Reduced engagement with social stimuli in 6-month-old infants with later autism spectrum disorder: a longitudinal prospective study of infants at high familial risk

In a repeated-measures ANOVA on P400 latency by age (6 or 12 months), gender (male, female), stimulus (face, object), and laterality (left, right), P400 latency peaked earlier to faces than objects (F(1,99) = 6.70, p = 0.011, η ² = 0.011), was faster over right than left electrodes (F(1,99) = 4.77, p = 0.031, η ² = 0.046), and had a shorter latency at 12 than 6 months (F(1,99) = 18.1, p < 0.001, η ² = 0.15). Male infants also showed faster P400 latencies than female infants (M female = 539.5, M male = 506.1; F(1,101) = 4.9, p = 0.03, η ² = 0.045). In a repeated-measures ANOVA on P400 amplitude by age (6 or 12 months), gender (male, female), stimulus (face, object), and laterality (left, right), P400 amplitude was greater to objects than faces (F (1,99) = 23.4, p < 0.001, η ² = 0.19). This is consistent with previous work [41, 43, 64].

In repeated-measures ANOVAs on Nc amplitude for early and late subcomponents separately by age (6 or 12 months), gender (male, female), and stimulus (face, object), Nc overall amplitude was more negative to objects than faces for both the early and late subcomponents (early F(1,104) = 9.9, p < 0.001, η ² = 0.09; late F(1,104) = 7.0, p = 0.009, η ² = 0.063). For the early Nc component, there was a significant interaction between age and gender (F(1,104) = 5.1, p = 0.026, η ² = 0.047) and a main effect of gender (early: F(1,104) = 4.01, p = 0.048, η ² = 0.037), driven by the fact that age-related change was significant in males (F(1,53) = 3.94, p = 0.05, η ² = 0.07) but not females (F(1,51) = 1.53, p = 0.22, η ² = 0.03). For the late Nc component, amplitudes were generally more negative at 12 than 6 months (F(1,104) = 10.3, p = 0.002, η ² = 0.09).

In repeated-measures ANOVAs on Nc duration by age (6 or 12 months), gender (male, female), and stimulus (face, object), the duration of the Nc was longer to objects than faces (F(1,84) = 3.9, p = 0.05, η ² = 0.045) and longer at 6 months than 12 months (F(1,84) = 19.9, p < 0.001, η ² = 0.19). Again, these results are comparable to previous work; for example, typically developing 3- to 4-year-old children show a more negative Nc to objects than faces [41].

To establish which of the ERP findings were interrelated, we examined patterns of correlations between P400 latency and amplitude and Nc latency and amplitude to faces. At 6 months but not 12 months, faster P400 latency to faces over the left hemisphere was correlated with a shorter duration Nc to faces (r(41) = 0.38, p = 0.014) and a less negative Nc response to faces (r(50) = −0.4, p = 0.006). Similarly, a less negative Nc to faces was highly correlated with a shorter Nc latency (r(42) = −0.9, p < 0.001). Taken together, these findings confirm (as expected) that a fast P400 latency to faces and a shorter and less negative Nc are interrelated and may be associated with lesser attention capture by social stimuli.

Experiment 1 confirmed that our paradigms elicit normative patterns of responding in typically developing infants. This includes a faster and smaller P400 to faces than objects, a smaller and shorter Nc to faces than objects, and a longer peak look to faces than objects during habituation.

In Experiment 2, we used these paradigms to examine differences in attention capture by faces and objects in infants at high risk for ASD. We reasoned that if attention capture by social stimuli were reduced in infants with later ASD, we would see an exaggeration of the faster P400 to faces and the smaller and shorter Nc to faces versus objects (reflecting an exaggeration of the typical tendency for greater attention capture by objects versus faces in this age range). Further, we predicted that we would see a reduction in the duration of the peak look to faces during the habituation paradigm.

