Cross-modal coherence and incoherence of early infant interactive behavior: links to attachment in infants born very preterm or full-term

ABSTRACT       Infants exhibit flexibly organized configurations of facial, vocal, affective, and motor behavior during caregiver-infant interactions that convey convergent messages about their internal states and desires. Prior work documents that greater cross-modal discrepancy at 4 months predicts disorganized attachment. Here, we evaluated whether: very preterm (VPT) or full-term (FT) status predicts cross-modal coherence or incoherence in infants’ behavior with the caregiver at 3 months; and, regardless of prematurity, whether cross-modal interactive coherence or incoherence predicts 12-month attachment. Participants included 155 infants (85 FT; 70 VPT), and their mothers followed from birth to 12 months (corrected age). Infants’ cross-modal coherent and incoherent responses were scored microanalytically from videotaped en-face interactions. Infants’ attachment security was evaluated during Ainsworth’s Strange Situation. Infants born VPT exhibited more incoherent cross-modal responses and insecure attachment than infants born FT. Regardless of prematurity, infants’ coherent and incoherent cross-modal interactive behaviors at 3 months predicted different attachment patterns at 12 months.


Introduction
A fundamental aspect of Bowlby's attachment theory is its focus on the biological bases of infant behaviors that elicit caregiving and promote the formation of the infant-caregiver attachment relationship (Bowlby, 1969). Behaviors such as looking and smiling at the caregiver signal infants' desire to initiate or maintain social interaction with them, and behaviors such as reaching or crawling toward the caregiver serve to bring the infant into proximity or physical contact with their caregiver. Other behaviors such as gaze aversion, angry fussing, resisting contact, or moving away from the caregiver have the predictable outcome of interrupting or terminating social interaction with the caregiver. These interactive behaviors emerge early in life and gradually become organized within an "attachment behavior system" (Bowlby, 1969).
During early caregiver-infant social interactions, infants typically communicate their states, needs, and interests to their caregiver using a variety of co-occurring behaviors and emotional displays (e.g. gazing at the caregiver while smiling; or averting gaze, head, and body away from the caregiver while fussing). In early work, these coordinated displays were termed "affective configurations" (Weinberg & Tronick, 1994). Cross-modal interactive behaviors that are coherent convey convergent information to caregivers about infants' current state and needs, making it easier for caregivers to read their signals accurately and respond to them appropriately. Infants exhibit this multiplicity of coordinated interactive behaviors in flexible ways that vary according to internal and external cues and afford the infancy greater efficacy in goal-corrected responses. In this way, coherence in infants' interactive behavior, flexibly organized, is thought to contribute to positive and reciprocal caregiver-infant interactions. In turn, early attuned, positive social interactions ("attachment-in-the-making," Bowlby, 1969) are associated with the development of secure attachment relationships in later infancy (Feldman, 2007).
Not all infants signal their attachment needs to caregivers in a clear, coherent manner. This cross-modal incoherence refers to interactive behaviors expressed by infants during early caregiver-infant interactions that convey discrepant and contradictory information, such as reaching for the caregiver while gaze averting. Caregivers may find infants' incoherent cross-modal behavioral responses difficult to read and respond to accurately. This is problematic because, when different modalities convey discrepant information, social communicative signals can be misinterpreted, and the subsequent dyadic interaction disturbed. For instance, incoherent cross-modal interactive behavior may jeopardize the accomplishment of infants' interactive and attachment goals and interfere with their parents' ability to read their cues accurately and respond in a sensitive, appropriate manner. If persistent, infants' cross-modal incoherence during caregiver-infant social interactions could undermine formation of a clear-cut pattern of secure attachment by the end of the first year (Beebe et al., 2012).
To date, few studies have evaluated the coherence and incoherence of infants' interactive behavior during early caregiver-infant interactions or assessed their links to later attachment patterns. Infants might be expected to display incoherent cross-modal responses more frequently in early life because very young infants have only rudimentary capacities to express their current state, needs, and sense of the world to their caregiver (Weinberg et al., 2006). Incoherent interactive behavior may also be especially prevalent among biologically vulnerable infants such as those born very preterm (VPT; delivery at < 32 gestational weeks). The current longitudinal study aims to shed lighter on this issue by describing the prevalence of infants' coherent and incoherent interactive behaviors during early caregiverinfant interaction at 3 months of age, in a sample of infants born VPT and infants born FT, and evaluating whether infants' coherent and incoherent interactive behaviors are associated with their later attachment organization within each group.

