following the above-mentioned mass-univariate analyses, mean amplitude was calculated for different time-windows: 350-550 ms for STD and DEVF and a 420-620 ms for DEVD. Mean amplitude was extracted within these time windows (instead of peak amplitude) because, as expected, clearly identifiable peaks were not present in the STD condition 29,38 . In addition, to investigate the latency of the component, peak latency was extracted in the same time-windows for the two difference waveforms (MMRF and MMRD). We decided to extract peak latency of the difference waveforms because, as previously stated, clearly identifiable peaks were not present in the STD condition.
Statistical analysis. Differences in the mean amplitude of the P3 component were investigated using two separate 3-way ANOVA models in order to directly contrast each 'deviant' stimulus type with the 'standard' 29 . This analytical strategy allowed us to investigate the amplitude of the mismatch response as a measure of auditory discrimination. Model 1 compared P3 mean amplitude to Stimulus Type (STD vs. DEVF) by Hemisphere (left vs. right) by Group (TD vs. HR-ASD vs. HR-LI; 2 × 2 × 3). Model 2 was a parallel analysis with STD and DEVD in Stimulus Type.
Finally, Pearson's product moment correlation and Mann-Whitney U test analyses were conducted to assess associations between infant ERP components (mean amplitude of P3 and peak latency of MMR separated by stimulus type) (1) 20-month-old language abilities (expressive vocabulary percentile score -LDS) (2) 20-month-old screening score of ASD risk (M-CHAT).