Within the high-risk group, we correlated key experimental variables (mean peak look duration to faces and objects at 6 months, P400 latency to faces over the left and right hemispheres, and late Nc amplitude) with key behavioral variables (Mullen verbal and nonverbal standard scores collected concurrently with experimental variables; and ADOS total scores at 24 m). There were no significant correlations between experimental variables and concurrent behavioral variables, confirming that results were not confounded with concurrent developmental level (rs < .25, ps > 0.07). However, higher 24-month ADOS total scores were significantly correlated with later peak look position to faces at 6 months (r(30) = 0.58, p = 0.001), shorter P4 latency to faces over the left hemisphere (r(18) = −0.46, p = 0.05), and marginally significantly with shorter mean peak look duration to faces and objects at 6 months (r(32) = −0.32, p = 0.08). These relations are illustrated in Fig. 4. Taken together, these results support categorical analyses in suggesting that shorter epochs of attention to social stimuli were related to later autistic symptomatology.

Habituation paradigms are the most widely used method of assessing visual attention in infancy [33]. Consistent with previous work, in the present study, a normative sample of low-risk typically developing infants and infants from low and high-risk groups who did not later develop ASD showed a longer peak look to faces than objects [63], decreased peak look duration over the first year [75], and dishabituated to a novel face or an object after delays of up to a minute [76]. However, infants who developed ASD showed a shorter peak look that was later in the habituation function for faces than objects. The “dual-process” account [32] of habituation argues that the position of a peak look in the sequence reflects a process of “sensitization” that is associated with a spike in parasympathetic arousal that increases attention to the stimulus [33]. The later peak look to faces observed in infants with later ASD (that was also continuously related to poorer social functioning across the high-risk group as a whole) may thus indicate that disruptions to sensitization are associated with later ASD symptoms. Sensitization is thought to be important in engaging deeper levels of processing and has been implicated in the facilitation of learning by infant-directed speech [34, 35]. Disrupted sensitization could have a negative impact on social development. Interestingly, a longitudinal study of typically developing infants revealed a cluster who showed rapid habituation times that did not decreases with age; these infants showed atypical developmental decreases in sensitization (reflected in heart rate deceleration) and poor language skills at 24 months [73, 77]. Although preliminary, a similar cluster was observed in the present dataset and contained a disproportionate number of infants with later ASD. These infants also showed poorer social functioning at outcome than infants with later ASD with other patterns of early habituation data. Shorter peak looks and the delayed peak look to faces in infants with later ASD could thus reflect altered timing of the physiological sensitization response to social stimuli, and thus delayed or disrupt engagement of deeper levels of attention.

The faster P400, and smaller and less sustained Nc ERP response, may also reflect a reduced depth of processing for social stimuli. The Nc component has been extensively studied in infancy. Because the Nc is modulated by novelty and stimulus salience [37], and is larger (more negative) to stimuli presented during physiologically defined states of attention [39], the Nc is thought to reflect attention engagement [40]. Thus, the smaller and shorter Nc observed in infants with later ASD would be expected to reflect reduced attention engagement to faces. The function of the P400 in infancy is less clear but may relate to semantic aspects of extracting information from faces [40]. Results with typically developing infants in experiment 1 indicate that faster P400 latency relates to a smaller and shorter Nc, consistent with our hypothesis that a faster P400 latency to faces may reflect reduced depth of processing. Other work with toddlers and young children with ASD has also noted atypicalities in both the P400 and Nc response to faces. For example, toddlers and young children with ASD show developmentally delayed modulation of the Nc by facial familiarity, such that responses resemble those seen in younger typically developing toddlers [27]. The same group of toddlers with ASD also show atypical modulation of the P400 by facial familiarity; specifically, typically developing toddlers showed a larger P400 amplitude to unfamiliar than familiar faces, while toddlers with ASD did not [41]. Further, the same group of children with ASD tested at age 4 to 5 years showed a slower Nc peak latency to faces than objects, which was reversed in children from that group who had received 2 years of intensive treatment that improved social functioning [59]. Thus, our results in infancy are consistent with the sensitivity of the P400 and Nc ERP components to atypical social attention in the early development of children with ASD. Future work should determine whether our infant attention metrics are also sensitive to the effects of early intervention [78].