Early face to face interactions: Infants' and caregivers' contributions
Many scholars of early social development point out that, to engage in positive and reciprocal interactions, infants and their caregivers must engage in mutually responsive, co-regulated interchanges, in which each partner works to match the intensity, quality, and timing of their responses to those of their partner's (Banella & Tronick, 2019;Beebe et al., 2010). The quality of mutual regulation that is achieved by the dyad is never perfect and is affected by the dynamic interplay of three critical processes . The first is the integrity and capacity of young infants' central nervous system to organize and control physiologic states and behavior. The second is the integrity and clarity of infants' communicative system, including the central nervous system centers that control and generate messages and meanings and the accompanying motor, attention, and emotional systems that allow infants to manifest these messages (e.g. facial expressions, gaze direction, bodily postures, and gestures). One of the earliest emerging functions of infants' communicative system is to express their state and goals to caregivers, which allows caregivers to apprehend the extent to which they are succeeding or failing in reading and responding to their infants' cues accurately. The third process, reciprocal to the second, is caregivers' capacity to appropriately read their infant's communicative signals accurately and their willingness to take appropriate responsive action.
Together, these processes contribute to the relative success or failure of the caregiverinfant mutual regulatory process, i.e. the capacity of each interactant -infant and caregiver -to perceive and respond to the communicative signals of the partner, so that each partner can achieve their goals Feldman, 2007). These mutual regulatory processes are critical because infants' successful engagement with the world of people and things depends on the effectiveness of the caregiver-infant communicative system in regulating young infants' dysregulated states and re-establishing their positive engagement (Beeghly et al., 2016;. Although caregivers play an important role by scaffolding infants' efforts to initiate and maintain their engagement in positive interactions, young infants also contribute actively to the quality of these interactions. By three months of age, most infants have gained better control of their head and upper-body posture, which supports their capacity for controlling gaze direction, coordinating visual-motor behaviors, and producing early gestures. However, individual differences in biological vulnerability (such as VPT birth) may delay or undermine infants' capacity to exhibit coherent cross-modal interactive behavior. Compared to infants born FT, infants born VPT at the same (corrected) age have more difficulty organizing and controlling their behavioral responses and physiological states during early social interactions, including their capacity for regulating states of arousal, autonomic, motor, and social behaviors, and the quality, coherence, and contingency of their affective responses (Mouradian et al., 2000).
In a systematic review (Ritchie et al., 2015), infants born VPT differed significantly from infants born FT in the display and timing of their behavioral and affective responses during dyadic social interactions at 6 months of age. They also exhibited lower inhibitory control and poorly coordinated social responses. In several longitudinal studies, infants born preterm exhibited more problems in regulating attention and emotions than infants born FT, and had a higher rate of internalizing and externalizing behavior problems in later life (Aarnoudse-Moens et al., 2009;Butcher et al., 2002;Clark et al., 2008;Feldman, 2009;Gerstein et al., 2017;Treyvaud et al., 2013;Wolf et al., 2002).
It therefore seems reasonable to expect that the early neurodevelopmental difficulties of infants born VPT may undermine their ability to coordinate their displays of affect, gaze direction, and other behaviors in a coherent fashion during early social interactions. Infants' incoherent or discrepant cross-modal interactive displays may send misleading or conflicting messages to the caregiver. For instance, when caregivers smile or vocalize to their infant, and their infant exhibits an incoherent (mixed or discrepant) cross-modal interactive response (e.g. smiling while averting gaze, the head, or the body), caregivers may perceive that their infant dislikes or wants to stop interacting with them. Caregivers may also find it difficult to read infants' communicative signals accurately, and consequently, may be more likely to misinterpret and respond inappropriately to them (Wolke et al., 2013). If this misattuned dyadic pattern is repeated over time, it could lead to the formation of insecure attachment relationships in later infancy.

Early face to face interaction and infant attachment
Several investigators have described individual differences in infants' social responses in face-to-face mother-infant interactions and found that some behaviors predict infants' later attachment organization (e.g. Braungart-Rieker et al., 2001E. Z. Tronick et al., 1982;Ekas et al., 2013;Kogan & Carter, 1996). In a meta-analytic review of this literature, Mesman and colleagues found that infants who exhibit more positive affect (d = .23) and less negative affect (d = .24) are more likely to be classified as securely attached to their mothers at 12 months of age, compared to other infants (Mesman et al., 2009).
Other investigators have described characteristics of the dyadic interchange (matching, mismatching, and repair) and how each partner contributes to the interaction (e.g. Forcada-Guex et al., 2011). However, relatively few have carried out longitudinal research to evaluate whether the cross-modal coherence or incoherence of infants' affective and behavioral displays are associated with their attachment organization at the end of the first year. In one notable exception, Beebe et al. (2010) used micro-analyses to describe the type, intensity, orientation, and modality of altered mother-infant contingency during face-to-face interactions at 4 months. They found that infants who with their mothers had exhibited behaviorally incoherent configurations of cross-modal behavior at 4 months were more likely to be classified as having disorganized attachment at 12 months. In their work, the authors also highlight the range of intrapersonal and interpersonal communication disturbances that result from early discordant and contradictory interactions.

Very preterm birth, cross-modal coherence of early infant interactive behavior, and attachment
Biological vulnerabilities such as preterm birth may undermine infants' ability to signal their attachment needs clearly to caregivers. Compared to infants born FT or moderate-to late preterm, infants born VPT, have greater difficulties with emotion regulation and expression. Yaari et al. (2018) demonstrated that infants born VPT exhibit more negative affect and gaze aversion than infants born FT or moderate-to-late preterm. Infants born VPT also have difficulties establishing mutually co-regulated interactions with their caregivers (Sansavini et al., 2015), particularly during early infancy (Korja et al., 2012). Given the greater interactive difficulties of infants born VPT, we aimed in the current study to learn more about these challenges by comparing the cross-modal coherence and incoherence of early interactive behavior in a sample of infants born VPT and infants born FT.
Several studies also indicate that infants born VPT have a higher incidence of insecure attachment compared to infants born FT (e.g. Ruiz et al., 2018;Udry-Jørgensen et al., 2011;Wille, 1991;Wolke et al., 2013), although findings vary across studies (Korja et al., 2012). However, the specific early predictors of attachment in the VPT population are understudied. Although a growing body of literature indicates that infants' early regulatory behavior predicts their later attachment organization, the results vary according to infants' birth risk status, including preterm birth. Some investigators report that infants born prematurely with very low weight, who exhibited intensive and negative affect, and struggled to return to normal interaction after exposure to the maternal still-face, were more likely to be classified as insecurely attached in late infancy (Fuertes et al., 2022). However, to our knowledge, the coherence and incoherence of infants' early cross-modal interactive behavior in VPT and FT groups have not been described and compared, nor have their associations with later attachment been evaluated.
The Present Study. The primary goal of the current longitudinal study was to further our understanding about the cross-modal coherence and incoherence of infants' interactive behavior with the caregiver in early infancy, by first describing and comparing the prevalence of each pattern in a sample of infants born VPT and infants born FT, and then determining whether each pattern was associated with infants' later attachment within each group. To accomplish this goal, we addressed four specific aims.
(1) The first aim was to assess the frequency of coherent and incoherent cross-modal interactive behavior in each group (VPT and FT) during mother-infant face-to-face interaction at 3 months (corrected age), and to test whether the frequency of coherent and incoherent cross-modal interactive behaviors differed in the VPT and FT groups. We expected that infants in each group would exhibit both coherent and incoherent cross-modal interactive behaviors, but that infants born FT would be more likely than infants born VPT to exhibit coherent cross-modal behavior, and infants born VPT would be more likely to display incoherent crossmodal interactive behavior (i.e. discrepant or contradictory signaling). These hypotheses were made based on the greater neurological and developmental immaturity and vulnerability of young infants born VPT in the literature (Aarnoudse-Moens et al., 2009;Wolf et al., 2002). Also, because mothers of infants born prematurely experience higher levels of psychological symptoms than mothers of infants born FT (Vigod et al., 2010), we carried out exploratory analyses to test the association of this variable with infants' coherent and incoherent crossmodal interactive behavior at 3 months and their attachment patterns at 12 months within the VPT group.
(2) The second aim was to determine whether the frequency of infants' coherent and incoherent cross-modal interactive behavior at 3 months (corrected age) in each group was associated with infant-mother attachment at 12 months (corrected age). Based on findings from prior research (e.g. Beebe, 2020), we expected that infants who exhibited a higher frequency of coherent cross-modal interactive behaviors (regardless of their birth status) would be more likely to have a secure attachment with their caregiver at 12 months (Beebe et al., 2010). (3) A third aim was exploratory and built on the results of the second aim. The goal was to evaluate whether the frequency of infants' coherent and incoherent cross-modal interactive responses predicted unique variance in attachment outcomes at 12 months, after controlling for potential confounding variables such as infant birth risk factors (e.g. gestational weeks at delivery, birth weight) and demographic/ familial risk factors such as low maternal age or education. We expected that, even after controlling for relevant covariates, the cross-modal coherence and incoherence of infants' early interactive behavior would be significant predictors of their later attachment classification. This hypothesis was made because infants are active partners in social interactions, and their behavior contributes significantly to the quality of emerging mother-infant relationships over time (e.g. Banella & Tronick, 2019;Beebe et al., 2010). Coherent cross-modal patterns of interactive behavior are easier for caregivers to read and respond to appropriately than incoherent patterns, thus, they should foster more positive and reciprocal social interactions (e.g. Braungart-Rieker et al., 2001) that, over time, may lead to secure attachment.

Sample characteristics and recruitment
Participants included 70 infants born VPT (40 girls, 30 boys) and 85 infants born FT (39 girls, 46 boys), and their mothers, who met eligibility requirements and were followed longitudinally from the infants' birth to 12 months postpartum (corrected for gestational weeks). Following recruitment, all eligible 155 dyads participated in two study visits and had complete data at 3 and 12 months. The distribution of male and female infants in each group did not differ significantly (χ 2 = 1.948; p = ns). All participants resided in the greater Lisbon and Porto metropolitan areas of Portugal. Most mothers were Portuguese-Caucasian in race/ethnicity and from urban working-and low-to-middle-class socioeconomic backgrounds. Recruitment took place from 2017 to 2019. Three research assistants contacted potential participants in the maternity wards and NICUs of Lisbon and Porto metropolitan hospitals and explained the study's purpose and procedures to them.
To be eligible to participate, infants and mothers had to meet the following inclusion criteria. Infants in the VPT group had to be: (a) less than 32 weeks of gestational age at delivery and (b) very low birth weight, i.e. less than 2500 grams. All infants had to be free of sensory or neuromotor disabilities, serious health conditions (e.g. chronic lung disease), or congenital anomalies at the time of recruitment. Mothers in both the VPT and FT groups had to be (a) at least 18 years old at the time of delivery; (b) their infant's primary caregiver; and (c) free of mental health problems including clinical depression or drug/ alcohol addiction problems, as determined by medical record review and maternal selfreport.
Information about mothers' mental health status at intake was obtained via medical record review. Current prenatal health vigilance initiatives in Portugal mandate that health service providers routinely screen pregnant women for physical and mental health problems. When mothers screen positive for mental health symptoms such as depression, health professionals refer them for a formal diagnostic mental health assessment. Mothers who received a mental health diagnosis were excluded from the current study. Since mothers in the VPT group experienced premature labor and delivery, they did not attend all prenatal health visits and consequently had incomplete or nonexistent mental health screens. Therefore, mothers of infants born VPT were administered the Brief Symptoms Inventory at 3 months. Mothers with clinical symptomatology were referred to Early Interventions services and not included in this study.
A total of 185 mother-infant dyads were initially approached for recruitment. Of these, 27 infants born VPT and three infants born FT were excluded from the present study based on the inclusion criteria or because they had missing data at follow-up (e.g. the child was ill at the time of the data collection, the family moved away, or the family was lost to follow-up for other personal reasons). This resulted in a final sample of 155 dyads who met inclusion criteria and had complete data at both the 3-month and 12-month visits.
Results of t-tests within the VPT group indicated that the 70 participants in the VPT group did not differ significantly from the 27 non-participants on birth weight, [t = 1.012, ns], gestational age [t = .983, ns], maternal age [t = .999, ns], or maternal education [t = 1.9045, ns]. This suggests that there was no differential attrition within the VPT group based on key medical or sociodemographic risk status variables.
Perinatal and demographic characteristics. Descriptive statistics for the perinatal and demographic characteristics of participants in the FT and VPT groups are provided in Table 1.
Perinatal characteristics. As expected, infants in the FT group had a significantly higher number of gestational weeks at birth, heavier birthweights, and higher Apgar scores at 1 minute and 5 minutes after delivery than infants born VPT. Consistent with population statistics in Portugal (Numerato et al., 2017), all infants in the VPT group were admitted to the Neonatal Intensive Care Unit (NICU) after delivery but varied in the length of time they stayed there (range = 9-209 days, M = 50.66; SD = 35.27). In contrast, no infants in the FT group were admitted to the NICU. The mean number of days in the NICU for infants in the VPT group is also consistent with that reported in European population statistics for this population, with average days in the NICU ranging from 46.2 (Italy) to 61.0 (Sweden) (see revision in Numerato et al., 2017). Most infants born VPT in this study were between 28 and 32 gestational weeks at delivery, and only five infants were less than 27 weeks of gestation. These five infants born extremely preterm were included in the study because their NICU stay was less than 3 months and they had no further health complications. Demographic characteristics. Demographic information was obtained via maternal interviews and medical record reviews at the time of recruitment. Participants in the FT and VPT groups differed significantly on some demographic variables (see Table 1). Mothers in the FT group were slightly younger and had more years of completed education than mothers in the VPT group. Infants born FT had more siblings than infants born VPT. Moreover, all infants in the FT group lived with both parents in the same household, whereas four mothers in the VPT group were single and raising their infants alone. Mothers in the FT and VPT groups also differed in employment status at the time of recruitment. Six mothers in the FT group were unemployed, whereas 10 mothers in the VPT group were unemployed.