Our interpretation of reduced depth of social attention engagement is consistent with other works. For example, Chawarska and colleagues [17, 18] have observed reduced monitoring of social scenes at 6 months. This is particularly apparent when faces were accompanied by speech, suggesting that deficits may be exaggerated during more complex naturalistic presentations. Further, Jones and Klin [21] noted a declining pattern of attention to the eyes of faces between 2 and 6 months, suggesting an emerging profile of disrupted social attention. Finally, Wass and colleagues [79] noted shorter durations of individual fixations during scanning of complex scenes (although comparison of fixations to faces versus non-faces did not reach significance). The present study adds to this literature by showing that there are atypicalities not only in the direction of visual attention but also in its temporal dynamics. Taken together, there is mounting evidence that there are early disruptions in the depth, direction, and quality of social attention in the first 6 months of life that precede the emergence of clear behavioral symptoms of ASD. Such disruptions could reduce the quality of social information processing, leading to cascading deficits in social behaviors that emerge over time [6].

However, other studies have not noted differences in social attention in infants with later ASD. For example, Ozonoff and colleagues [11] found that infants with later ASD showed a typical number of gazes to faces per minute during a cognitive task at 6 months, and Elsabbagh and colleagues [19, 20] found normal rapid orienting to a face in a complex array and normal shifting of attention between the eyes and mouth of a person telling nursery rhymes. One potential explanation stemming from the present study requires us to differentiate between processes of attention orienting and attention maintenance (variously called sustained attention or attention holding), which are subserved by different neural systems [80]. Typically, physiologically defined states of sustained attention follow an initial orienting response, and only emerge 1 to 2 s after a period of looking to a stimulus begins [81]. It may be that attention orienting to social stimuli is relatively typical at 6 months in infants with later ASD [82], but attention maintenance is disrupted. The present study is consistent with this pattern: we observed atypicalities in the duration and quality of attention, while controlling initial attention capture (since stimuli were presented while infants were looking at the screen). Further, atypicalities were noted in the late but not early Nc components, consistent with this hypothesis. Studies that measure only the number of gazes to a face per minute [11] and studies that use complex arrays that elicit very short individual fixations of average duration 600 ms [19] may not detect overall deficits in attention allocation. In contrast, studies that measure overall monitoring of a social stimulus over a longer period [17, 18] may be more sensitive to the accumulated effects of shorter individual attention epochs. If this explanation is correct, future work should contrast attention orienting and attention engagement within the same paradigm with infants with later ASD; one valuable approach would be to use heart rate-defined phases of attention to separate effects on orienting and attention maintenance.

Six-month-old high-risk infants who later met criteria for ASD showed significantly shorter peak looks to faces than high-risk infants without early ASD. Dishabituation was unaffected, indicating that shortened habituation times did not reflect failure to learn about the stimulus. Rather, infants with later ASD showed a similar magnitude of dishabituation despite significantly shorter habituation times. These results are in apparent contrast to previous work showing increased habituation times to faces versus houses in clinically-referred toddlers with an ASD diagnosis [22]. What could account for this disparity? In the study by Webb and colleagues, prolonged habituation times were only apparent in those with the highest level of symptoms (ADOS total scores over 17). Consistent with other work with high-risk infants, the present sample tended to be more mildly affected: indeed, only one child had an ADOS score over 17 at 24 months. Possibly, there is a complex relation between sustained attention and levels of ASD symptoms. Alternatively, developmental stage may be a critical factor, particularly since no group differences were observed in patterns of habituation at 12 months in the present study. Indeed, several recent studies of high-risk infants reveal opposite patterns of disruption to those seen in young children with ASD. For example, long-range underconnectivity has been observed in children with ASD [83], while others have found overconnectivity in high-risk infants [84] and infants with later ASD [85, 86]. Further, while children with ASD show a longer latency pupillary constriction to a sudden increase in luminance [87], infants at high familial risk show a faster constriction [88]. Temperament trajectories also suggest that high-risk infants are less active than controls in infancy but more active than controls in toddlerhood [89]. Thus, it is possible that ASD-related disruptions to sustained attention are expressed differently in infancy and toddlerhood.