Procedures
All procedures in this study were conducted according to the ethical guidelines presented in the Declaration of Helsinki, with written informed consent obtained from all mothers before any assessment or data collection took place. All procedures were approved by the Ethics Committees of all Health Units and Hospitals involved, as well as by the Portuguese Data Protection Commission.
At 3 and 12 months postpartum (corrected age for infants born VPT), recruited mothers were recontacted to schedule a follow-up visit to the laboratory with their infant. At the 3-month visit, mother-infant dyads were videotaped during face-to-face interactions, and at the 12-month visit, dyads were observed during Ainsworth's Strange Situation (Ainsworth et al., 1978). At the 3-month visit, mothers of infant born VPT also reported on their psychological symptoms using the Brief Symptom Inventory (BSI, Derogatis & Melisaratos, 1983).
Face-to-Face Interactions (FFI). During the FFI, mothers were instructed to play with their infants en face, as they normally would at home without using toys or pacifiers for three minutes. Mothers and infants were videotaped using two cameras, one focused on the mother's face and upper torso, and the other focused on the infant's face and body. Both cameras were connected to an image mixer software that generated a timesynchronized split-screen image of each partner on a single video record.
The Strange Situation (SS, Ainsworth et al., 1978) is a 21-minute laboratory-based observational paradigm consisting of eight successive 3-minute episodes (or less if the infant is distressed) of increasing mild stress for the infant, including being introduced to an unfamiliar playroom, interacting with an unfamiliar adult stranger, and two motherinfant separations and reunions.

Measures
Infants' coherent and incoherent cross-modal interactive behavior. To describe infants' coherent and incoherent cross-modal interactive behavior, three steps of analyses were performed. In the first step, before the current study began, three coders observed 20 FFI videos and described infants' coherent and incoherent cross-modal configurations using narratives describing co-occurring gaze direction, vocalizations, body directions, and facial expressions. The coders agreed on 10 coherent configurations and 17 incoherent configurations.
Following this preliminary work, in a second step, to confirm the configurations obtained in the narratives and during the coders' conference, a single trained coder blinded to infant birth status, background variables, and the study's aims scored discrete infant behaviors second-by-second using the Infant Regulatory Scoring System (IRSS, E. Z. Tronick & Weinberg, 1990) and Noldus Observer computer-assisted software using the same 20 cases used in narratives. Specifically, the following categories of infant behavior and affect were scored in separate passes: the infant's direction of gaze (looking at mother, looking at an object, looking away, or closing eyes), vocalizations (positive, neutral, or negative), pick-me-up gestures (reaching arms toward the mother, the chair, or the infants' own body), affective facial expressions (positive, negative, and neutral), selfcomforting behaviors (self-oral, object-oral, oral-mom, touches self, self-clasp, rock), distancing behaviors (getaway, arch, pull/push away, or screen out) and autonomic stress indicators (e.g. spitting up or hiccups). A second coder scored second-by-second 25% of the cases for reliability purposes. Intercoder reliability was good (Cohen's kappa =.74).
After each behavior was scored, the frequency of each infant coherent and incoherent cross-modal interactive configuration (e.g. looking at the mother and smiling) was ascertained with univariate statistics. The behavioral configurations exhibited by at least 60% of the infants in each group (FT and VPT) were included in the current study in two main categories: i) Coherent cross-modal interactive behaviors and ii) Incoherent cross-modal interactive behaviors.
Coherent cross-modal interactive behaviors included the following three combinations of infant affect and behaviors (all were scored while the mother was trying to engage the infant in interaction): (i) Looking at the mother and smiling, ii) Looking at the mother and vocalizing (positive or neutral vocalizations), and iii) Looking at the mother and directing their arms and body in the mother's direction. Each behavioral configuration was computed (present or not present) second-by-second for each infant in both the FT and VPT groups. The total frequency of all instances of coherent cross-modal configurations exhibited by each infant was used in the statistical analyses of the current study.
Incoherent (discrepant) cross-modal interactive behaviors included the following six configurations of infant affect and behaviors (all were scored while the mother was trying to engage the infant in interaction): (i) Looking away and smiling, ii) Looking at an object and smiling, iii) Looking away and vocalizing, iv) Looking at an object and vocalizing, v) Closing eyes and verbally protesting while directing their arms and body in mother's direction; and vi) Looking away turning head, while directing arms in mother's direction. As was the case for coherent cross-modal behavior patterns, the total frequency of incoherent cross-modal behavior combinations was tallied for each infant and used in the statistical analyses.
A few coherent and incoherent cross-modal behavioral configurations were excluded from this study because only one or very few infants exhibited them, and because our goal was to evaluate infants' common coherent and incoherent cross-modal interactive responses. For descriptive purposes, the excluded configurations, and the number of number of infants who exhibited them are presented in Table 2.
To evaluate inter-coder reliability, two blinded coders then independently scored a randomly selected set of 60 videotapes for infant coherent and incoherent crossmodal interactive behaviors (40 from the FT sample and 20 from the VPT sample). The mean Cohen's kappa coefficient for the inter-rater agreement was 0.79, denoting a very good agreement. To evaluate intra-observer reliability over time (coder drift), the first coder then re-scored 20 randomly selected cases after a three-month interval (Kappa coefficient k = 0.90, indicating excellent reliability).
Brief Symptom Inventory (BSI, Derogatis & Melisaratos, 1983). Mothers of infants born VPT reported on their psychological distress symptoms on the BSI at the 3-month visit. The BSI is a 53-item instrument consisting of self-reported symptoms drawn from the Symptom Checklist-90-R, designed to assess a range of psychological symptoms tapping emotional, somatic, and interpersonal dimensions of distress. Participants rate symptoms on a 5-point Likert scale. Items are scored on a scale of 1 (strongly disagree) to 5 (strongly agree), with total possible scores ranging from 0 to 212. On the BSI, higher scores (on each of the corresponding subscales and on the global indices of distress) indicate greater psychological distress experienced by the respondent. The BSI has been used with a wide range of populations and is known to have good reliability and validity (Derogatis & Melisaratos, 1983). Mothers' total BSI score in the VPT group was evaluated in the current study.
Attachment classifications. Videotapes of infants' attachment behavior in the Strange Situation at 12 months were scored by two trained, reliable coders following the procedures developed by Ainsworth et al. (1978). The coders who scored attachment differed from the face-to-face interaction coders and were unaware of infants' cross-modal (i) Looking at the mother, smiling and reaching their arms toward mother direction (ii) Smiling, vocalizing and offering his/her arms in mother direction (iii) Looking at mother, smiling and spit bubbles (iv) Looking at mother and laugh (v) Vocalizing positively and sucking mother hand (vi) Pulling mother hand on his/her direction 3 3 1 1 1 1 (i) Smiling with his/her eyes close (ii) Pulling away from mother while vocalizing positively (iii) Pulling away from mother while vocalizing and smiling (iv) Closing his/her eyes while vocalizing positively (v) Swing his/her body back and forth while vocalizing positively (vi) Swing his/her body left and right with vocalizing positively (vii) Swing his/her body left and right when the mother was holding his/her hands and talking at him (viii) Swing his/her hands in the air while vocalizing positively toward mother (ix) Smiling and vocalizing negatively (x) Turning her/his head (left and right) when the mother was holding his/her hands Only three cases were scored as D in the FT sample and only seven in the VPT sample. Because there were not enough D cases within each group to permit a reliable withingroup evaluation of disorganized attachment, analyses including the D attachment category were carried out only in the whole sample.

Analytic plan
Preliminary analyses were conducted to evaluate the distributional properties of the study variables. Then, several sets of statistical analyses were conducted to address the specific aims of the study. First, the significance of differences in the frequency of infants' coherent and incoherent cross-modal interactive behavior in the FT and VPT groups was evaluated using student t-tests. Second, the distribution of attachment patterns for the FT and VPT groups at 12 months was compared using chi-square tests. Additionally, the Bonferroni test correction was used as a measure of proportional reduction in Type 1 error. Third, within the VPT group, Pearson correlations were used to assess the association between maternal psychological symptoms (BSI total score) and infants' coherent and incoherent behavior frequencies. Fourth, one-way ANOVA was used to evaluate the association between infants' coherent and incoherent cross-modal interactive behavior frequencies and the four patterns of attachment for the sample as a whole. This analysis was repeated within the VPT and FT groups separately (excluding disorganized attachment, due to the low number of D cases within each group). Fifth, one-way ANOVA was used to evaluate the association between perinatal and demographic/family variables and the three patterns of attachment. In one-way ANOVA, Tukey´s post hoc tests were used to test differences between specific attachment groups. Finally, logistic regression analyses were used to identify predictors of attachment security from the following variables: gestational age, gestational birth weight, infant coherent cross-modal interactive behavior, and infant incoherent cross-modal interactive behavior.
Effect size for the t-test analyses was calculated with Cohen's d, for the crosstab analyses with Cramer's, and for the ANOVA analyses with Eta squared. No missing data imputation was carried out because all families in the current sample participated in both study visits at 3 and 12 months and had complete data.

Differences in infants' coherent and incoherent cross-modal interactive behavior in the FT and VPT groups
The first aim was to describe the frequency of infants' coherent and incoherent crossmodal interactive behaviors during the FFI at 3 months (both scored in the context of the mother attempting to engage the infant in social interaction) and to evaluate whether this frequency differed significantly between the FT and VPT groups. See Table 3 for descriptive statistics. Infants born FT had significantly higher frequencies of coherent cross-modal behavior than infants born VPT, with a very large effect size (Cohen, 1988). Specific group differences were observed for looking at the mother and smiling, looking at the mother and vocalizing, and the total frequency of coherent cross-modal behaviors. In contrast, infants born VPT displayed more frequent incoherent cross-modal interactive behaviors than infants born FT. Specifically, infants born VPT were significantly more likely than infants born FT to exhibit the following incoherent cross-modal behavioral combinations: Looking away and smiling; Looking away and vocalizing; and Looking away turning head, while directing arms in mother's direction (all scored in the context of the mother attempting to engage the infant in social interaction). The magnitude of these group differences was moderate.

Differences in infants' coherent and incoherent cross-modal interactive behavior according to attachment organization
The third aim was to ascertain whether the frequency of infants' coherent and incoherent cross-modal interactive behavior at 3 months was associated with their attachment organization at 12 months. To evaluate this aim, we first evaluated the distribution of infant attachment classifications within the two groups and tested the significance of between-group differences. Table 4, infants in the FT and VPT groups differed significantly in attachment organization with large effect sizes. Infants born FT were also more likely than infants born VPT to have a secure attachment [χ 2 (2) = 18.033, p < .001]. Moreover, the insecure-avoidant attachment pattern was more prevalent in the VPT group (28.6%) than in the FT group (10.6%). Importantly, results of Cramer's V provided evidence for significant sample size effects, [ϕ C = .341, p < .001]. Also, disorganized attachment is more prevalent in the VPT group (10%) compared to the FT group (3.5%).