Of note, 12-month-old infants with typical development (Experiment 1) and those without later ASD (Experiment 2) showed a decrease in peak look durations with age. There are several interpretations of this finding. Possibly, normative decreases in peak look duration reflect increased encoding speed across the first year of life, rather than changes in attention. Factors driving the variance in look duration with age may thus differ from those driving variance in look duration between diagnosis groups. Alternatively, there may be decreases in attention engagement to simple static stimuli across the first year. Indeed, several studies show that while look durations to simple stimuli typically decrease, look durations to more complex dynamic stimuli increase across the first year (for review [90]). This may reflect decreasing attention capture by simple static stimuli with developmental time and highlights the importance of both studies with highly controlled stimuli and studies involving more complex naturalistic settings.

One alternative explanation of our results should be considered. The duration of individual looks during a habituation paradigm is influenced not only by sustained attention but also by the speed at which infants encode the stimulus [30, 49]. Thus, one possible alternative explanation is that infants who go on to ASD at 6 months are actually more efficient at structural face encoding. The observed faster P400 latencies to faces in the ASD+ group could also be considered a reflection of more rapid perceptual processing, which may contribute to more rapid encoding. Increased efficiency of perceptual encoding could potentially be related to greater experience with faces in the early development of infants with later ASD. Indeed two reports have observed greater attention to faces or eyes in the first 6 months in infants with later ASD relative to low-risk typically developing infants [11, 21]. This related increase in experience with faces would provide the potential to support faster learning [91]. Subsequently, the progressive decrease in interest in faces seen in infants with later ASD over the second year of life may fail to reinforce this initially advanced developmental trajectory of the face recognition system, leading to the gradual emergence of relatively slower learning [22] and delayed neural correlates of face processing [27]. This is consistent with models that propose that the emergence of ASD reduces social interest, which in turn affects the experience-dependent process of social learning [15]. Examining the developmental inter-relation between naturalistic measures of social attention and neurocognitive measures of social processing in high-risk infants will provide a further test of this model.

Better encoding or perceptual processing would, however, not be consistent with a range of other findings from our study. First, this would not account for the finding of a shorter and smaller Nc response to faces, since the Nc has not typically been associated with basic aspects of perceptual processing. Rather, the Nc is widely accepted as an index of attention engagement [40]. The correlation between the less negative and shorter Nc component with a faster P400 response in our large normative cohort in experiment 1 is consistent with the findings of experiment 2, in which infants with later ASD showed both a faster P400 response to faces and a smaller and shorter Nc. Taken together, this data suggests that the shorter P400 latency to faces, reduced Nc duration and amplitude in infants with later ASD may together reflect reduced attention engagement to social stimuli. Of note, although our ERP findings by outcome group are preliminary because of the relatively small number of infants with later ASD, we also observed a continuous association between P400 latency to faces (not objects) and later ADOS scores within the large high-risk group as a whole, supporting this finding. Second, the finding of delayed sensitization to faces (significantly later peak look to faces at 6 months) in the present dataset is not consistent with faster perceptual processing or encoding. Delayed sensitization also did not only occur at the group level: there was a strong continuous relation between delayed sensitization to faces and higher later ADOS scores. This delayed sensitization is strongly suggestive of reduced or altered attention engagement with faces in infants with later symptoms of ASD. Further, a cluster of infants showed delayed peak look to faces, and shorter peak look to faces and objects. The infants with later ASD within this cluster (n = 4/6) were significantly more impaired than infants with later ASD who did not show this pattern. These results are not consistent with shorter peak looks reflecting a “strength.” Third, there were no differences in the latency of the P1 or N290 ERP components (see Additional file 2) between infants with and without later ASD, and these are the two components that have been most closely associated with perceptual aspects of face processing. No difference in the latency of P1 or N290 ERP components has also been reported in other cohorts of infants with later ASD [36]. If perceptual processing or encoding of faces were more rapid in infants with later ASD, differences would be expected on these components. Finally, while a range of other studies have observed altered attention to social stimuli in young infants with later ASD (e.g., [17, 18]), there have been no reports consistent with more efficient encoding or perceptual processing of face stimuli (though see [79, 92] for possible evidence of more efficient processing of nonsocial stimuli). Thus, the weight of evidence from this study and from the previous literature supports our interpretation of reduced attention engagement to social stimuli at 6 months.