Distribution of infant attachment classifications in the FT and VPT groups. As seen in
Associations between infants' coherent and incoherent cross-modal interactive behavior at 3 months and attachment organization at 12 months. As presented in Table 5, securely attached infants (regardless of birth status) were more likely to exhibit coherent crossmodal interactive behaviors with their mothers during FFI at 3 months, and less likely to express incoherent cross-modal interactive behaviors, compared to infants with insecuredisorganized, insecure-avoidant or insecure-ambivalent attachment. Also, infants with  insecure-avoidant attachment had lower frequency of coherent cross-modal interactive behaviors than infants with disorganized attachment.
Moreover, the following specific categories of coherent cross-modal interactive behaviors were displayed significantly more frequently among infants later classified as securely attached, compared to infants classified as insecurely attached: looking at the mother and smiling, looking at the mother and vocalizing, and looking at the mother and directing their arms and body in her direction (see Table 5).
In turn, the frequency of specific types of infants' incoherent cross-modal interactive behaviors at 3 months was lower among infants with a secure attachment than in infants with an insecure attachment, including disorganized attachment. No significant differences were found among infants with an insecure attachment (including disorganized attachment) regarding the frequency of incoherent cross-modal interactive behaviors.
Moreover, specific categories of incoherent cross-modal interactive behaviors, including directing their arms in the mother's direction with closed eyes while verbally protesting, looking away and vocalizing, or looking away and turning their head away occurred more frequently among infants who were later classified as insecure-ambivalent with medium-high effect sizes (Lakens, 2013). In these categories, infants with a disorganized attachment had a higher frequency of incoherent cross-modal interactive behaviors than infants classified with a secure attachment. Vocalizing or smiling while looking at an object occurred more often among infants who later were classified as having an insecure-avoidant attachment than among infants with other attachment organizations, with a medium-high effect size (Lakens, 2013).
A similar set of analyses was performed separately within each group (FT and VPT). Infants with a disorganized attachment were not included in these analyses due to the low number of D cases within each group. Results are presented in Tables 6 and 7. The results of these within-group analyses are similar to those observed across groups.
Regarding incoherent cross-modal interactive behaviors, Looking to an object while mother tries to engage and Looking to an object and smiling while mother tries to engage were more prevalent among infants with an insecure-avoidant attachment compared to other infants, in both the FT and VPT groups. However, Looking away and smiling as the mother tries to engage was more likely in infants with an insecure-ambivalent attachment in the FT group, whereas in the VPT group this behavior was more likely among infants with an insecure-avoidant attachment. The incoherent cross-modal interactive behaviors: Looking to an object and vocalizing as the mother tries to engage; Close eyes while verbally protesting and directing their arms and body in the mother's direction; and Looking away turning her/his head, while directing their arms in the mother's direction were more prevalent among infants with an insecure-ambivalent attachment compared to other infants, in both the FT and VPT groups.
Associations between infants' coherent and incoherent cross-modal interactive behavior and mothers' psychological symptoms in the VPT group. BSI was administered only to mothers of infants born VPT in the current study. Within the VPT group, mothers' selfreported psychological symptoms on the BSI (total score) at 3 months postpartum were positively correlated with the following incoherent cross-modal interactive behavior configurations (all coded in the context of the mother attempting to engage the infant in interaction): infants directing their arms in the mother's direction while closing their eyes (r = .265; p < .030), looking at an object and smiling (r = .257; p < .036) and the total number of infants' incoherent cross-modal interactive behaviors (r = .253; p < .039). However, no significant correlations were found between the BSI total score and coherent cross modal interactive behaviors. Also, no significant associations were found between mothers' BSI total score and infants' attachment organization within the VPT group.

Predictors of attachment
The fourth aim was to conduct binary logistic regressions to test whether the crossmodal coherence or incoherence of infants' interactive behavior during mother-infant interactions at 3 months predicted unique variance in infants' attachment outcomes at 12 months, after controlling for perinatal or demographic covariates. As presented in Table 8, preliminary results indicated that perinatal factors (but not demographic factors) were significantly associated with attachment classifications at 12 months. Infants with secure attachment had a higher gestational age at delivery, heavier birth weight, and higher Apgar scores at the first and the fifth minute post-birth, compared to infants with insecure attachment classifications. Thus, only these perinatal variables were evaluated as potential covariates in the binary logistic regressions.
Prior to conducting the regressions, multicollinearity analyses were carried out. Results indicated that there was high collinearity (VIF above 10) between gestational weeks and birth weight, and between Apgar score at first and fifth minute (VIF above 3). Therefore, only gestational age and 5-minute Apgar score were included as covariates in the binary logistic regressions. These variables were chosen because they had higher effect sizes in the bivariate analyses. Secure attachment was contrasted with all insecure attachment patterns. The model fit indices were 182.0 and Nagelkerk r 2 = .942. Results of the covariate-controlled binary logistic regressions indicated that only infants' coherent (p = .001) and incoherent (p = .001) cross-modal interactive behaviors were significantly associated with attachment security. See Table 9.

Discussion
According to Bowlby (1969), attachment relationships form gradually across the first year and reflect the quality of infant-caregiver dyadic interactions across the first year ("attachment-in-the-making"). During early social interactions, infants communicate their needs and goals to caregivers using cross-modal configurations of affect and behavior. However, not all infants can organize their communicative behaviors in a clear and coherent manner to achieve their attachment goals. Infants with biological vulnerabilities such as preterm birth may have more difficulty communicating their needs and goals to caregivers than infants born FT. Little is known about the characteristics of the cross-modal coherence and incoherence in infants' early interactive behavior in groups of infants varying in biological risk, and even less is known about their associations with later attachment formation. The main goal of this study, therefore, was to increase our understanding of infants' coherent and incoherent cross-modal interactive behavior with the caregiver at 3 months (corrected age) by describing and comparing their prevalence among infants born FT or VPT and evaluating their associations with infants' later attachment organization at 12 months (corrected age).
As expected, infants in each group (FT and VPT) exhibited both coherent and incoherent patterns of cross-modal behaviors during mother-infant en-face interaction at 3 months (corrected age). In previous studies that evaluated dyadic matching/mismatching/repair processes in mother-infant interactions (e.g. Cohn & Tronick, 1988;E. Tronick, 2005), dyadic mismatching and disengagement occur frequently, even during interactions in which both partners are described as engaging in reciprocal interactions and sharing positive affect. The high prevalence of dyadic mismatching is thought to occur because young infants are immature, and the quality of early social interactions stems from a complex and dynamic infant -adult communication system that integrates each partner's contributions. These contributions are multi-faceted and include positive/negative affect; social orientation of responses, such as social-or self-oriented responses; the coherence of facial, vocal, affective, and kinesthetic responses, dyadic states of consciousness (i.e. shared meaning of both partners), and mutual regulatory processes that scaffold infants' engagement with the world of people, objects, and themselves (E. . Weinberg and Tronick (1994) also report that mismatching in configurations of facial, vocal, and gestural behaviors are likely to occur during early mother-infant interactions. The relatively frequent occurrence of incoherent clusters of infants' crossmodal interactive behavior and dyadic mismatching may reflect the fact that the dyadic communication system is still emerging and requires frequent dyadic adjustments and adaptations stemming from changing dyadic matching, mismatching, and reparation processes. Thus, it is not surprising that in the present study we found incoherent combinations of cross-modal interactive behavior among infants in both the FT and VPT groups. Typically, these interactive responses reflect the ongoing processing of the emerging infant-caregiver communication system. Nevertheless, infants born VPT exhibited more frequent incoherent cross-modal interactive responses than infants born FT. This result raises concerns for this population because coherent and incoherent interactive responses were differentially associated with infants' later attachment organization at 12 months, which in turn is thought to be foundational to later socio-emotional competence. Additionally, perinatal variables (i.e. gestational age, birth weight, and Apgar scores) were each significantly correlated with infants' coherent and incoherent cross-modal interactive behavior. These results support our hypothesis that the greater neurological and developmental immaturity and vulnerability of young infants born VPT are associated with a poorer capacity for behavioral regulation, which in turn may contribute to a higher prevalence of incoherent cross-modal interactive behavioral responses during early mother-infant interactions. Consistent with this hypothesis, others report that infants born VPT have poorer scores on ratings of motor coordination and less optimal psychomotor development at 6 months than infants born FT (Ritchie et al., 2015). VPT birth occurs during a particularly vulnerable stage of the organization of the central nervous system. Disruptions associated with VPT birth can negatively impact the development of infants' capacity for developing autonomic stability, motor maturity, state organization, and the regulation of attention, affect, and social behavior, all of which contribute to early communicative behavior (Mcgowan et al., 2022). Attesting to these difficulties, we found a significant between-group difference in the percentage of secure attachment (FT: 64.7% vs VPT: 31.4%). Although, we need to stress that other Portuguese samples reported slightly lower values of attachment security in infants born FT (approximately 54%; see revision in Faria et al., 2014).
Our results further indicate that a higher prevalence of infants' coherent cross-modal interactive behavior in early life is associated with their later attachment security in both the VPT and FT groups, whether we consider the total frequency of coherent cross-modal behaviors or the particular coherent cross-modal behaviors (e.g. Looking at mother and smiling, Looking at mother and vocalizing, Looking at mother and directing their arms and body toward mother). This finding is consistent with similar reports in the preterm literature showing that early positive, reciprocal, and contingent interactions as well as with infants' more frequent positive affect during mother-infant en face interactions are linked to later attachment security (e.g. Forcada-Guex et al., 2006;Poehlmann & Fiese, 2001). In the current study, infants born VPT or FT who exhibited contradictory (incoherent) cross-modal responses during early social interactions (e.g. directing their arms in the mother's direction while closing their eyes and verbally protesting or looking away by turning the head) were more likely to be classified as having an insecure-ambivalent attachment by the end of the first year.
Additionally, we performed further analyses computing both groups together (VPT and PT). As reported in past studies, we also found that infants with disorganized attachment were more likely to exhibit a lower frequency of incoherent cross-modal interactive behaviors and a higher frequency of incoherent cross-modal interactive behaviors than infants with secure attachment. For instance, Beebe et al. (2010) found that infants' discrepant affect during early en-face interactions is associated with later attachment disorganization. However, our findings indicate that these infants also exhibit a higher frequency of coherent interactive behaviors than insecure-avoidant infants. Contrary to our hypotheses, infants with a disorganized attachment were not the ones with the lowest scores of coherent cross-modal interactive behaviors nor the ones with highest scores of incoherent cross-modal interactive behaviors. It will be interesting to study whether these results persist later on infants development e.g. at 9 months of age). Considering that socioemotional developmental changes depend on the continuous bidirectional and reciprocal influences between the infant and their social context, the inconsistency associated with disorganized attachment may be only emergent at 3 months.
Interestingly, all attachment patterns in our sample were associated with distinct profiles of coherent/incoherent cross-modal interactive behaviors. Infants who exhibited incoherent responses during early social interactions (e.g. directing their arms in the mother's direction while closing their eyes and verbally protesting or looking away by turning the head) were more likely to be classified as having an insecure-ambivalent attachment. In contrast, infants with an insecure-avoidant attachment had a lower frequency of coherent cross-modal interactive behaviors and were more likely to vocalize or smile to an object or without focus. Each attachment pattern seems coherent or incoherent in its particular way, indicating the lability of early socioemotional processes. Additionally, infants with an insecure-avoidant attachment may be at high risk for reciprocal en-face interactions since they presented the lowest total of coherent crossmodal interactive behaviors (including infants with disorganized attachment). In contrast, infants with a secure attachment exhibited a lower frequency of incoherent cross-modal interactive behaviors and a higher frequency of coherent cross-modal interactive behaviors than other infants.
In sum, our findings add to a growing literature indicating a significant association between the coherence or incoherence of infants' cross-modal interactive behavior observed as early as 3 months of life and their later attachment organization. Infants appear to be inherently motivated to communicate with caregivers and establish dyadic states of consciousness with them (E. Tronick et al., 2020). This "attachment-in-themaking" process (Bowlby, 1969) starts early in the first year and occurs repeatedly in the context of everyday dyadic interactions that occur moment-by-moment and over time, in which the infant and adult mutually regulate and scaffold one another's social or object engagement (Beeghly et al., 2016). Over time, both partners construct common meanings during social interactions that reflect their history of shared emotions, intentions, motor responses, and physiological states, and which generate new dyadic states of consciousness (E. Tronick, 2005), which contributes to the quality of early social interactions and emerging attachment relationships .
The associations between perinatal factors and attachment quality that were identified in the current study are also noteworthy. In bivariate analyses, infants' gestational age, birth weight, and Apgar scores at the first and fifth minutes were all significantly higher among infants classified as securely attached. However, results of the binary logistic regressions controlling for perinatal and demographic variables revealed that these associations were no longer significant when infants' coherent or incoherent crossmodal interactive behaviors were also included in the model. Thus, the primary predictors of attachment security in this study are the cross-modal coherence and incoherence of infants' early interactive behavior rather than variables denoting their level of birth or demographic risk. These findings are consistent with the idea that infants' behavioral integration and modulation actively contribute to the quality of mother-infant social interactions. If infants' early coherent and incoherent cross-modal behavioral responses are relatively stable over time, they may reflect ongoing iterative interactive processes in the mother-child relationship, which may, in turn, contribute to later attachment formation (Cox et al., 2010;Gerstein & Poehlmann-Tynan, 2015;Mikulincer et al., 2003). This possibility should be evaluated in future research.
A possible confounding factor in this research is that mothers of infants with biological vulnerability, such as infants born VPT, are more likely than mothers of infants born FT to experience postpartum psychological distress symptoms (Vigod et al., 2010). In the current study, mothers' self-reported psychological symptoms on the BSI (total score) were evaluated in the VPT group at 3 months postpartum. Results of within-group bivariate correlations revealed that higher BSI total scores were associated with a higher frequency of infants' incoherent cross-modal interactive behaviors, but not with the frequency of infants' coherent cross modal interactive behaviors or their later attachment organization. Prior studies found that infants of depressed mothers are characterized as showing more anger and less affective sharing in free-play situations with their mothers (Stein et al., 1991), and that these results may be exacerbated in infants born preterm (Korja et al., 2008). It is possible that infants born preterm rely more on their social partners to support their communicative efforts, while mothers with depression may struggle to respond appropriately to and support their infants' interactive needs (e.g. to reinforce positive and coherent behaviors). We speculate that, when infants born VPT have depressed mothers and display coherent messages, their mothers are able to respond appropriately. However, when infants' interactive behaviors are incoherent, their depressed mothers may feel overwhelmed, responding with flat and withdrawn or with harsh and intrusive behaviors. Future studies should assess the effects of more comprehensive measures of maternal mental health and well-being on the coherence and incoherence of infant cross-modal interactive behavior and their links to later attachment, in samples of infants varying in biological vulnerability.
Taken together, the results from this study are consistent with the assumption that the formation of infant attachment relationships is linked to, and mediated by, multiple mechanisms and interactive processes that begin in early life, both for infants born FT and infants born VPT (e.g. Fuertes et al., 2009Fuertes et al., , 2021Gonçalves et al., 2020;Montirosso et al., 2010). Specifically, findings from our study suggest that greater coherence in infants' early cross-modal social interactive behaviors with their caregiver is associated with later attachment security. Possible correlates of these behaviors with infant development such as rapid brain development, motor development, and infants' growing capacity for emotion regulation should be evaluated further at older infant ages and in populations varying in biological risk in future research.