One important discussion point is some apparent inconsistencies between the present ERP results and a previous study [36]. Elsabbagh and colleagues presented repeated pictures of intact and scrambled faces to 6- to 10-month-old HR infants. There were no group differences in P400 latency to static faces in relation to HR-ASD+ outcome, although infants with later ASD did show less sensitivity over the P400 component to gaze shifts in a second set of stimuli [36]. However, Elsabbagh and colleagues did observe a faster P400 latency to faces than noise stimuli in infants with later ASD that was not observed in other groups, and this is consistent with the current dataset. Thus, the modulation of P400 latency by social versus nonsocial content was consistent across studies, but the overall decrease in P400 latency to faces was only observed in the present report. One possible explanation is clearly the relatively small sample size included in both studies in relation to the range of functioning levels seen in children with ASD. A second possibility is that the difference in findings reflects a developmental shift in which effects are found in the younger sample in this study (M = 6.1 months, range 6 to 8) but not the slightly older sample included in Elsabbagh’s work (M = 7.9 months, range 6 to 10) or in our 12-month-old sample. Alternatively, infants with later ASD may also have a particular advantage in the more challenging task of processing the trial-unique stimuli used in the present study relative to the repeated pictures used by Elsabbagh and colleagues.

Important in our interpretations, the present study used two-dimensional static screen-based stimuli. This has been a common strategy in studies of face processing in older children and adults with ASD [22, 59, 93], and in young infants [44, 94], and allows more precise matching of stimuli between social and nonsocial conditions. However, it will be important to examine in future work whether such results generalize to more naturalistic social stimuli. In early infancy, EEG markers may show greater sensitivity to social stimuli in live or dynamic contexts [36, 95, 96], and a range of evidence suggests that infants learn more effectively from live than recorded stimuli (e.g., [97, 98]). Thus, examining aspects of attention engagement to live dynamic social and nonsocial stimuli will be an important step for future work.

We related infant markers to ASD diagnosis at age 2 years in order to examine early learning mechanisms that might underlie the early appearance of ASD symptoms. This has become an increasingly common strategy in work with high-risk infants [86, 99, 100]. Diagnostic stability of clinical diagnoses made at age 2 years is typically high [4, 101, 102]. However, the relationship between variance in diagnosis trajectory (e.g., the presence or absence of regression) [103‐105] based on outcome age and the underlying attention mechanisms or neurocognitive differences is unclear. There continue to be dynamic changes in ASD diagnosis, and symptom expression across the lifespan and emphasis on very early mechanisms will allow the development of more targeted pre-diagnostic interventions for high-risk infants [78].

Lastly, an important limitation is the relatively small number of children with later ASD in the ERP analyses. Despite this smaller group, we had sufficient power to detect significant and complementary effects at 6 months within both methodologies employed. The ERP findings were significantly correlated with habituation variables at 6 months, suggesting that the two experimental measures were tapping related processes. The number of children who met later criteria for ASD was comparable with recent studies of infants with later ASD [18, 21, 86]. Several aspects of the present data are also consistent with previous work. (1) Elsabbagh and colleagues also observed a faster P400 to faces than phase-scrambled faces in 6-month-old infants with later ASD, which is replicated in the present dataset. (2) Wass and colleagues report shorter fixation durations during viewing of a static scene at 6 months [79], which is consistent in direction with the shorter peak looks observed in the present dataset. (3) The direction of effects observed in infants with later ASD across different ERP components (a faster P400 to faces, and a shorter and smaller Nc to faces) is consistent with the internal structure of ERP findings from the large sample of typically developing infants tested in experiment 1 (in which a faster P400 response to faces correlated with a shorter and smaller Nc to faces). (4) Data from the habituation paradigm are also consistent with reductions in social attention at 6 months reported by other groups [17, 18]. Thus, our results are supported by both internal and external validation approaches.