Study limitations, strengths, and future directions
Our study has both limitations and strengths. One limitation is the relatively high number of infants born VPT, compared to infants born FT, who were lost to follow-up. Notably, however, there was no evidence for differential attrition within the VPT group on key medical or demographic risk factors. In our future research, we intend to study the lost cases in each group to learn more about possible reasons for this difference. A second limitation is that maternal and family-level risk and protective factors (other than demographics) were not evaluated. Moreover, mothers' current psychological symptoms at the 3-month visit were assessed only in the VPT group. Although mothers' higher psychological symptoms were associated with a greater frequency of infant incoherent crossmodal interactive behaviors in the VPT group, they were not linked to coherent behaviors or infants' later attachment organization. In future research, the role of maternal postnatal depression or anxiety, perceived social support, maternal interactive behavior, and familylevel factors such as cohesion, conflict, or chaos should be evaluated in both VPT and FT groups, to better understand the impact on caregivers' psychological well-being on the coherence of infants' cross-modal interactive behavior and their links to later attachment formation.
Strengths of this study include its longitudinal, prospective design, the use of direct observations of mother-infant interaction at 3 months postpartum and attachment at 12 months, the use of microanalytic scoring methods to evaluate the cross-modal coherence and incoherence of infant's early interactive behavior, the inclusion of infants born FT and VPT varying in birth risk, and careful evaluation of birth, demographic risk factors, and for mothers in the VPT group, maternal psychological symptoms as potential covariates. Very little research to date has evaluated the coherence and incoherence of infants' crossmodal interactive behavior during early en-face interactions in infants born FT or VPT and their associations with later attachment; thus, we believe the current findings expand our knowledge about this topic and raise promising new directions for future research in the attachment field.